โŒ

Reading view

There are new articles available, click to refresh the page.

Gestiรณn de la cartera de TI: cรณmo optimizar los activos tecnolรณgicos para generar valor empresarial

En el รกmbito financiero, la gestiรณn de carteras consiste en seleccionar estratรฉgicamente un conjunto de inversiones alineadas con los objetivos financieros y la tolerancia al riesgo del inversor. Este mismo enfoque puede aplicarse a la cartera de sistemas de TI, con una salvedad clave: ademรกs del rendimiento financiero, la funciรณn de TI debe evaluar cada activo tambiรฉn desde el punto de vista de su rendimiento operativo.

La TI actual combina sistemas heredados, plataformas en la nube y tecnologรญas emergentes o de vanguardia, como la inteligencia artificial (IA). Cada una de estas categorรญas incluye activos crรญticos para la misiรณn, pero no todos los sistemas aportan el mismo valor empresarial, financiero o de mitigaciรณn de riesgos. ยฟCรณmo pueden los CIO optimizar el rendimiento global de su cartera tecnolรณgica?

A continuaciรณn, se exponen cinco criterios de evaluaciรณn para maximizar el valor de la cartera de TI.

Activos crรญticos

Los sistemas mรกs crรญticos para el funcionamiento diario del negocio constituyen una categorรญa en sรญ mismos. Pueden ser visibles o estar ocultos en lo mรกs profundo de la pila tecnolรณgica, pero todos los activos deben evaluarse en funciรณn de su grado de criticidad.

Por ejemplo, una soluciรณn ERP puede ser un sistema imprescindible 24ร—7 porque interactรบa con una cadena de suministro global y soporta gran parte del negocio de la empresa. En cambio, una aplicaciรณn de recursos humanos o un sistema de analรญtica de marketing probablemente podrรญan permanecer inactivos durante un dรญa si el personal dispone de soluciones alternativas.

Este mismo anรกlisis debe aplicarse a servidores, redes y sistemas de almacenamiento. ยฟQuรฉ recursos son absolutamente imprescindibles y de cuรกles se podrรญa prescindir, aunque sรณlo sea de forma temporal?

A medida que el departamento de TI identifica estos activos crรญticos, conviene revisar esta clasificaciรณn con los usuarios finales y la direcciรณn para asegurar un consenso real.

Utilizaciรณn de los activos

Segรบn Zylo, proveedor especializado en la gestiรณn de inventario, licencias y renovaciones de SaaS, โ€œel 53% de las licencias de SaaS no se utilizan o estรกn infrautilizadas de media, por lo que identificar el software inactivo deberรญa ser una prioridadโ€. Este problema del โ€œsoftware sin usarโ€ no se limita al SaaS: tambiรฉn aparece en sistemas antiguos y modernos infrautilizados, servidores y discos obsoletos, o tecnologรญas de red que no se utilizan, pero por las que se sigue pagando.

En muchos casos, este fenรณmeno se debe a que el departamento de TI estรก tan centrado en proyectos que no dispone de tiempo para revisar inventarios y niveles de obsolescencia. Como resultado, productos antiguos permanecen en catรกlogo y se renuevan automรกticamente.

Si se quiere maximizar el rendimiento y la rentabilidad de la cartera de TI, este problema debe abordarse de forma sistemรกtica. Cuando el propio departamento no puede dedicar tiempo suficiente, puede recurrirse a consultores externos que realicen una evaluaciรณn de uso de activos para identificar aquellos que nunca se utilizan o lo hacen de forma marginal, para reutilizarlos o retirarlos.

Riesgo de los activos

El objetivo de una cartera de TI es reunir activos relevantes hoy y que sigan siรฉndolo maรฑana. Por ello, es imprescindible evaluar el riesgo asociado a cada recurso tecnolรณgico.

ยฟExiste riesgo de que el proveedor retire el producto o lo deje obsoleto? ยฟEs estable el propio proveedor? ยฟDispone el departamento de TI del talento necesario para seguir operando determinados sistemas, independientemente de su calidad, como aplicaciones heredadas personalizadas escritas en COBOL o Assembler? ยฟSe estรก encareciendo en exceso el funcionamiento de un sistema o de un hardware concreto? ยฟExiste una hoja de ruta clara para integrar los sistemas actuales con las tecnologรญas que llegarรกn en el futuro?

Cuando un activo se considera de riesgo, deben definirse estrategias para sacarlo de esa situaciรณn o planificar su sustituciรณn.

Valor de la propiedad intelectual

Conozco a un CIO del sector hotelero que presume de que su sistema de reservas y el mainframe en el que se ejecuta no han fallado en 30 aรฑos. Atribuye gran parte de ese รฉxito al cรณdigo personalizado y al sistema operativo especializado que utiliza la compaรฑรญa, y tanto รฉl como su comitรฉ de direcciรณn lo consideran una ventaja competitiva frente a sus rivales.

No es un caso aislado. Muchas organizaciones operan con su propia โ€œreceta secreta de TIโ€, que impulsa directamente el negocio. Esta receta puede ser un sistema heredado o un algoritmo de IA. Activos de este tipo, que constituyen autรฉntica propiedad intelectual tecnolรณgica, son un argumento sรณlido para su conservaciรณn dentro de la cartera.

TCO y ROI de los activos

ยฟRinden al mรกximo de su potencial todos los activos tecnolรณgicos? Al igual que ocurre con las inversiones financieras, las tecnologรญas deben demostrar que siguen generando un valor medible y sostenible. Los dos principales indicadores son el coste total de propiedad (TCO) y el retorno de la inversiรณn (ROI).

El TCO mide el valor de un activo a lo largo de su vida รบtil. Por ejemplo, unos servidores adquiridos para el centro de datos pudieron resultar rentables hace cuatro aรฑos, pero hoy ese mismo centro puede albergar racks obsoletos con tecnologรญa anticuada, cuando resulta mรกs eficiente trasladar la carga de trabajo a la nube.

El ROI, por su parte, se utiliza sobre todo al adquirir nueva tecnologรญa. Se definen mรฉtricas para determinar cuรกndo se recuperarรก la inversiรณn inicial. Una vez alcanzado el punto de equilibrio, se sigue monitorizando para comprobar cรณmo se materializa la rentabilidad o el ahorro esperado. Sin embargo, no todas las inversiones salen segรบn lo previsto: el caso de negocio puede cambiar o surgir complicaciones imprevistas que conviertan una inversiรณn prometedora en una carga.

Tanto en tรฉrminos de TCO como de ROI, la cartera de TI debe gestionarse de forma activa para eliminar activos que ya no aportan valor o que generan pรฉrdidas.

En resumen

La gestiรณn de la cartera de TI es una funciรณn esencial y continua del CIO, pero con demasiada frecuencia se aborda de forma reactiva: se sustituye un sistema cuando los usuarios lo reclaman o se retira un servidor cuando deja de funcionar.

Tampoco ayudan el CEO, el CFO y otros interlocutores clave durante la elaboraciรณn del presupuesto tecnolรณgico. Aunque suelen interesarse por el plazo de amortizaciรณn de una nueva inversiรณn, rara vez preguntan por la visiรณn global de la cartera de TI: cรณmo estรกn funcionando los activos en conjunto y quรฉ sistemas deberรกn sustituirse para mantener o incrementar el valor para la empresa.

Para mejorar su gestiรณn tecnolรณgica, los CIO deberรญan aprovechar el potencial de la gestiรณn de carteras. Esto implica crear una cartera formal de activos de TI y revisarla periรณdicamente con las รกreas del negocio que tienen influencia directa en los presupuestos.

Este enfoque conecta especialmente bien con el director financiero y el director general, acostumbrados a trabajar con carteras financieras y de riesgo. Una mayor visibilidad de la cartera tecnolรณgica tambiรฉn facilita al CIO presentar nuevas recomendaciones, justificar sustituciones o actualizaciones y obtener aprobaciones cuando sea necesario.

How adaptive infrastructure is evolving capabilities at the speed of business

Iโ€™m not normally fond of year-end technology retrospectives, but 2025 was indeed a year of quantum leaps in the art of the possible and it has filled us all with measured optimism paired with some healthy and well-earned skepticism where AI is concerned. When I put architecture in perspective, Iโ€™m inclined to take a longer view of automation in all its variations over a decade. Thatโ€™s why 2025 feels more like a footnote in a long series of events culminating in the perfect storm of opportunities weโ€™ve been contemplating for some time now.

The composable infrastructure revolution

Weโ€™ve been moving toward self-aware, composable infrastructure in architecture for a while now and infrastructure-as-code was merely the first major inflection.

Letโ€™s be honest, the old way of building IT infrastructure is breaking down. As an enterprise architect, the vicious cycle is very familiar. Tying agentic architecture demand-patterns to legacy infrastructure without careful consideration is fraught with peril. The old pattern is really predictable now: You provision systems, maintain them reactively and eventually retire them. Rinse and repeat.

That model is now officially unsustainable in the age of AI. Whatโ€™s taking its place? Composable and intelligent infrastructure that can proactively self-assemble, reconfigure and optimize on the fly to match what the business needs.

For IT leaders, this shift from rigid systems to modular, agent-driven infrastructure is both a breakthrough opportunity and a serious transformation challenge. And the numbers back this up: the global composable infrastructure market sits at USD $8.3 billion in 2025 and is projected to grow at 24.9% annually through 2032.

Whatโ€™s driving this hyper-accelerated growth? Geopolitical disruptions, supply chain chaos and AI advances are reshaping how and where companies operate. Business environments are being driven by reactive and dynamic agentic experiences, transactions and digital partnerships everywhere, all the time. Static infrastructure just canโ€™t deliver that kind of flexibility based on marketing exercises that describe solution offerings as โ€œon-demand,โ€ โ€œutility-based,โ€ โ€œadaptiveโ€ and โ€œcomposable.โ€ These are little more than half-truths.

A 2025 Forrester study commissioned by Microsoft found that 84% of IT leaders want solutions that consolidate edge and cloud operations across systems, sites and teams. As an architect in the consumer goods space, I found that our IT team would produce endless slide decks about composable enterprises ad nauseam, but infrastructure-as-code was the level of actual capability for some time.

Leaders wanted composable architecture that can pull together diverse components without hyperextended interoperability efforts. IBMโ€™s research reinforces this, showing that companies with modular architectures are more agile, more resilient and faster to market โ€” while also reducing the technical debt that slows everyone down.

The problem has been one of capacity and fitness for purpose. Legacy infrastructure and the underlying systems of record simply werenโ€™t designed with agentic AI patterns in mind. My conversations with pan-industry architecture colleagues reflect the same crisis of expectation and resilience around agentic architectures.

Consider McKinseyโ€™s 2025 AI survey that demonstrated 88% of organizations now use AI regularly in at least one business function and 62% are experimenting with AI agents. But most are stuck in pilot mode because their infrastructure canโ€™t scale AI across the business.

If there are any winners in this race, theyโ€™ve broken apart their monolithic systems into modular pieces that AI agents can orchestrate based on whatโ€™s actually happening in real time.

AI agents: The new orchestration layer

So, whatโ€™s driving this shift? Agentic AI โ€” systems that understand business context, figure out optimal configurations and execute complex workflows by pulling together infrastructure components on demand. This isnโ€™t just standard automation following rigid, brittle scripts. Agents reason about what to assemble, how to configure it and when to reconfigure as conditions change.

The adoption curve is steep. BCG and MIT Sloan Management Review found that 35% of organizations already use agentic AI, with another 44% planning to jump in soon. The World Economic Forum reports 82% of executives plan to adopt AI agents within three years. McKinseyโ€™s abovementioned State of AI research further highlights agentic AI as an emerging focus area for enterprise investment and describes AI agents as systems that can plan, take actions and orchestrate multi-step workflows with less human intervention than traditional automation.

As McKinsey puts it: โ€œWeโ€™re entering an era where enterprise productivity is no longer just accelerated by AI โ€” itโ€™s orchestrated by it.โ€ Thatโ€™s a fundamental change in how infrastructure works.

IBM is betting big on this future, stating that โ€œthe future of IT operations is autonomous, policy-driven and hybrid by design.โ€ Theyโ€™re building environments where AI agents can orchestrate everything โ€” public cloud, private infrastructure, on-premises systems, edge deployments โ€” assembling optimal configurations for specific workloads and contexts. The scope of automation ranges from helpful recommendations to closed-loop fixes to fully autonomous optimization.

What composable architecture actually looks like

I recall no shortage of Lego-induced architecture references to composability over the last decade. Sadly, we conflated them with domain services and not how business capabilities and automation could and should inform how the Legos are pieced together to solve problems. Traditional infrastructure comes as tightly integrated stacks โ€” hard to decompose, inflexible and reactive. The new composable model flips this, offering modular building blocks that agents can intelligently assemble and reassemble dynamically based on whatโ€™s needed right now.

Composability demands modularity and responsive automation

The foundation is extreme modularity โ€” breaking monolithic systems into discrete, independently deployable pieces with clean interfaces. Composable infrastructure lets you dynamically assemble and disassemble resources based on application demands, optimizing how pooled resources get allocated and improving overall efficiency.

This goes far beyond physical infrastructure to include services, data pipelines, security policies and workflows. When everything is modular and API-accessible, agents can compose complex solutions from simple building blocks and adapt in real time.

Bringing cloud and edge together

Enterprise organizations are no longer treating cloud and edge as separate worlds requiring manual integration. The new approach treats all infrastructure โ€” from hyperscale data centers to network edge โ€” as a unified resource pool that agents can compose into optimal configurations.

McKinsey identifies edge-cloud convergence as essential for agentic AI: โ€œAgents need real-time data access and low-latency environments. Combining edge compute (for inference and responsiveness) with cloud-scale training and storage is essential.โ€ They further highlight how Hewlett Packard Enterprise (HPE) expanded its GreenLake platform in late 2024 with composable infrastructure hardware for hybrid and AI-driven workloads โ€” modular servers and storage that let enterprises dynamically allocate resources based on real-time demand.

Agents running the show

Even IBM with its storied fixed-infrastructure history is all-in on agentic AI infrastructure capabilities โ€” including agents and Model Context Protocol (MCP) servers โ€” across its portfolio, making infrastructure components discoverable and composable by AI agents. These agents donโ€™t just watch the infrastructure state; they actively orchestrate resources across enterprise data and applications, creating optimal configurations for specific workloads.

Management interfaces across IBM cloud, storage, power and Z platforms are becoming MCP-compatible services โ€” turning infrastructure into building blocks that agents can reason about and orchestrate. Vendor-native agentic management solutions introduced similar AI-driven orchestration enhancements in 2024, letting large enterprises dynamically allocate resources across compute, storage and networking.

Self-aware and self-correcting infrastructure

Instead of manually configuring every component, composable architectures enable intent-based interfaces. You specify business objectives โ€” support 10,000 concurrent users with sub-100ms latency at 99.99% availability โ€” and agents figure out the infrastructure composition to make it happen.

Emerging intelligent infrastructure player Quali describes this as โ€œinfrastructure that understands itselfโ€ โ€” systems where agentic AI doesnโ€™t just demand infrastructure that keeps up, but infrastructure built from composable components that agents can understand and orchestrate.

Getting scale and flexibility in real time

Traditional infrastructure forces a choice: optimize for scale or build for adaptability. As architects, there are clear opposing trade-offs we must navigate successfully: Scale relative to adaptability, investment versus sustaining operations, tight oversight versus autonomy and process refactoring versus process reinvention.

Composable architectures solve this by delivering both. The dual nature of agentic AI โ€” part tool, part human-like โ€” doesnโ€™t fit traditional management frameworks. People are flexible but donโ€™t scale. Tools scale but canโ€™t adapt. Agentic AI on composable infrastructure gives you scalable adaptability โ€” handling massive workloads while continuously reconfiguring for changing contexts.

Self-composability and evolved governance

Agent-orchestrated infrastructure demands governance that balances autonomy with control. The earlier-mentioned MIT Sloan Management Review and BCG study found that most agentic AI leaders anticipate significant changes to governance and decision rights as they adopt agentic AI. They recommend creating governance hubs with enterprise-wide guardrails and dynamic decision rights rather than approving individual AI decisions one by one.

The answer lies in policy-based composition, defining constraints that bound agent decisions without prescribing exact configurations. Within those boundaries, agents compose and recompose infrastructure autonomously.

When AI agents continuously compose and recompose resources, you need governance frameworks that look nothing like traditional change management. A model registry that includes MCP connects different large language models while implementing guardrails for analytics, security, privacy and compliance. This treats AI as an agent whose decisions must be understood, managed and learned from โ€” not as an infallible tool.

Making it happen in 2026

What should IT leaders do? Here are the most critical moves from my perspective.

Redesign work around agents first. Use agentic AIโ€™s capacity to implement scalability and adapt broadly within parameterized governance automation rather than automating isolated tasks. Almost two-thirds of agentic AI leaders expect operating model changes. Build workflows that shift smoothly between efficiency and problem-solving modes.

Rethink roles for human-agent collaboration. Agents are an architectโ€™s new partner. Reposition your role as an architect in the enterprise to adopt and embrace portfolios of AI agents to coordinate workflows, and traditional management layers change. Expect fewer middle management layers, with managers evolving to orchestrate hybrid human-AI teams. Consider dual career paths for generalist orchestrators and AI-augmented specialists.

Keep investments tied to value. Agentic AI leaders anchor investments to value โ€” whether efficiency, innovation, revenue growth or some combination. Agentic systems are evolving from finite function agents to multi-agent collaborators, from narrow to broadly orchestrated tasks and other ecosystems and agents, and from operational to strategic human-mediated partnership.

The bottom line

The companies that will win in the next decade will recognize composability as the foundation of adaptive infrastructure. When every part of the technology stack becomes a modular building block and intelligent agents compose those blocks into optimal configurations based on real-time context, infrastructure becomes a competitive advantage instead of a constraint.

Organizations that understand agentic AIโ€™s dual nature and align their processes, governance, talent and investments accordingly will realize its full business value. My architectโ€™s perspective is that agentic AI will challenge established management approaches and, yes, even convince many of its ability to defy gravity. But with the right strategy and execution, it wonโ€™t just offer empty promises โ€” it will deliver results. Further, our grounded expectations around the capacity of aging infrastructure and legacy demand patterns must guide us in ensuring we make intelligent decisions.

The question isnโ€™t whether to embrace composable, agent-orchestrated infrastructure. Itโ€™s how fast you can decompose monolithic systems, build orchestration capabilities and establish the governance to make it work.

This article was made possible by our partnership with the IASA Chief Architect Forum. The CAFโ€™s purpose is to test, challenge and support the art and science of Business Technology Architecture and its evolution over time, as well as grow the influence and leadership of chief architects both inside and outside the profession. The CAF is a leadership community of the IASA, the leading non-profit professional association for business technology architects.ย 

This article is published as part of the Foundry Expert Contributor Network.
Want to join?

์˜คํ”ˆAI, ์„ธ๋ ˆ๋ธŒ๋ผ์Šค์™€ ์ดˆ๋Œ€ํ˜• ๊ณ„์•ฝ ์ฒด๊ฒฐยทยทยทAI ์ถ”๋ก  ์ธํ”„๋ผ ํ™•์žฅ ๋‚˜์„œ

์˜คํ”ˆAI๋Š” AI ์นฉ ์Šคํƒ€ํŠธ์—… ์„ธ๋ ˆ๋ธŒ๋ผ์Šค ์‹œ์Šคํ…œ์ฆˆ(Cerebras Systems)๋กœ๋ถ€ํ„ฐ ์ปดํ“จํŒ… ์šฉ๋Ÿ‰์„ ๊ตฌ๋งคํ•˜๋Š” ์ˆ˜์‹ญ์–ต ๋‹ฌ๋Ÿฌ ๊ทœ๋ชจ์˜ ๊ณ„์•ฝ์„ ์ฒด๊ฒฐํ–ˆ๋‹ค๊ณ  14์ผ ๋ฐํ˜”๋‹ค. ์ฑ—GPT ๊ฐœ๋ฐœ์‚ฌ์ธ ์˜คํ”ˆAI๋Š” ๊ธ‰์ฆํ•˜๋Š” ์‚ฌ์šฉ์ž ์ˆ˜์š”์— ๋Œ€์‘ํ•˜๊ณ , ๋ฐ์ดํ„ฐ์„ผํ„ฐ์™€ ๋„คํŠธ์›Œํฌ ์ž์›์— ๊ฐ€ํ•ด์ง€๋Š” ๋ถ€๋‹ด์ด ์ปค์ง€๋Š” ์ƒํ™ฉ์—์„œ ์ถฉ๋ถ„ํ•œ ์ธํ”„๋ผ๋ฅผ ํ™•๋ณดํ•˜๊ธฐ ์œ„ํ•œ ์กฐ์น˜๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

์ด๋ฒˆ ๊ณ„์•ฝ์— ๋”ฐ๋ผ ์˜คํ”ˆAI๋Š” ์„ธ๋ ˆ๋ธŒ๋ผ์Šค๊ฐ€ ์„ค๊ณ„ํ•œ ์นฉ์„ ํ™œ์šฉํ•ด ์ฑ—GPT ์ถ”๋ก  ์›Œํฌ๋กœ๋“œ ์ผ๋ถ€๋ฅผ ์šด์˜ํ•˜๊ฒŒ ๋œ๋‹ค. ์›”์ŠคํŠธ๋ฆฌํŠธ์ €๋„(WSJ)์˜ 14์ผ ๋ณด๋„์— ๋”ฐ๋ฅด๋ฉด ์˜คํ”ˆAI๋Š” 3๋…„์— ๊ฑธ์ณ ์ตœ๋Œ€ 750๋ฉ”๊ฐ€์™€ํŠธ ๊ทœ๋ชจ์˜ ์ปดํ“จํŒ… ์šฉ๋Ÿ‰์„ ๊ตฌ๋งคํ•˜๊ธฐ๋กœ ์•ฝ์ •ํ–ˆ๋‹ค.

์ด๋ฒˆ ํ–‰๋ณด๋Š” ๋Œ€๊ทœ๋ชจ AI ์„œ๋น„์Šค๊ฐ€ ์ „๋ ฅ ์ˆ˜๊ธ‰, ๋„คํŠธ์›Œํ‚น, ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๊ฐ„ ์—ฐ๊ฒฐ์„ฑ์— ์ƒ๋‹นํ•œ ๋ถ€๋‹ด์„ ์ฃผ๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ค€๋‹ค. ์˜คํ”ˆAI๋Š” ์—”๋น„๋””์•„์˜ ์ง€๋ฐฐ์ ์ธ GPU๋ณด๋‹ค ๋” ๋น ๋ฅด๊ณ  ๋น„์šฉ ํšจ์œจ์ ์ธ ๋Œ€์•ˆ์„ ์ฐพ๊ณ  ์žˆ๋‹ค.

์˜คํ”ˆAI ๊ฒฝ์˜์ง„์€ ํšŒ์‚ฌ๊ฐ€ ์ปดํ“จํŒ… ์šฉ๋Ÿ‰ ์ œ์•ฝ์— ์ง๋ฉดํ•ด ์žˆ๋‹ค๊ณ  ์ง€์†์ ์œผ๋กœ ๊ฒฝ๊ณ ํ•ด ์™”๋‹ค. ํ˜„์žฌ ์˜คํ”ˆAI์˜ ๋„๊ตฌ๋Š” ๋งค์ฃผ 8์–ต ๋ช…์ด ๋„˜๋Š” ์‚ฌ์šฉ์ž๊ฐ€ ์ด์šฉํ•˜๊ณ  ์žˆ๋‹ค. ์ด๋กœ ์ธํ•ด ์ธํ”„๋ผ๋ฅผ ํ™•์žฅํ•˜๊ธฐ ์œ„ํ•ด ๋” ๋งŽ์€ ํŒŒํŠธ๋„ˆ๊ฐ€ ํ•„์š”ํ•˜๋‹ค๋Š” ํŒ๋‹จ์ด ๋‚˜์˜ค๊ณ  ์žˆ๋‹ค.

์„ธ๋ ˆ๋ธŒ๋ผ์Šค์™€์˜ ๊ณ„์•ฝ์€ ์˜คํ”ˆAI๊ฐ€ ์ธํ”„๋ผ๋ฅผ ๋‹ค๊ฐํ™”ํ•˜๊ธฐ ์œ„ํ•ด ์ถ”์ง„ํ•ด ์˜จ ์ผ๋ จ์˜ ๋…ธ๋ ฅ์˜ ์—ฐ์žฅ์„ ์— ์žˆ๋‹ค. ์˜คํ”ˆAI๋Š” ๋ธŒ๋กœ๋“œ์ปด๊ณผ ๋งž์ถคํ˜• AI ์นฉ์„ ๊ฐœ๋ฐœํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, AMD์˜ ์ตœ์‹  ๊ฐ€์†๊ธฐ ๋„์ž…๋„ ๊ณ„ํšํ•˜๊ณ  ์žˆ๋‹ค. ์ด๋Š” ๋น„์šฉ์„ ์–ต์ œํ•˜๊ณ  ์—”๋น„๋””์•„์— ๋Œ€ํ•œ ์˜์กด๋„๋ฅผ ๋‚ฎ์ถ”๊ธฐ ์œ„ํ•œ ์ „๋žต์ด๋‹ค.

AI ์ธํ”„๋ผ ์žฌ์„ค๊ณ„

์˜คํ”ˆAI๊ฐ€ ์ „์šฉ ์ถ”๋ก  ์šฉ๋Ÿ‰์— ๋Œ€๊ทœ๋ชจ๋กœ ํˆฌ์žํ•œ ๊ฒƒ์€ ๋Œ€ํ˜• AI ํ”Œ๋žซํผ๋“ค์ด ๋‹จ์ผ ๊ฐ€์†๊ธฐ ๋ชจ๋ธ์„ ๋„˜์–ด, ์ง€์—ฐ ์‹œ๊ฐ„์— ๋ฏผ๊ฐํ•œ ์›Œํฌ๋กœ๋“œ๋ฅผ ์ง€์›ํ•˜๊ธฐ ์œ„ํ•ด ์ธํ”„๋ผ๋ฅผ ์žฌ๊ตฌ์„ฑํ•˜๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ค€๋‹ค.

์• ๋„๋ฆฌ์ŠคํŠธ๋“ค์€ ํ–ฅํ›„ AI ์›Œํฌ๋กœ๋“œ๊ฐ€ ๋”์šฑ ๋‹ค์–‘ํ•ด์ง€๊ณ  ์š”๊ตฌ ์ˆ˜์ค€๋„ ๋†’์•„์งˆ ๊ฒƒ์œผ๋กœ ๋ณด๊ณ  ์žˆ๋‹ค. ์ด์— ๋”ฐ๋ผ ์ถ”๋ก  ์„ฑ๋Šฅ์— ์ตœ์ ํ™”๋œ ์•„ํ‚คํ…์ฒ˜์— ๋Œ€ํ•œ ํ•„์š”์„ฑ์ด ์ปค์ง€๊ณ  ์žˆ์œผ๋ฉฐ, ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๋„คํŠธ์›Œํฌ์— ๊ฐ€ํ•ด์ง€๋Š” ๋ถ€๋‹ด๋„ ํ•œ์ธต ๊ฐ€์ค‘๋  ์ „๋ง์ด๋‹ค.

์ปจ์„คํŒ… ๊ธฐ์—… ์นด์šดํ„ฐํฌ์ธํŠธ๋ฆฌ์„œ์น˜์˜ ๋ฆฌ์„œ์น˜ ๋ถ€๋ฌธ ๋ถ€์‚ฌ์žฅ ๋‹ ์ƒค๋Š” โ€œ์ด๋Ÿฌํ•œ ๋ณ€ํ™”๋กœ ํ•˜์ดํผ์Šค์ผ€์ผ๋Ÿฌ๋Š” ์ปดํ“จํŒ… ์‹œ์Šคํ…œ์„ ๋‹ค๊ฐํ™”ํ•˜๊ณ  ์žˆ๋‹คโ€๋ผ๋ฉฐ โ€œ๋ฒ”์šฉ AI ์›Œํฌ๋กœ๋“œ์—๋Š” ์—”๋น„๋””์•„ GPU๋ฅผ ํ™œ์šฉํ•˜๊ณ , ๊ณ ๋„๋กœ ์ตœ์ ํ™”๋œ ์ž‘์—…์—๋Š” ์ž์ฒด AI ๊ฐ€์†๊ธฐ๋ฅผ, ์ดˆ์ €์ง€์—ฐ์ด ์š”๊ตฌ๋˜๋Š” ํŠนํ™” ์›Œํฌ๋กœ๋“œ์—๋Š” ์„ธ๋ ˆ๋ธŒ๋ผ์Šค์™€ ๊ฐ™์€ ์‹œ์Šคํ…œ์„ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ์‹โ€์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

์ด๋กœ ์ธํ•ด ํ•˜์ดํผ์Šค์ผ€์ผ ํ™˜๊ฒฝ์—์„œ ์šด์˜๋˜๋Š” AI ํ”Œ๋žซํผ์€ ๋‹จ์ผํ•˜๊ณ  ๋ฒ”์šฉ์ ์ธ ํด๋Ÿฌ์Šคํ„ฐ์—์„œ ๋ฒ—์–ด๋‚˜, ๋ณด๋‹ค ๊ณ„์ธตํ™”๋˜๊ณ  ์ด๊ธฐ์ข…์ด ํ˜ผํ•ฉ๋œ ์ธํ”„๋ผ ์ „๋žต์œผ๋กœ ์ธํ”„๋ผ ์ œ๊ณต์—…์ฒด๋ฅผ ์••๋ฐ•ํ•˜๊ณ  ์žˆ๋‹ค.

์‚ฌ์ด๋ฒ„๋ฏธ๋””์–ด๋ฆฌ์„œ์น˜ ์‚ฐ์—…์—ฐ๊ตฌ๊ทธ๋ฃน์˜ ๋ถ€์‚ฌ์žฅ ํ”„๋ผ๋ถ€ ๋žŒ์€ โ€œ์˜คํ”ˆAI๊ฐ€ ์„ธ๋ ˆ๋ธŒ๋ผ์Šค ๊ธฐ๋ฐ˜ ์ถ”๋ก  ์šฉ๋Ÿ‰์„ ๋„์ž…ํ•œ ๊ฒƒ์€ AI ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์„ค๊ณ„ ๋ฐฉ์‹ ์ „๋ฐ˜์—์„œ ๋‚˜ํƒ€๋‚˜๋Š” ๊ด‘๋ฒ”์œ„ํ•œ ๋ณ€ํ™”๋ฅผ ๋ฐ˜์˜ํ•œ๋‹คโ€๋ผ๋ฉฐ โ€œ์ด๋Š” ์—”๋น„๋””์•„๋ฅผ ๋Œ€์ฒดํ•˜๊ธฐ ์œ„ํ•œ ์›€์ง์ž„์ด๋ผ๊ธฐ๋ณด๋‹ค, ์ถ”๋ก ์ด ํ™•๋Œ€๋˜๋ฉด์„œ ๋‚˜ํƒ€๋‚˜๋Š” ๋‹ค๊ฐํ™” ํ๋ฆ„์— ๊ฐ€๊น๋‹คโ€๋ผ๊ณ  ๋ถ„์„ํ–ˆ๋‹ค.

๋žŒ์€ ์ด ์ •๋„ ๊ทœ๋ชจ์— ์ด๋ฅด๋ฉด ์ธํ”„๋ผ๋Š” AI ๊ณต์žฅ๊ณผ ์œ ์‚ฌํ•œ ํ˜•ํƒœ๋ฅผ ๋ ๊ฒŒ ๋œ๋‹ค๋ฉฐ, ๋„์‹œ ๊ทœ๋ชจ์˜ ์ „๋ ฅ ๊ณต๊ธ‰๊ณผ ๊ณ ๋ฐ€๋„ ๋™์„œ ํŠธ๋ž˜ํ”ฝ ๋„คํŠธ์›Œํ‚น, ์ €์ง€์—ฐ ์ธํ„ฐ์ปค๋„ฅํŠธ๊ฐ€ ์ตœ๋Œ€ ์—ฐ์‚ฐ ์„ฑ๋Šฅ(FLOPS)๋ณด๋‹ค ๋” ์ค‘์š”ํ•ด์ง„๋‹ค๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

ํ…Œํฌ์ธ์‚ฌ์ด์ธ ์˜ ๋ฐ˜๋„์ฒด ์• ๋„๋ฆฌ์ŠคํŠธ ๋งˆ๋‹ˆ์‹œ ๋ผ์™“์€ โ€œ์ด ์ˆ˜์ค€์—์„œ๋Š” ๊ธฐ์กด์˜ ๋ž™ ๋ฐ€๋„, ๋ƒ‰๊ฐ ๋ชจ๋ธ, ๊ณ„์ธตํ˜• ๋„คํŠธ์›Œํฌ๊ฐ€ ๋” ์ด์ƒ ์‹ค์šฉ์ ์ด์ง€ ์•Š๋‹คโ€๋ผ๋ฉฐ โ€œ์ถ”๋ก  ์›Œํฌ๋กœ๋“œ๋Š” ํ•™์Šต์ฒ˜๋Ÿผ ๊ฐ„ํ—์ ์œผ๋กœ ๋ฐœ์ƒํ•˜๋Š” ์ž‘์—…์ด ์•„๋‹ˆ๋ผ ์ง€์†์ ์ด๊ณ  ์ง€์—ฐ์— ๋ฏผ๊ฐํ•œ ํŠธ๋ž˜ํ”ฝ์„ ์ƒ์„ฑํ•˜๊ธฐ ๋•Œ๋ฌธ์—, ๋” ํ‰ํ‰ํ•œ ๋„คํŠธ์›Œํฌ ํ† ํด๋กœ์ง€์™€ ๊ณ ์ฐจ์ˆ˜ ์Šค์œ„์นญ, ์—ฐ์‚ฐยท๋ฉ”๋ชจ๋ฆฌยท์ธํ„ฐ์ปค๋„ฅํŠธ์˜ ๊ธด๋ฐ€ํ•œ ํ†ตํ•ฉ์„ ์š”๊ตฌํ•œ๋‹คโ€๋ผ๊ณ  ์ „ํ–ˆ๋‹ค.

๊ทœ๋ชจ ํ™•์žฅ์— ๋”ฐ๋ฅธ ๋ณต์žก์„ฑ ์ฆ๊ฐ€

์—”๋น„๋””์•„์˜ GPU ๊ธฐ๋ฐ˜ ๋ชจ๋ธ์€ ์—ฌ์ „ํžˆ ์—…๊ณ„ ํ‘œ์ค€์œผ๋กœ ์ž๋ฆฌ ์žก๊ณ  ์žˆ์ง€๋งŒ, AI ํด๋Ÿฌ์Šคํ„ฐ๊ฐ€ ํ™•์žฅ๋ ์ˆ˜๋ก ํŠนํžˆ ์ธํ„ฐ์ปค๋„ฅํŠธ ์ˆ˜์š”๊ฐ€ ๋Š˜์–ด๋‚˜๋ฉด์„œ ๊ตฌ์กฐ๋Š” ์ ์  ๋” ๋ณต์žกํ•ด์ง€๊ณ  ์ „๋ ฅ ํšจ์œจ์„ฑ๋„ ๋–จ์–ด์ง€๊ณ  ์žˆ๋‹ค.

๋žŒ์€ โ€œ์„ธ๋ ˆ๋ธŒ๋ผ์Šค์˜ ์›จ์ดํผ ์Šค์ผ€์ผ ์•„ํ‚คํ…์ฒ˜๋Š” ๋‹ค์ค‘ GPU ํŒจ๋ธŒ๋ฆญ์—์„œ ํ•„์—ฐ์ ์œผ๋กœ ๋ฐœ์ƒํ•˜๋Š” ํ†ต์‹  ์˜ค๋ฒ„ํ—ค๋“œ(์—ฌ๋Ÿฌ GPU๊ฐ€ ๋ฐ์ดํ„ฐ๋ฅผ ์ฃผ๊ณ ๋ฐ›๋Š” ๊ณผ์ •์—์„œ ๋ฐœ์ƒํ•˜๋Š” ํ†ต์‹  ์ง€์—ฐ๊ณผ ์ž์› ์†Œ๋ชจ)๋ฅผ ์ค„์—ฌ ์ถ”๋ก  ์ฒ˜๋ฆฌ๋Ÿ‰๊ณผ ๋น„์šฉ ์ธก๋ฉด์—์„œ ์ž ์žฌ์ ์ธ ์ด์ ์„ ์ œ๊ณตํ•œ๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

๋‹ค๋งŒ ์• ๋„๋ฆฌ์ŠคํŠธ๋“ค์€ ์ธํ”„๋ผ ๋‹ค๊ฐํ™”๊ฐ€ ์ž์ฒด์ ์ธ ์šด์˜ ๊ณผ์ œ๋ฅผ ๋™๋ฐ˜ํ•œ๋‹ค๊ณ  ๊ฒฝ๊ณ ํ•œ๋‹ค.

๋ผ์™“์€ โ€œ์ด๋Ÿฌํ•œ ๋ณ€ํ™”๋Š” ์šด์˜ ๋ณต์žก์„ฑ์„ ๋†’์ธ๋‹คโ€๋ฉฐ โ€œ์ด๊ธฐ์ข… ๊ฐ€์†๊ธฐ๋ฅผ ์šด์˜ํ•˜๋ ค๋ฉด ์—ฌ๋Ÿฌ ์†Œํ”„ํŠธ์›จ์–ด ์Šคํƒ๊ณผ ์„œ๋กœ ๋‹ค๋ฅธ ์žฅ์•  ์œ ํ˜•์„ ๊ด€๋ฆฌํ•ด์•ผ ํ•˜๊ณ , ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ „๋ฐ˜์—์„œ ๋” ๋ณต์žกํ•œ ์šฉ๋Ÿ‰ ์˜ค์ผ€์ŠคํŠธ๋ ˆ์ด์…˜์ด ํ•„์š”ํ•˜๋‹คโ€๊ณ  ์–ธ๊ธ‰ํ–ˆ๋‹ค.

์˜คํ”ˆAI๋Š” ์ˆ˜์š” ๋ณ€๋™์„ฑ์ด ํฐ ํ™˜๊ฒฝ์—์„œ ํŠนํ™”๋œ ์ปดํ“จํŒ… ์šฉ๋Ÿ‰์— ๋Œ€ํ•œ ์žฅ๊ธฐ์  ํˆฌ์ž ์•ฝ์ •์„ ๋งž์ถ”๋Š” ๋™์‹œ์—, ์›Œํฌ๋กœ๋“œ๋ฅผ ์‹ค์‹œ๊ฐ„์œผ๋กœ ๊ฐ€์žฅ ํšจ์œจ์ ์ธ ๊ฐ€์†๊ธฐ๋กœ ๋™์ ์œผ๋กœ ๋ถ„์‚ฐํ•ด์•ผ ํ•˜๋Š” ๊ณผ์ œ์— ์ง๋ฉดํ•ด ์žˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ์˜ค์ผ€์ŠคํŠธ๋ ˆ์ด์…˜๊ณผ ์ž์› ํ™œ์šฉ์—์„œ ๋ฐœ์ƒํ•  ์ˆ˜ ์žˆ๋Š” ๊ณต๋ฐฑ์„ ์ตœ์†Œํ™”ํ•˜๋Š” ๊ฒƒ์ด ํ•ต์‹ฌ ๊ณผ์ œ๋กœ ๊ผฝํžŒ๋‹ค.

์ธํ”„๋ผ๋ฅผ ์—ฌ๋Ÿฌ ์•„ํ‚คํ…์ฒ˜๋กœ ๋ถ„์‚ฐํ•˜๋ฉด์„œ ์ด๋Ÿฌํ•œ ํˆฌ์ž ์ž์‚ฐ์˜ ์ˆ˜๋ช… ์ฃผ๊ธฐ๋ฅผ ๊ด€๋ฆฌํ•˜๋Š” ๋ฌธ์ œ๋„ ๋˜ ๋‹ค๋ฅธ ๋ถ€๋‹ด์ด๋‹ค. ์ƒค๋Š” โ€œ์‹ค๋ฆฌ์ฝ˜์˜ ์ˆ˜๋ช… ์ฃผ๊ธฐ์ธ 18~24๊ฐœ์›”๊ณผ ์‹œ์„ค์˜ ์ˆ˜๋ช… ์ฃผ๊ธฐ์ธ 15~20๋…„ ์‚ฌ์ด์˜ ๊ฒฉ์ฐจ๊ฐ€ ์ ์  ๋ฒŒ์–ด์ง€๊ณ  ์žˆ๋‹คโ€๋ฉฐ โ€œ์นฉ ํ˜์‹  ์†๋„๋ฅผ ๊ณ ๋ คํ•˜๋ฉด, 100์–ต ๋‹ฌ๋Ÿฌ(์•ฝ 13์กฐ ์›)๋ฅผ ๋„˜๋Š” ํŠนํ™” ํ•˜๋“œ์›จ์–ด๊ฐ€ ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ์™„์ „ํžˆ ๊ฐ€๋™๋˜๊ธฐ๋„ ์ „์— ๊ธฐ์ˆ ์ ์œผ๋กœ ๊ตฌ์‹์ด ๋  ์œ„ํ—˜์ด ์žˆ๋‹คโ€๊ณ  ์ง€์ ํ–ˆ๋‹ค.

AI ์ธํ”„๋ผ๊ฐ€ ํ™•์žฅ๋ ์ˆ˜๋ก ์ „๋ ฅ, ๋ƒ‰๊ฐ, ๋„คํŠธ์›Œํ‚น์ด ์ง€๋ฐฐ์ ์ธ ์ œ์•ฝ ์š”์ธ์œผ๋กœ ๋ถ€์ƒํ•˜๊ณ  ์žˆ๋‹ค. ๋žŒ์€ โ€œ์ปดํ“จํŒ… ์•„ํ‚คํ…์ฒ˜๋ฅผ ์ „๋ ฅ๋ง ๊ทœ๋ชจ์˜ ์ „๋ ฅ ๊ณต๊ธ‰๊ณผ ํšจ์œจ์ ์ธ ๋ฐ์ดํ„ฐ ์ด๋™์— ๋งž์ถœ ์ˆ˜ ์žˆ๋Š” ๋ฒค๋”๊ฐ€ ์ฐจ์„ธ๋Œ€ AI ์ธํ”„๋ผ์˜ ๋ฐฉํ–ฅ์„ ๊ทœ์ •ํ•˜๊ฒŒ ๋  ๊ฒƒโ€์ด๋ผ๊ณ  ์ „ํ–ˆ๋‹ค.
dl-ciokorea@foundryco.com

Your 100 Billion Parameter Behemoth is a Liability

The "bigger is better" era of AI is hitting a wall. We are in an LLM bubble, characterized by ruinous inference costs and diminishing returns. The future belongs to Agentic AI powered by specialized Small Language Models (SLMs). Think of it as a shift from hiring a single expensive genius to running a highly efficient digital factory. Itโ€™s cheaper, faster, and frankly, the only way to make agents work at scale.

IT portfolio management: Optimizing IT assets for business value

In finance, portfolio management involves the strategic selection of a collection of investments that align with an investorโ€™s financial goals and risk tolerance.ย 

This approach can also apply to ITโ€™s portfolio of systems, with one addition: IT must also assess each asset in that portfolio for operational performance.

Todayโ€™s IT is a mix of legacy, cloud-based, and emerging or leading-edge systems, such as AI. Each category contains mission-critical assets, but not every system performs equally well when it comes to delivering business, financial, and risk avoidance value to the enterprise. How can CIOs optimize their IT portfolio performance?

Here are five evaluative criteria for maximizing the value of your IT portfolio.

Mission-critical assets

The enterpriseโ€™s most critical systems for conducting day-to-day business are a category unto themselves. These systems may be readily apparent, or hidden deep in a technical stack. So all assets should be evaluated as to how mission-critical they are.

For example, it might be that your ERP solution is a 24/7 โ€œmust haveโ€ system because it interfaces with a global supply chain that operates around the clock and drives most company business. On the other hand, an HR application or a marketing analytics system could probably be down for a day with work-arounds by staff.

More granularly, the same type of analysis needs to be performed on IT servers, networks and storage. Which resources do you absolutely have to have, and which can you do without, if only temporarily?

As IT identifies these mission-critical assets, it should also review the list with end-users and management to assure mutual agreement.

Asset utilization

Zylo, which manages SaaS inventory, licenses, and renewals, estimates that โ€œ53% of SaaS licenses go unused or underused on average, so finding dormant software should be a priority.โ€ This โ€œshelfwareโ€ problem isnโ€™t only with SaaS; it can be found in underutilized legacy and modern systems, in obsolete servers and disk drives, and in network technologies that arenโ€™t being used but are still being paid for.

Shelfware in all forms exists because IT is too busy with projects to stop for inventory and obsolescence checks. Consequently, old stuff gets set on the shelf and auto-renews.

The shelfware issue should be solved if IT portfolios are to be maximized for performance and profitability. If IT canโ€™t spare the time for a shelfware evaluation, it can bring in a consultant to perform an assessment of asset use and to flag never-used or seldom-used assets for repurposing or elimination.

Asset risk

The goal of an IT portfolio is to contain assets that are presently relevant and will continue to be relevant well into the future. Consequently, asset risk should be evaluated for each IT resource.

Is the resource at risk for vendor sunsetting or obsolescence? Is the vendor itself unstable? Does IT have the on-staff resources to continue running a given system, no matter how good it is (a custom legacy system written in COBOL and Assembler, for example)? Is a particular system or piece of hardware becoming too expense to run? Do existing IT resources have a clear path to integration with the new technologies that will populate IT in the future?

For IT assets that are found to be at risk, strategies should be enacted to either get them out of โ€œriskโ€ mode, or to replace them.

Asset IP value

There is a CIO I know in the hospitality industry who boasts that his hotel reservation program, and the mainframe it runs on, have not gone down in 30 years. He attributes much of this success to custom code and a specialized operating system that the company uses, and he and his management view it as a strategic advantage over the competition.

He is not the only CIO who feels this way. There are many companies that operate with their โ€œown IT special sauceโ€ that makes their businesses better. This special sauce could be a legacy system or an AI algorithm. Assets like these that become IT intellectual property (IP) present a case for preservation in the IT portfolio.

Asset TCO and ROI

Is every IT asset pulling its weight? Like monetary and stock investments, technologies under management must show they are continuing to produce measurable and sustainable value. The primary indicators of asset value that IT uses are total cost of ownership (TCO) and return on investment (ROI).

TCO is what gauges the value of an asset over time. For instance, investments in new servers for the data center might have paid off four years ago, but now the data center has an aging bay of servers with obsolete technology and it is cheaper to relocate compute to the cloud.

ROI is used when new technology is acquired. Metrics are set that define at what point the initial investment into the technology will be recouped. Once the breakeven point has been reached, ROI continues to be measured because the company wants to see new profitability and/or savings materialize from the investment. Unfortunately, not all technology investments go as planned. Sometimes the initial business case that called for the technology changes or unforeseen complications arise that turn the investment into a loss leader.

In both cases, whether the issue is TCO or ROI, the IT portfolio must be maintained in a way such that losing or wasted assets are removed.

Summing it up

IT portfolio management is an important part of what CIOs should be doing on an ongoing basis, but all too often, it is approached in a reactionary mode โ€” for example, with a system being replaced only when users ask for it to be replaced, or a server needing to be removed from the data center because it fails.

The CEO, the CFO, and other key stakeholders whom the CIO deals with during technology budgeting time donโ€™t help, either. While they will be interested in how long it will take for a new technology acquisition to โ€œpay for itself,โ€ no one ever asks the CIO about the big picture of IT portfolio management: how the overall assets in the IT portfolio are performing, and which assets will require replacement for the portfolio to sustain or improve company value.

To improve their own IT management, CIOs should seize the portfolio management opportunity. They can do this by establishing a portfolio for their companyโ€™s IT assets and reviewing these assets periodically with those in the enterprise who have direct say over IT budgets.

IT portfolio management will resonate with the CFO and CEO because both continually work with financial and risk portfolios for the business. Broader visibility of the IT portfolio will also make it easier for CIOs to present new technology recommendations and to obtain approvals for replacing or upgrading existing assets when these actions are called for.

See also:

์ „๋ ฅ ๋ถ€์กฑยทํƒ„์†Œ ํฌ์ง‘ยทAI ์ž๋™ํ™” ๅค–ยทยทยท2026๋…„ ๋ฐ์ดํ„ฐ์„ผํ„ฐ์˜ ์ฃผ์š” ๋ณ€์ˆ˜ 5๊ฐ€์ง€

์—…ํƒ€์ž„ ์ธ์Šคํ‹ฐํŠœํŠธ(์ดํ•˜ ์—…ํƒ€์ž„)์˜ โ€˜2026๋…„ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ „๋งโ€™ ์กฐ์‚ฌ์— ๋”ฐ๋ฅด๋ฉด, ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์‚ฐ์—…์€ ๊ธฐ์ˆ ์  ๋Œ€์‘๋งŒ์œผ๋กœ ํ•ด๊ฒฐํ•˜๊ธฐ ์–ด๋ ค์šด ์ „๋ ฅ ์ˆ˜๊ธ‰ ๊ณผ์ œ์— ์ง๋ฉดํ•˜๊ณ  ์žˆ๋‹ค.

์—…ํƒ€์ž„ ๋ฆฌ์„œ์น˜ ์ด๊ด„ ๋””๋ ‰ํ„ฐ์ธ ์•ค๋”” ๋กœ๋ Œ์Šค๋Š” ์„ฑ๋ช…์„ ํ†ตํ•ด โ€œํ•ต์‹ฌ ๋””์ง€ํ„ธ ์ธํ”„๋ผ๋Š” ๊ณ„์†ํ•ด์„œ ํ™•๋Œ€๋˜๊ณ  ์žˆ์ง€๋งŒ, AI๊ฐ€ ์ˆ˜์š” ๊ตฌ์กฐ๋ฅผ ์–ด๋–ป๊ฒŒ ๋ณ€ํ™”์‹œํ‚ฌ์ง€๋ฅผ ๋‘๊ณ  ๋ถˆํ™•์‹ค์„ฑ์ด ์ปค์ง€๋ฉด์„œ ํˆฌ์ž ๊ณ„ํš๊ณผ ํšŒ๋ณต ์ „๋žต ์ˆ˜๋ฆฝ์ด ํ•œ์ธต ์–ด๋ ค์›Œ์ง€๊ณ  ์žˆ๋‹ค. ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์„ค๊ณ„์™€ ๊ตฌ์ถ• ๋ฐฉ์‹๋„ ์ ์ฐจ ๋‹ค์–‘ํ™”๋˜๊ณ  ์žˆ์œผ๋ฉฐ, ํƒ„์†Œ ํฌ์ง‘ ๊ธฐ์ˆ , AI, ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ž๋™ํ™” ๋ถ„์•ผ์—์„œ ํˆฌ์ž์™€ ๊ธฐ์ˆ  ๊ฐœ๋ฐœ์ด ํ™•๋Œ€๋  ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒํ•œ๋‹คโ€๋ผ๊ณ  ์ „๋งํ–ˆ๋‹ค.

์‹ค์ œ๋กœ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๊ฑด์„ค ์†๋„์™€ ์ƒˆ๋กœ์šด ๋ฐœ์ „ ๋ฐ ์†ก์ „ ์„ค๋น„๋ฅผ ์‹ค์ œ๋กœ ๊ฐ€๋™ํ•˜๊ธฐ๊นŒ์ง€ ๊ฑธ๋ฆฌ๋Š” ์‹œ๊ฐ„ ์‚ฌ์ด์˜ ๊ฒฉ์ฐจ๊ฐ€ ์ ์ฐจ ๋ฒŒ์–ด์ง€๊ณ  ์žˆ๋‹ค๋Š” ์ ์ด ๊ฐ€์žฅ ํฐ ์šฐ๋ ค ์š”์ธ์œผ๋กœ ์ง€๋ชฉ๋๋‹ค.

๋กœ๋ Œ์Šค๋Š” โ€œ์˜ฌํ•ด ์ „๋ง์€ 2025๋…„ ์˜ˆ์ธก์„ ๋ฐ”ํƒ•์œผ๋กœ ํ•˜๋ฉฐ, ์—…๊ณ„์˜ ์ง€์†์ ์ธ ์„ฑ์žฅ๊ณผ ๊ทธ์— ๋”ฐ๋ฅธ ๊ณผ์ œ์— ์ดˆ์ ์„ ๋งž์ถ˜๋‹ค. ํ˜„์žฌ๋กœ์„  AI๊ฐ€ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์„ฑ์žฅ ์†๋„๋ฅผ ๋†’์ด๋Š” ๊ฐ•๋ ฅํ•œ ๋ณ€ํ™” ์š”์ธ์ธ ๊ฒƒ์œผ๋กœ ํ‰๊ฐ€ํ•˜๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ์—…ํƒ€์ž„์ด ์ œ์‹œํ•œ 2026๋…„ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์‚ฐ์—…์˜ ์ฃผ์š” ์ „๋ง 5๊ฐ€์ง€๋ฅผ ์†Œ๊ฐœํ•œ๋‹ค.

1. AI ์ธํ”„๋ผ ์—…์ฒด์˜ ์ ๋ฆผ ํ˜„์ƒ ์‹ฌํ™”

์—…ํƒ€์ž„์— ๋”ฐ๋ฅด๋ฉด, ๋งŽ์€ ๊ธฐ์—…์ด AI ๋„๊ตฌ๋ฅผ ์‹คํ—˜ํ•˜๋Š” ๋™์•ˆ AI๋ฅผ ๋’ท๋ฐ›์นจํ•˜๋Š” ์ธํ”„๋ผ๋Š” ๋ช‡๋ช‡ ์†Œ์ˆ˜ ์—…์ฒด์— ์ ์  ๋” ์ง‘์ค‘๋˜๋Š” ํ๋ฆ„์„ ๋ณด์ด๊ณ  ์žˆ๋‹ค.

์—…ํƒ€์ž„ ๋ฆฌ์„œ์น˜ ๋””๋ ‰ํ„ฐ์ธ ๋‹ค๋‹ˆ์—˜ ๋น„์กฐ๋Š” โ€œ๊ธฐ์—…๊ณผ ์†Œ๋น„์ž ์ „๋ฐ˜์—์„œ AI ๋„๊ตฌ ํ™œ์šฉ์ด ํ™•๋Œ€๋˜๊ณ  ์žˆ์ง€๋งŒ, ๊ทธ๋ ‡๋‹ค๊ณ  ํ•ด์„œ AI ์—ฐ์‚ฐ์ด๋‚˜ AI ์ธํ”„๋ผ๊นŒ์ง€ ๊ณ ๋ฅด๊ฒŒ ๋ถ„์‚ฐ๋˜๋Š” ๊ฒƒ์€ ์•„๋‹ˆ๋‹ค. AI ์—ฐ์‚ฐ ์ธํ”„๋ผ์˜ ์ง‘์ค‘๋„๋Š” ์•ž์œผ๋กœ ๋ช‡ ๋…„๊ฐ„ ์˜คํžˆ๋ ค ๋” ๋†’์•„์งˆ ๊ฒƒโ€์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

์—…ํƒ€์ž„์€ ๋Œ€๋ถ€๋ถ„ ๊ธฐ์—…์˜ AI ์ธํ”„๋ผ ํˆฌ์ž๊ฐ€ ๋น„๊ต์  ์ œํ•œ์ ์ธ ์ˆ˜์ค€์— ๋จธ๋ฌผ๊ณ  ์žˆ๋‹ค๊ณ  ๋ถ„์„ํ–ˆ๋‹ค. ๊ธฐ์—…์€ ์ถ”๋ก (inference)๊ณผ ์ผ๋ถ€ ํ•™์Šต์—๋งŒ ํˆฌ์ž๋ฅผ ์ง‘์ค‘ํ•˜๊ณ  ์žˆ์„ ๊ฐ€๋Šฅ์„ฑ์ด ํฌ๋ฉฐ, ์ถ”๋ก  ์ค‘์‹ฌ์˜ ์›Œํฌ๋กœ๋“œ๋Š” ๋Œ€๊ทœ๋ชจ ์šฉ๋Ÿ‰ ํ™•๋Œ€๋ฅผ ํ•„์š”๋กœ ํ•˜์ง€ ์•Š๋Š”๋‹ค๋Š” ์„ค๋ช…์ด๋‹ค.

๋น„์กฐ๋Š” โ€œAI ์—ฐ์‚ฐ ์ธํ”„๋ผ๊ฐ€ ์†Œ์ˆ˜ ์—…์ฒด์— ์ ๋ฆฌ๋Š” ํ๋ฆ„์€ ํ–ฅํ›„ ๋ช‡ ๋…„๊ฐ„ ์†Œํญ์ด์ง€๋งŒ ๊ณ„์† ๊ฐ•ํ™”๋  ๊ฒƒโ€์ด๋ผ๋ฉฐ โ€œ2026๋…„ ๋ง๊นŒ์ง€ ์ƒ์„ฑํ˜• AI์™€ ์ธ์ ‘ ์›Œํฌ๋กœ๋“œ๋ฅผ ์šด์˜ํ•˜๊ธฐ ์œ„ํ•ด ์ „ ์„ธ๊ณ„ ๋ฐ์ดํ„ฐ์„ผํ„ฐ์— ์•ฝ 10๊ธฐ๊ฐ€์™€ํŠธ(GW)์˜ ์ƒˆ๋กœ์šด IT ๋ถ€ํ•˜๊ฐ€ ์ถ”๊ฐ€๋  ๊ฒƒโ€๋ผ๊ณ  ์ „๋งํ–ˆ๋‹ค. ๊ทธ๋Š” ์ด์–ด โ€œ์ „ ์„ธ๊ณ„์ ์œผ๋กœ ์•ฝ 1,300๋งŒ~1,500๋งŒ ๊ฐœ์˜ GPU์™€ ๊ฐ€์†๊ธฐ๊ฐ€ ๋ฐฐ์น˜๋œ๋‹ค๋Š” ์˜๋ฏธ์ด๋ฉฐ, ์ด๋“ค ๋Œ€๋ถ€๋ถ„์€ ์Šˆํผ์ปดํ“จํŒ… ๋ฐฉ์‹์œผ๋กœ ๊ตฌ์ถ•๋  ๊ฒƒ์œผ๋กœ ๋ณธ๋‹คโ€๋ผ๊ณ  ๋งํ–ˆ๋‹ค.

2. ๋ฐ์ดํ„ฐ์„ผํ„ฐ ํ™•์žฅ ์†๋„๋ฅผ ๋”ฐ๋ผ๊ฐ€์ง€ ๋ชปํ•˜๋Š” ์ „๋ ฅ ๊ณต๊ธ‰

์—…ํƒ€์ž„์€ ๊ฐ€์žฅ ์‹œ๊ธ‰ํ•œ ๊ณผ์ œ๋กœ, ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ 3๋…„ ์ด๋‚ด์— ๊ฑด์„คํ•  ์ˆ˜ ์žˆ๋Š” ๋ฐ˜๋ฉด ๋ฐœ์ „ ์„ค๋น„ ๊ตฌ์ถ•์—๋Š” ํ›จ์”ฌ ๋” ์˜ค๋žœ ์‹œ๊ฐ„์ด ๊ฑธ๋ฆฐ๋‹ค๋Š” ์ ์„ ๊ผฝ์•˜๋‹ค.

์—…ํƒ€์ž„ ๋ฆฌ์„œ์น˜ ์• ๋„๋ฆฌ์ŠคํŠธ์ธ ๋งฅ์Šค ์Šค๋ชฐ๋ฝ์Šค๋Š” โ€œํƒœ์–‘๊ด‘์ด๋‚˜ ํ’๋ ฅ ๋ฐœ์ „ ๋‹จ์ง€๋ฅผ ๊ตฌ์ถ•ํ•˜๋Š” ๋ฐ๋Š” 3~6๋…„์ด ๊ฑธ๋ฆฌ๊ณ , ๋ณตํ•ฉํ™”๋ ฅ๋ฐœ์ „์†Œ๋Š” ์•ฝ 6๋…„์ด ์†Œ์š”๋œ๋‹ค. ๊ธฐ์กด ์›์ž๋ ฅ๋ฐœ์ „์†Œ์˜ ๊ฒฝ์šฐ ๋‚™๊ด€์ ์œผ๋กœ ๋ณด๋”๋ผ๋„ 10๋…„ ์ด์ƒ์ด ํ•„์š”ํ•˜๋‹คโ€๋ผ๊ณ  ์ง„๋‹จํ–ˆ๋‹ค.

์—…ํƒ€์ž„์€ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๊ทœ๋ชจ๊ฐ€ ์ƒ๋Œ€์ ์œผ๋กœ ์ž‘๊ณ  ์„ฑ์žฅ ํ๋ฆ„์ด ์˜ˆ์ธก ๊ฐ€๋Šฅํ•˜๋˜ ์‹œ๊ธฐ์—๋Š” ์ด๋Ÿฌํ•œ ๊ฒฉ์ฐจ๋ฅผ ๊ด€๋ฆฌํ•  ์ˆ˜ ์žˆ์—ˆ์ง€๋งŒ, ํ˜„์žฌ๋Š” ํ”„๋กœ์ ํŠธ ๊ทœ๋ชจ๊ฐ€ ์ˆ˜์‹ญ์—์„œ ๋งŽ๊ฒŒ๋Š” ์ˆ˜๋ฐฑ ๋ฉ”๊ฐ€์™€ํŠธ ๋‹จ์œ„๋กœ ์ปค์ง€๋ฉด์„œ ๋‹จ๊ธฐ๊ฐ„์— ํ•„์š”ํ•œ ์ „๋ ฅ์„ ํ™•๋ณดํ•˜๋Š” ๊ฒƒ์ด ๊ฑฐ์˜ ๋ถˆ๊ฐ€๋Šฅํ•ด์กŒ๋‹ค๊ณ  ์ง€์ ํ–ˆ๋‹ค.

์Šค๋ชฐ๋ฝ์Šค๋Š” โ€œ๋งŽ์€ ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ 2026๋…„๋ถ€ํ„ฐ ๋ณธ๊ฒฉ์ ์œผ๋กœ ๋“œ๋Ÿฌ๋‚  ์œ„๊ธฐ์˜ ๊ทœ๋ชจ์™€ ์‹ฌ๊ฐ์„ฑ์— ์ถฉ๋ถ„ํžˆ ๋Œ€๋น„ํ•˜์ง€ ๋ชปํ•œ ์ƒํƒœ์ผ ๊ฐ€๋Šฅ์„ฑ์ด ๋†’๋‹ค. ๋ฌด์—‡์„, ์–ด๋–ป๊ฒŒ, ์–ด๋””์— ๊ฑด์„คํ•  ์ˆ˜ ์žˆ๋Š”์ง€๋ฅผ ๊ฒฐ์ •ํ•˜๋Š” ํ•ต์‹ฌ ์š”์†Œ๋Š” ๋ฐœ์ „๊ณผ ๋ฐฐ์ „ ์„ค๋น„๊ฐ€ ๋  ๊ฒƒ์ด๋ฉฐ, ์ด๋Ÿฌํ•œ ์ „๋ ฅ ์œ„๊ธฐ๋Š” ์ˆ˜๋…„๊ฐ„ ์ด์–ด์งˆ ์ˆ˜ ์žˆ๋‹คโ€๋ผ๊ณ  ๋‚ด๋‹ค๋ดค๋‹ค.

์—…ํƒ€์ž„์€ ์ฃผ์š” ์ง€์—ญ์—์„œ ์ „๋ ฅ ๋น„์šฉ์ด ์ƒ์Šนํ•ด ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๊ฑด์„ค์‚ฌ๊ฐ€ ์ ์ฐจ 2์ฐจ ์‹œ์žฅ์œผ๋กœ ๋ˆˆ์„ ๋Œ๋ฆด ๊ฒƒ์œผ๋กœ ์ „๋งํ–ˆ๋‹ค. ๋˜ํ•œ ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ์ „๋ ฅ๋ง์— ์—ฐ๊ฒฐ๋˜๋Š” ์กฐ๊ฑด์œผ๋กœ, ์ „๋ ฅ ์ˆ˜์š”๊ฐ€ ์ง‘์ค‘๋˜๋Š” ์‹œ๊ฐ„๋Œ€์—๋Š” ์™ธ๋ถ€ ์ „๋ ฅ ์‚ฌ์šฉ์„ ์ค„์ด๊ณ  ์ž์ฒด ๋ฐœ์ „ ์„ค๋น„๋ฅผ ๊ฐ€๋™ํ•˜๋Š” ๋“ฑ ์ˆ˜์š” ์กฐ์ ˆ์— ํ˜‘๋ ฅํ•ด์•ผ ํ•  ๊ฐ€๋Šฅ์„ฑ๋„ ๋†’๋‹ค๊ณ  ๋ถ„์„ํ–ˆ๋‹ค.

3. ํƒ„์†Œ ํฌ์ง‘ ๊ธฐ์ˆ ์— ์ฃผ๋ชฉ

์ผ๋ถ€ ๋ฐ์ดํ„ฐ์„ผํ„ฐ๋Š” ์ฒœ์—ฐ๊ฐ€์Šค ๋ฐœ์ „์„ ํ†ตํ•ด ์ „๋ ฅ์„ ํ™•๋ณดํ•˜๋ ค ํ•˜๊ณ  ์žˆ์ง€๋งŒ, ๊ทธ ๊ณผ์ •์—์„œ ๋˜ ๋‹ค๋ฅธ ๊ณผ์ œ์— ์ง๋ฉดํ•˜๊ณ  ์žˆ๋‹ค. ์„ฑ์žฅ์— ํ•„์š”ํ•œ ์ „๋ ฅ ๊ณต๊ธ‰์„ ์šฐ์„ ํ•  ๊ฒƒ์ธ์ง€, ์•„๋‹ˆ๋ฉด ๊ฐœ๋ณ„ ์‹œ์„ค์— ์„ค์ •๋œ ํƒ„์†Œ์ค‘๋ฆฝ ๋ชฉํ‘œ๋ฅผ ์ถฉ์กฑํ•  ๊ฒƒ์ธ์ง€๋ฅผ ๋‘๊ณ  ํŒ๋‹จํ•ด์•ผ ํ•˜๋Š” ์ƒํ™ฉ์ด๋‹ค.

์ด๋Ÿฐ ์ƒํ™ฉ์—์„œ ์—…ํƒ€์ž„์€ ํƒ„์†Œ ํฌ์ง‘ ๋ฐ ์ €์žฅ(CCS) ์‹œ์Šคํ…œ์ด 2026๋…„ ์ดํ›„ ์ด๋ก ์  ๋…ผ์˜๋ฅผ ๋„˜์–ด ์‹ค์šฉ ๋‹จ๊ณ„๋กœ ์ „ํ™˜๋  ๊ฒƒ์œผ๋กœ ์ „๋งํ–ˆ๋‹ค. CCS๋Š” ๋ฐœ์ „์ด๋‚˜ ์‚ฐ์—… ๊ณต์ •์—์„œ ๋ฐœ์ƒํ•œ ์ด์‚ฐํ™”ํƒ„์†Œ๋ฅผ ๋ฐฐ์ถœํ•˜์ง€ ์•Š๊ณ  ํฌ์ง‘ํ•ด ์ €์žฅํ•˜๊ฑฐ๋‚˜ ๋‹ค๋ฅธ ์žฅ์†Œ๋กœ ์šด์†กํ•ด ์ฒ˜๋ฆฌํ•˜๋Š” ๊ธฐ์ˆ ์ด๋‹ค. ์ด๋Š” ๊ทธ๋™์•ˆ ๊ฒฝ์ œ์„ฑ์ด ๋‚ฎ๋‹ค๋Š” ํ‰๊ฐ€๋ฅผ ๋ฐ›์•„์™”์ง€๋งŒ, ์—…ํƒ€์ž„์€ ๊ธฐ์ˆ  ์„ฑ์ˆ™๋„ ํ–ฅ์ƒ๊ณผ ๋Œ€์•ˆ ์—๋„ˆ์ง€์›์˜ ๋น„์šฉ ์ƒ์Šน, ํƒ„์†Œ ์ƒ์‡„ ๋น„์šฉ ์ฆ๊ฐ€ ๋“ฑ์ด ๋งž๋ฌผ๋ฆฌ๋ฉด์„œ CCS์˜ ๊ฒฝ์Ÿ๋ ฅ์ด ์žฌํ‰๊ฐ€๋˜๊ณ  ์žˆ๋‹ค๊ณ  ๋ถ„์„ํ–ˆ๋‹ค. ๋‹ค๋งŒ ํฌ์ง‘ํ•œ ํƒ„์†Œ๋ฅผ ํ˜„์ง€์— ์ €์žฅํ•  ์ˆ˜ ์žˆ๊ฑฐ๋‚˜, ์ €๋น„์šฉ์œผ๋กœ ์šด์†ก์ด ๊ฐ€๋Šฅํ•œ ์ง€์—ญ์—์„œ๋งŒ ํ˜„์‹ค์ ์ธ ์„ ํƒ์ง€๊ฐ€ ๋  ์ˆ˜ ์žˆ๋‹ค๊ณ  ๋ง๋ถ™์˜€๋‹ค.

๋กœ๋ Œ์Šค๋Š” โ€œ์ด๋ฅผ ๊ตฌํ˜„ํ•˜๋Š” ํ•œ ๊ฐ€์ง€ ๋ฐฉ์‹์ด ์ฒœ์—ฐ๊ฐ€์Šค ๋ฐœ์ „์ด๋ฉฐ, ์ด๋Š” ์ „๋ ฅ๋ง์— ์—ฐ๊ฒฐ๋œ ๊ฐ€์Šค ํ„ฐ๋นˆ์ด๋‚˜ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๋ถ€์ง€ ๋‚ด์— ์„ค์น˜๋œ ์ž์ฒด ๊ฐ€์Šค ํ„ฐ๋นˆ ํ˜•ํƒœ๋กœ ์šด์˜๋˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ๋‹ค. ์ตœ๊ทผ CCS ๊ธฐ์ˆ ์„ ํ†ตํ•ฉํ•  ์ˆ˜ ์žˆ๋Š” ๊ธฐ์—…๊ณผ ํ˜‘๋ ฅํ•˜๋ ค๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ๋ˆˆ์— ๋„๊ฒŒ ๋Š˜๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ๊ทธ๋Š” ์ด์–ด โ€œ์ด๋ฒˆ ์ „๋ง์˜ ํ•ต์‹ฌ์€ ์ผ๋ถ€ ์ง€์—ญ์—์„œ CCS๊ฐ€ ๋Œ€๊ทœ๋ชจ๋กœ ์ ์šฉ ๊ฐ€๋Šฅํ•œ ํ˜„์‹ค์ ์ธ ์„ ํƒ์ง€๋กœ ๋ฐ”๋€Œ๊ณ  ์žˆ๋‹ค๋Š” ์ โ€์ด๋ผ๋ฉฐ, โ€œ๊ธฐ์ˆ  ๋ฐœ์ „๊ณผ ๋Œ€์•ˆ ์—๋„ˆ์ง€์›์˜ ๋†’์€ ๋น„์šฉ ๋“ฑ ์—ฌ๋Ÿฌ ์š”์ธ์ด ๋งž๋ฌผ๋ฆฌ๋ฉด์„œ CCS๊ฐ€ ์‹ค์ œ๋กœ ๋„์ž…ํ•  ์ˆ˜ ์žˆ๋Š” ๊ธฐ์ˆ ๋กœ ์ž๋ฆฌ ์žก๊ณ  ์žˆ๊ณ , ํŠน์ • ํ™˜๊ฒฝ์—์„œ๋Š” ์ถฉ๋ถ„ํžˆ ํ™•์žฅ ๊ฐ€๋Šฅํ•œ ํ•ด๋ฒ•์ด ๋˜๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ๋ถ„์„ํ–ˆ๋‹ค.

4. ๊ทœ๋ชจ ํ™•๋Œ€๊ฐ€ ์ƒˆ๋กœ์šด ๊ณผ์ œ๋กœ ๋ถ€์ƒ

์—…ํƒ€์ž„์€ ๋ฐ์ดํ„ฐ์„ผํ„ฐ์˜ ๊ทœ๋ชจ๊ฐ€ ์ปค์ง€๊ณ  ํŠน์ • ์ง€์—ญ์— ์ง‘์ค‘๋ ์ˆ˜๋ก ์šด์˜ ์•ˆ์ •์„ฑ ๋ฆฌ์Šคํฌ๋„ ํ•จ๊ป˜ ์ปค์ง€๊ณ  ์žˆ๋‹ค๊ณ  ๊ฒฝ๊ณ ํ–ˆ๋‹ค.

๋กœ๋ Œ์Šค๋Š” โ€œ๊ฐœ๋ณ„ ๋ฐ์ดํ„ฐ์„ผํ„ฐ์˜ ๋ฌผ๋ฆฌ์  ๊ทœ๋ชจ๋ฟ ์•„๋‹ˆ๋ผ, ๋Œ€๊ทœ๋ชจ ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ํŠน์ • ์ง€์—ญ์— ๋ฐ€์ง‘๋˜๋Š” ํ˜„์ƒ ์ž์ฒด๊ฐ€ ์ƒ๋‹นํ•œ ๋ฆฌ์Šคํฌ๋ฅผ ๋งŒ๋“ค์–ด๋‚ด๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

ํŠนํžˆ ๋…ธํ›„ํ™”๋œ ์†ก์ „ ์„ค๋น„์™€ ๊ฐ„ํ—์ ์œผ๋กœ ๋ฐœ์ „๋˜๋Š” ์žฌ์ƒ ์—๋„ˆ์ง€์˜ ๋น„์ค‘ ํ™•๋Œ€๊ฐ€ ์ „๋ ฅ๋ง ๋ถˆ์•ˆ์ •์„ฑ์„ ํ‚ค์šฐ๊ณ  ์žˆ์œผ๋ฉฐ, ๋Œ€๊ทœ๋ชจ ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ํ•œ ์ง€์—ญ์— ๋ชฐ๋ ค ์ „๋ ฅ๋ง ์•ˆ์ •์„ฑ์— ์ง์ ‘์ ์ธ ์˜ํ–ฅ์„ ๋ฏธ์น˜๊ณ  ์žˆ๋‹ค. ์—…ํƒ€์ž„์€ ์ด๋Ÿฌํ•œ ๋ณ€ํ™”์— ๊ทœ์ œ ๋‹น๊ตญ๋„ ์ฃผ๋ชฉํ•˜๊ณ  ์žˆ๋‹ค๊ณ  ์ „ํ–ˆ๋‹ค. ์‹ค์ œ๋กœ ์—ฌ๋Ÿฌ ๊ตญ๊ฐ€์—์„œ ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ๋” ํฐ ์ „์•• ๋ณ€๋™์„ ๊ฒฌ๋”œ ์ˆ˜ ์žˆ๋„๋ก ์š”๊ตฌํ•˜๊ณ , ์ „๋ ฅ๋ง ์ด์ƒ ๋ฐœ์ƒ ์‹œ ๋ถ€ํ•˜ ์ฐจ๋‹จ์„ ํ”ผํ•˜๊ฑฐ๋‚˜ ์ง€์—ฐํ•˜๋„๋ก ํ•˜๋Š” ์ƒˆ๋กœ์šด ๊ทœ์ œ๋ฅผ ๊ฒ€ํ† ํ•˜๊ณ  ์žˆ๋‹ค.

5. ๋ฐ์ดํ„ฐ์„ผํ„ฐ AI ์ž๋™ํ™”, ์‹ค์šด์˜ ๋‹จ๊ณ„๋กœ ์ „ํ™˜

2026๋…„์—๋Š” AI ๊ธฐ๋ฐ˜ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ž๋™ํ™”๋„ ์‹คํ—˜ ๋‹จ๊ณ„๋ฅผ ๋ฒ—์–ด๋‚˜, ์ผ์ƒ์ ์ธ ์šด์˜์„ ์ง€์›ํ•˜๋Š” ์ˆ˜์ค€์œผ๋กœ ์ „ํ™˜๋  ์ „๋ง์ด๋‹ค.

์—…ํƒ€์ž„์˜ ๋ฆฌ์„œ์น˜ ์• ๋„๋ฆฌ์ŠคํŠธ์ธ ๋žœ๋“œ ํƒˆ๋ฆฌ๋ธŒ๋Š” โ€œ2026๋…„์—๋Š” AI์— ๊ธฐ๋ฐ˜ํ•œ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์šด์˜์ด ๋ชฉํ‘œ ์ง€ํ–ฅ์ ์œผ๋กœ ์ดˆ๊ธฐ ๋‹จ๊ณ„์— ์ ‘์–ด๋“ค ๊ฒƒโ€์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ๊ทธ๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ž๋™ํ™”๊ฐ€ ๋ƒ‰๊ฐ ๋ฐ ์ „๋ ฅ ์ตœ์ ํ™”๋ฅผ ์œ„ํ•œ ๊ฐ•ํ™” ํ•™์Šต, ํ•˜์ด๋ธŒ๋ฆฌ๋“œ ๋””์ง€ํ„ธ ํŠธ์œˆ, ์—…๋ฌด ํ๋ฆ„์„ ์ง€์›ํ•˜๋Š” ์‚ฐ์—…์šฉ ์ฝ”ํŒŒ์ผ๋Ÿฟ, ์‹ค์‹œ๊ฐ„ ์„ผ์„œ ๋ฐ์ดํ„ฐ์— ๋Œ€์‘ํ•ด ์ž๋™์œผ๋กœ ๋ฐ˜์‘ํ•˜๋Š” ๊ณ ๋„ํ™”๋œ ๊ทœ์น™ ๊ธฐ๋ฐ˜ ์˜ค์ผ€์ŠคํŠธ๋ ˆ์ด์…˜ ๋“ฑ์„ ์ค‘์‹ฌ์œผ๋กœ ์ „๊ฐœ๋  ๊ฒƒ์ด๋ผ๊ณ  ๋‚ด๋‹ค๋ดค๋‹ค.

๊ทธ๋Š” โ€œ์•ž์œผ๋กœ 1~2๋…„ ๋™์•ˆ ์—…๊ณ„๋Š” ์™„์ „ํ•œ ๋ฌด์ธยท์ž์œจ ์šด์˜์œผ๋กœ ๊ฐ€๊ธฐ๋ณด๋‹ค๋Š”, ์‚ฌ๋žŒ์ด ๊ด€๋ฆฌยท๊ฐ๋…ํ•˜๋Š” ๋ฒ”์œ„ ์•ˆ์—์„œ ์‹ค์šฉ์ ์ธ ์ž๋™ํ™”๋ฅผ ๋‹จ๊ณ„์ ์œผ๋กœ ๋„์ž…ํ•˜๋Š” ๋ฐฉํ–ฅ์œผ๋กœ ์›€์ง์ผ ๊ฒƒโ€์ด๋ผ๋ฉฐ โ€œ์ด๋Ÿฌํ•œ ์ž๋™ํ™”๋Š” ๋‹น์žฅ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์šด์˜์„ ๋Œ€์ฒดํ•˜๊ธฐ๋ณด๋‹ค๋Š” ์—…๋ฌด ๋ถ€๋‹ด์„ ์ค„์ด๊ณ  ๋” ์•ˆ์ •์ ์œผ๋กœ ์šด์˜๋˜๋„๋ก ์ง€์›ํ•˜๋Š” ๋ฐ ๋ชฉ์ ์ด ์žˆ๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

์ดˆ๋Œ€ํ˜• ํด๋ผ์šฐ๋“œ ์—…์ฒด์™€ ์ฝ”๋กœ์ผ€์ด์…˜ ์—…์ฒด๋Š” ๋น„๊ต์  ๋น ๋ฅด๊ฒŒ AI ์ž๋™ํ™” ๋„์ž…์„ ์ถ”์ง„ํ•  ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒ๋˜์ง€๋งŒ, ๋Œ€๋ถ€๋ถ„์˜ ์—…์ฒด๋Š” ์ ์ง„์ ์ธ ์—…๊ทธ๋ ˆ์ด๋“œ๋ฅผ ํ†ตํ•ด ์‹ ์ค‘ํ•˜๊ฒŒ ์ ‘๊ทผํ•  ๊ฐ€๋Šฅ์„ฑ์ด ํฌ๋‹ค. ์—…ํƒ€์ž„์€ ์ „๋ ฅ ์ œ์•ฝ์„ ์™„ํ™”ํ•  ์ˆ˜ ์žˆ๋Š” ๋น„๊ต์  ์‰ฌ์šด ํ•ด๋ฒ•์ด ์ด๋ฏธ ๋Œ€๋ถ€๋ถ„ ์†Œ์ง„๋๋‹ค๋ฉฐ, AI ์ง€์› ๊ฒฝ์Ÿ์ด ์ „๋ ฅ๋ง๊ณผ ๊ณต๊ธ‰๋ง, ๊ธฐ์กด์˜ ์šด์˜ ์•ˆ์ •์„ฑ ์ „๋žต์— ๋ถ€๋‹ด์„ ์ฃผ๊ณ  ์žˆ๋‹ค๊ณ  ์ง€์ ํ–ˆ๋‹ค.

๋กœ๋ Œ์Šค๋Š” โ€œAI๊ฐ€ ์ง€๋ฐฐ์ ์ธ ํ™”๋‘๋กœ ๋– ์˜ค๋ฅธ ์ง€๋Š” ์˜ค๋ž˜๋˜์ง€ ์•Š์•˜๋‹คโ€๋ผ๋ฉฐ โ€œ๋งŽ์€ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์—…์ฒด๊ฐ€ ๋ช…ํ™•ํ•œ AI ์ˆ˜์š”๊ฐ€ ํ™•์ธ๋˜๊ธฐ ์ „๊นŒ์ง€๋Š” ์‹ ์ค‘ํ•œ ๊ณ„ํš ๋‹จ๊ณ„์— ๋จธ๋ฌผ๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ์ „ํ–ˆ๋‹ค.
dl-ciokorea@foundryco.com

โ€œAI ๋ฐ์ดํ„ฐ์„ผํ„ฐ ํ™•์žฅ์˜ ์ตœ๋Œ€ ๊ฑธ๋ฆผ๋Œ์€ ์ „๋ ฅ๋ง ์—ฐ๊ฒฐ ์ง€์—ฐโ€ ๊ตฌ๊ธ€

๊ตฌ๊ธ€์€ ์ „๋ ฅ๋ง ์—ฐ๊ฒฐ ์ง€์—ฐ์ด ๋ฐ์ดํ„ฐ์„ผํ„ฐ์— ์ „๋ ฅ์„ ๊ณต๊ธ‰ํ•˜๋Š” ๋ฐ ๊ฐ€์žฅ ํฐ ์žฅ์• ๋ฌผ์ด ๋˜๊ณ  ์žˆ๋‹ค๋ฉฐ, ์ง€์—ฐ ๊ธฐ๊ฐ„์ด ์ตœ๋Œ€ 12๋…„์— ์ด๋ฅผ ์ˆ˜ ์žˆ๋‹ค๊ณ  ๋‚ด๋‹ค๋ดค๋‹ค. AI ์›Œํฌ๋กœ๋“œ๊ฐ€ ๊ธ‰์ฆํ•˜๋Š” ๊ฐ€์šด๋ฐ, ๊ตฌ๊ธ€์€ ํด๋ผ์šฐ๋“œ ์šฉ๋Ÿ‰ ํ™•์žฅ์„ ์ œํ•œํ•˜๋Š” ์ธํ”„๋ผ ์œ„๊ธฐ๊ฐ€ ๋ณธ๊ฒฉํ™”๋˜๊ณ  ์žˆ๋‹ค๊ณ  ๊ฐ•์กฐํ–ˆ๋‹ค.

๊ตฌ๊ธ€์˜ ๊ธ€๋กœ๋ฒŒ ์ง€์†๊ฐ€๋Šฅ์„ฑ ๋ฐ ๊ธฐํ›„ ์ •์ฑ… ์ด๊ด„์ธ ๋งˆ์Šค๋“  ํ•œ๋‚˜๋Š” ์ง€๋‚œ 15์ผ ์—ด๋ฆฐ ๋ฏธ๊ตญ๊ธฐ์—…์—ฐ๊ตฌ์†Œ(AEI) ํ–‰์‚ฌ์—์„œ โ€œ์ „๋ ฅ๋ง์˜ ํ•ต์‹ฌ ๊ณผ์ œ๋Š” ์†ก์ „ ์žฅ๋ฒฝโ€์ด๋ผ๋ฉฐ, ์—ฌ๋Ÿฌ ์ง€์—ญ์˜ ์ „๋ ฅ ๊ธฐ์—…์ด ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์‹ ๊ทœ ์—ฐ๊ฒฐ๊นŒ์ง€ 4๋…„์—์„œ 10๋…„์ด ๊ฑธ๋ฆฐ๋‹ค๊ณ  ์•ˆ๋‚ดํ•˜๊ณ  ์žˆ๋‹ค๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ๊ทธ์ค‘ ์ผ๋ถ€ ๊ธฐ์—…์€ ์ „๋ ฅ๋ง ์—ฐ๊ฒฐ ์š”์ฒญ์„ ๊ฒ€ํ† ํ•˜๋Š” ๋ฐ๋งŒ 12๋…„์ด ํ•„์š”ํ•˜๋‹ค๊ณ  ํ†ต๋ณดํ•œ ๊ฒƒ์œผ๋กœ ์•Œ๋ ค์กŒ๋‹ค.

์‹ค์ œ๋กœ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ „๋ ฅ ์†Œ๋น„๊ฐ€ ๊ธ‰๊ฒฉํžˆ ๋Š˜์–ด๋‚  ๊ฒƒ์ด๋ผ๋Š” ์ „๋ง์€ ๊พธ์ค€ํžˆ ์ œ๊ธฐ๋˜๊ณ  ์žˆ๋‹ค. ์ง€๋‚œํ•ด 12์›” ๋กœ๋ Œ์Šค๋ฒ„ํด๋ฆฌ๊ตญ๋ฆฝ์—ฐ๊ตฌ์†Œ๊ฐ€ ๋ฐœํ‘œํ•œ ์—ฐ๊ตฌ์— ๋”ฐ๋ฅด๋ฉด, ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ „๋ ฅ ์‚ฌ์šฉ๋Ÿ‰์€ 2023๋…„ 176ํ…Œ๋ผ์™€ํŠธ์‹œ(TWh)์—์„œ 2028๋…„ 325~580TWh ์ˆ˜์ค€์œผ๋กœ ์ฆ๊ฐ€ํ•  ์ „๋ง์ด๋‹ค. ํ•˜์ง€๋งŒ ์ „๋ ฅ๋ง์€ ์ด์ฒ˜๋Ÿผ ์ฆ๊ฐ€ํ•˜๋Š” ์ˆ˜์š”๋ฅผ ๊ฐ๋‹นํ•˜์ง€ ๋ชปํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ๋ฏธ๊ตญ์—์„œ๋งŒ ๋ฐœ์ „ ๋ฐ ์—๋„ˆ์ง€ ์ €์žฅ ์„ค๋น„ ๋“ฑ ์•ฝ 2,300๊ธฐ๊ฐ€์™€ํŠธ(GW) ๊ทœ๋ชจ์˜ ์ „๋ ฅ ์šฉ๋Ÿ‰์ด ์•„์ง๋„ ์ „๋ ฅ๋ง ์—ฐ๊ฒฐ์„ ๊ธฐ๋‹ค๋ฆฌ๊ณ  ์žˆ๋Š” ์ƒํ™ฉ์ด๋‹ค.

๊ธฐ์—…์˜ IT ์ธํ”„๋ผ๋ฅผ ์ด๊ด„ํ•˜๋Š” CIO์—๊ฒŒ ์ „๋ ฅ๋ง ๋ฌธ์ œ๋Š” ํด๋ผ์šฐ๋“œ ์šฉ๋Ÿ‰ ํ™•์žฅ์— ๋Œ€ํ•œ ๊ธฐ์กด ์ธ์‹์„ ๋ฐ”๊ฟ”๋†“๊ณ  ์žˆ๋‹ค. ๊ทธ๋ ˆ์ดํ•˜์šด๋“œ ๋ฆฌ์„œ์น˜์˜ ์ˆ˜์„ ์• ๋„๋ฆฌ์ŠคํŠธ ์‚ฐ์นซ ๋น„๋ฅด ๊ณ ๊ธฐ์•„๋Š” โ€œ์ด์ œ ํด๋ผ์šฐ๋“œ๊ฐ€ ๋ฌดํ•œํ•˜๋‹ค๋Š” ์ „์ œ๋ฅผ ๋ฒ„๋ ค์•ผ ํ•œ๋‹ค. ์ง€๊ธˆ์˜ ๋ณ‘๋ชฉ์€ ๋ฐ˜๋„์ฒด๋‚˜ ์†Œํ”„ํŠธ์›จ์–ด๊ฐ€ ์•„๋‹ˆ๋ผ ์ „๊ธฐ ๊ณต๊ธ‰โ€์ด๋ผ๊ณ  ์ง€์ ํ–ˆ๋‹ค.

์ „๋ ฅ๋ง ์ „๋ฐ˜์œผ๋กœ ํ™•์‚ฐ๋˜๋Š” ๊ตฌ์กฐ์  ์ง€์—ฐ

๊ตฌ๊ธ€์ด ๊ฒช๊ณ  ์žˆ๋Š” ์—ฐ๊ฒฐ ์ง€์—ฐ์€ ํŠน์ • ๊ธฐ์—…์ด ์•„๋‹ˆ๋ผ ์ „๋ ฅ๋ง ์ „๋ฐ˜์— ๋ˆ„์ ๋œ ๋ฌธ์ œ๋‹ค. ๋ฒ„ํด๋ฆฌ๊ตญ๋ฆฝ์—ฐ๊ตฌ์†Œ ์ž๋ฃŒ์— ๋”ฐ๋ฅด๋ฉด ์ „๋ ฅ๋ง ์—ฐ๊ฒฐ ๋Œ€๊ธฐ ๊ธฐ๊ฐ„์€ 2000~2007๋…„์— ์ถ”์ง„๋œ ํ”„๋กœ์ ํŠธ์˜ ๊ฒฝ์šฐ 2๋…„์ด ์ฑ„ ๊ฑธ๋ฆฌ์ง€ ์•Š์•˜์ง€๋งŒ, 2018~2024๋…„์—๋Š” ํ‰๊ท  4๋…„ ์ด์ƒ์œผ๋กœ 2๋ฐฐ ๋„˜๊ฒŒ ๋Š˜์–ด๋‚ฌ๋‹ค. 2000~2019๋…„ ์‚ฌ์ด ์ „๋ ฅ๋ง ์—ฐ๊ฒฐ์„ ์‹ ์ฒญํ•œ ์ „์ฒด ์„ค๋น„ ๊ฐ€์šด๋ฐ 2024๋…„ ๋ง๊นŒ์ง€ ์‹ค์ œ ์ƒ์—… ์šด์ „์— ๋„๋‹ฌํ•œ ๋น„์œจ์€ 13%์— ๋ถˆ๊ณผํ–ˆ๋‹ค.

์ด ๊ฐ™์€ ์ง€์—ฐ์˜ ๊ทผ๋ณธ ์›์ธ์œผ๋กœ๋Š” ๊ธ‰์ฆํ•˜๋Š” ์ „๋ ฅ ์ˆ˜์š”๋ฅผ ๊ฐ๋‹นํ•˜์ง€ ๋ชปํ•˜๋Š” ๋…ธํ›„ ์†ก์ „ ์ธํ”„๋ผ๊ฐ€ ์ง€๋ชฉ๋๋‹ค. ํ•œ๋‚˜๋Š” โ€œ์ง€์—ญ ๊ฐ„ ์†ก์ „์„ ๋กœ๋ฅผ ์ƒˆ๋กœ ๊ตฌ์ถ•ํ•˜๋ ค๋ฉด ํ—ˆ๊ฐ€ ์ ˆ์ฐจ๋งŒ 7~11๋…„์ด ์†Œ์š”๋œ๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ๊ทธ๋Š” โ€œSPP(Southwest Power Pool) ์˜ˆ์ธก์— ๋”ฐ๋ฅด๋ฉด ์†ก์ „ ์ธํ”„๋ผ ํ™•์ถฉ ์†๋„๊ฐ€ ์ง€๊ธˆ์ฒ˜๋Ÿผ ์ˆ˜์š”๋ฅผ ๋”ฐ๋ผ๊ฐ€์ง€ ๋ชปํ•  ๊ฒฝ์šฐ, ์ „๋ ฅ์„ ๊ณต๊ธ‰ํ•˜์ง€ ๋ชปํ•˜๋Š” โ€˜๋ถ€ํ•˜ ์†์‹คโ€™ ์ƒํ™ฉ์ด ์ตœ๋Œ€ 115์ผ์— ์ด๋ฅผ ์ˆ˜ ์žˆ๋‹คโ€๋ผ๊ณ  ์–ธ๊ธ‰ํ–ˆ๋‹ค.

์ „๋ ฅ๋ง ์ „๋ฐ˜์— ๊ฑธ์นœ ์ง€์—ฐ์€ ๊ธฐ์—…์œผ๋กœ ํ•˜์—ฌ๊ธˆ ํด๋ผ์šฐ๋“œ ์šฉ๋Ÿ‰์— ๋Œ€ํ•œ ๊ธฐ๋ณธ ๊ฐ€์ •์„ ์žฌ๊ณ ํ•˜๊ฒŒ ํ•˜๊ณ  ์žˆ๋‹ค. ๋ถ๋ถ€ ๋ฒ„์ง€๋‹ˆ์•„์™€ ์‚ฐํƒ€ํด๋ผ๋ผ์ฒ˜๋Ÿผ ์ดˆ๋Œ€ํ˜• ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๊ตฌ์ถ•์˜ ์ตœ์ ์ง€๋กœ ๊ผฝํžˆ๋˜ ์ง€์—ญ์กฐ์ฐจ ๊ณต๊ธ‰ํ•  ์ „๋ ฅ์ด ๋น ๋ฅด๊ฒŒ ์†Œ์ง„๋˜๊ณ  ์žˆ๋Š” ์ƒํ™ฉ์ด๋‹ค.

์ธํ”„๋ผ ์ œ์•ฝ์€ ํด๋ผ์šฐ๋“œ ์‹œ์žฅ์˜ ๊ฒฝ์Ÿ ๊ตฌ๋„๋„ ๋ฐ”๊พธ๊ณ  ์žˆ์œผ๋ฉฐ, ๊ธฐ์ˆ  ์—ญ๋Ÿ‰์ด ์•„๋‹ˆ๋ผ ์ „๋ ฅ์—์˜ ์ ‘๊ทผ์„ฑ์ด ๊ฒฝ์Ÿ์˜ ํ•ต์‹ฌ ์š”์†Œ๋กœ ๋– ์˜ค๋ฅด๊ณ  ์žˆ๋‹ค. ๊ณ ๊ธฐ์•„๋Š” โ€œ์ด์ œ ๊ฒฝ์Ÿ์€ ๋ˆ„๊ฐ€ ๋” ๋งŽ์€ GPU ์ธ์Šคํ„ด์Šค๋ฅผ ๋จผ์ € ๋‚ด๋†“๋А๋ƒ์˜ ๋ฌธ์ œ๊ฐ€ ์•„๋‹ˆ๋‹ค. ๋ˆ„๊ฐ€ ๋จผ์ € ์ „๋ ฅ๋ง์— ์ ‘๊ทผํ•˜๋А๋ƒ์˜ ์‹ธ์›€์œผ๋กœ ๋ฐ”๋€Œ๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ๋งํ–ˆ๋‹ค.

๋Œ€์•ˆ์œผ๋กœ ๋– ์˜ค๋ฅธ โ€˜์ฝ”๋กœ์ผ€์ด์…˜โ€™

๊ธฐ์กด ์ „๋ ฅ๋ง ์—ฐ๊ฒฐ์„ ์ˆ˜๋…„์”ฉ ๊ธฐ๋‹ค๋ฆด ์ˆ˜ ์—†๊ฒŒ ๋˜๋ฉด์„œ, ์ฃผ์š” ํด๋ผ์šฐ๋“œ ์—…์ฒด๋Š” ๋ฐœ์ „์†Œ ๋ฐ”๋กœ ์˜†์— ๋ฐ์ดํ„ฐ์„ผํ„ฐ๋ฅผ ์ง“๋Š” ์ฝ”๋กœ์ผ€์ด์…˜ ๋ฐฉ์‹์„ ๋Œ€์•ˆ์œผ๋กœ ์„ ํƒํ•˜๊ณ  ์žˆ๋‹ค. ์ด๋Š” ์†ก์ „๋ง์„ ๊ฑฐ์น˜์ง€ ์•Š๊ณ  ๋ฐœ์ „์›๊ณผ ์ง์ ‘ ์—ฐ๊ฒฐํ•˜๋Š” ๋ฐฉ์‹์ด๋‹ค. ๊ณ ๊ธฐ์•„๋Š” ์ „๋ ฅ ๊ณต๊ธ‰์ด ์ œํ•œ๋œ ์ง€์—ญ์˜ ์ฝ”๋กœ์ผ€์ด์…˜ ์ˆ˜์š”๊ฐ€ ๊ธ‰์ฆํ•˜๋ฉด์„œ ๊ณ„์•ฝ ๊ฐ€๊ฒฉ์ด ์•ฝ 20% ์ƒ์Šนํ–ˆ๋‹ค๊ณ  ๋ฐํžˆ๋ฉฐ, ์ด ๋น„์šฉ์ด ์ง€์—ญ๋ณ„ ๊ฐ€๊ฒฉ ์ฐจ์ด๋ฅผ ํ†ตํ•ด ํด๋ผ์šฐ๋“œ ๊ณ ๊ฐ์—๊ฒŒ ์ „๊ฐ€๋˜๊ณ  ์žˆ๋‹ค๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

๊ตฌ๊ธ€๋„ ์ด ๊ฐ™์€ ๋ฐฉ์•ˆ์„ ๊ฒ€ํ† ํ•˜๊ณ ๋Š” ์žˆ์ง€๋งŒ, ํ•œ๋‚˜๋Š” ํšŒ์‚ฌ๊ฐ€ ์—ฌ์ „ํžˆ ์ „๋ ฅ๋ง์— ์—ฐ๊ฒฐ๋œ ๋ฐ์ดํ„ฐ์„ผํ„ฐ๋ฅผ ์ตœ์šฐ์„  ๋ฐฉ์‹์œผ๋กœ ์‚ผ๊ณ  ์žˆ๋‹ค๊ณ  ๊ฐ•์กฐํ–ˆ๋‹ค. ๊ทธ๋Š” โ€œ์ „๋ ฅ์„ ๋” ๋นจ๋ฆฌ ํ™•๋ณดํ•˜๊ธฐ ์œ„ํ•œ ์„ ํƒโ€์ด๋ผ๋ฉด์„œ๋„, ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ๋ฐœ์ „์†Œ ์ „๋ ฅ์„ ๋‹จ๋…์œผ๋กœ ์‚ฌ์šฉํ•˜๋Š” ๊ณ ๋ฆฝํ˜• ์‹œ์„ค์ด ์•„๋‹ˆ๋ผ ์ „๋ ฅ๋ง์— ์—ฐ๊ฒฐ๋ผ ๋‹ค๋ฅธ ์ˆ˜์š”์ฒ˜์™€ ์ „๋ ฅ์„ ํ•จ๊ป˜ ์‚ฌ์šฉํ•˜๋Š” ํ˜•ํƒœ๋ฅผ ์œ ์ง€ํ•˜๊ธธ ์›ํ•œ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

๊ทธ์— ๋”ฐ๋ฅด๋ฉด ๋‹ค๋ฅธ ํด๋ผ์šฐ๋“œ ์—…์ฒด ์—ญ์‹œ ์ „๋ ฅ ํšŒ์‚ฌ์™€ ์ง์ ‘ ํ˜‘์ƒ์— ๋‚˜์„œ๊ฑฐ๋‚˜ ๋ฐœ์ „์†Œ ์ธ๊ทผ ํ† ์ง€๋ฅผ ํ™•๋ณดํ•˜๊ณ , ๋ฐฐํ„ฐ๋ฆฌ๋ถ€ํ„ฐ ์†Œํ˜• ๋ชจ๋“ˆ ์›์ž๋กœ(SMR)์— ์ด๋ฅด๊ธฐ๊นŒ์ง€ ๋‹ค์–‘ํ•œ ์ „๋ ฅ ์ธํ”„๋ผ์— ๋Œ€ํ•œ ํˆฌ์ž ๊ฐ€๋Šฅ์„ฑ์„ ๊ฒ€ํ† ํ•˜๊ณ  ์žˆ๋‹ค.

๋‹ค๋งŒ ์ฝ”๋กœ์ผ€์ด์…˜์€ ์ƒˆ๋กœ์šด ์•ˆ์ •์„ฑ ๋ฌธ์ œ๋ฅผ ๋™๋ฐ˜ํ•œ๋‹ค. ๊ณ ๊ธฐ์•„๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ์ „๋ ฅ๋ง๊ณผ ๋ถ„๋ฆฌ๋œ ์ƒํƒœ์—์„œ ๋‹จ์ผ ๋ฐœ์ „์›์— ์˜์กดํ•  ๊ฒฝ์šฐ ๊ธฐ์กด์˜ ์ด์ค‘ํ™” ์ „๋žต์„ ์ ์šฉํ•˜๊ธฐ ์–ด๋ ค์›Œ์งˆ ์ˆ˜ ์žˆ๋‹ค๊ณ  ์ง€์ ํ–ˆ๋‹ค. ๊ทธ๋Š” โ€œ์ „ํ†ต์ ์ธ ํด๋ผ์šฐ๋“œ ๋ฆฌ์ „์€ ์ด์ค‘ ์ „๋ ฅ ๊ณต๊ธ‰๋ง๊ณผ ์ค‘๋ณต ๋ณ€์ „์†Œ, ๋„คํŠธ์›Œํฌ ์ „๋ฐ˜์—์„œ ์ „๋ ฅ์„ ๋Œ์–ด์˜ฌ ์ˆ˜ ์žˆ๋Š” ๊ตฌ์กฐ๋ฅผ ๊ฐ–์ถฐ ์™”๋‹ค. ๋ฐ˜๋ฉด ์ฝ”๋กœ์ผ€์ด์…˜ ๋ฐ์ดํ„ฐ์„ผํ„ฐ๋Š” ์ข…์ข… ๊ณ ๋ฆฝ๋œ ์„ฌ์ฒ˜๋Ÿผ ์šด์˜๋œ๋‹คโ€๋ผ๊ณ  ๋งํ–ˆ๋‹ค.

์ž„์‹œ ๋Œ€์‘์„ ๋„˜์–ด ๊ตฌ์กฐ์  ํ•ด๋ฒ• ํ•„์š”

์ฝ”๋กœ์ผ€์ด์…˜์ด ๋‹จ๊ธฐ์ ์ธ ๋Œ€์‘์ฑ…์ด ๋  ์ˆ˜๋Š” ์žˆ์ง€๋งŒ, ์†ก์ „ ์šฉ๋Ÿ‰ ๋ถ€์กฑ์ด๋ผ๋Š” ๊ทผ๋ณธ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๋ ค๋ฉด ๋ณด๋‹ค ํญ๋„“์€ ์ •์ฑ…์  ๊ฐœ์ž…์ด ํ•„์š”ํ•˜๋‹ค๋Š” ์ง€์ ๋„ ๋‚˜์˜จ๋‹ค. ์ด์™€ ๊ด€๋ จํ•ด ๋ฏธ๊ตญ ์—ฐ๋ฐฉ ๊ทœ์ œ ๋‹น๊ตญ์€ ์ง€๋‚œ 2023๋…„ ์ „๋ ฅ๋ง ์—ฐ๊ฒฐ ์ ˆ์ฐจ๋ฅผ ๊ฐœ์„ ํ•˜๊ธฐ ์œ„ํ•ด ๋‚˜์„  ๋ฐ” ์žˆ๋‹ค. ๊ธฐ์กด์— ์‹ ์ฒญ ์ˆœ์„œ๋Œ€๋กœ ์ฒ˜๋ฆฌํ•˜๋˜ ๋ฐฉ์‹ ๋Œ€์‹ , ์ค€๋น„๊ฐ€ ๊ฐ€์žฅ ์ž˜ ๋œ ์‚ฌ์—…์„ ์šฐ์„  ์—ฐ๊ฒฐํ•˜๋Š” โ€˜์ค€๋น„ ์šฐ์„ โ€™ ํด๋Ÿฌ์Šคํ„ฐ ๋ชจ๋ธ์„ ๋„์ž…ํ•˜๊ณ , ๋ถ€์ง€ ํ™•๋ณด ์š”๊ฑด์„ ๊ฐ•ํ™”ํ•˜๋Š” ํ•œํŽธ ๋ณด์ฆ๊ธˆ ๋ถ€๋‹ด๋„ ๋†’์˜€๋‹ค. ๋‹ค๋งŒ ์‹ค์ œ ์ดํ–‰์€ ์—ฌ์ „ํžˆ ๋”๋””๊ณ  ์ „๋ ฅ๋ง ์ž์ฒด์˜ ์šฉ๋Ÿ‰ ๋ถ€์กฑ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๊ธฐ์—๋Š” ์—ญ๋ถ€์กฑ์ธ ์ƒํ™ฉ์ด๋‹ค.

๋ฏธ๊ตญ ์—๋„ˆ์ง€๋ถ€๋Š” ์ฒจ๋‹จ ์†ก์ „ ๊ธฐ์ˆ ์„ ํ™œ์šฉํ•  ๊ฒฝ์šฐ ์ตœ๋Œ€ 100GW์˜ ์ถ”๊ฐ€ ์šฉ๋Ÿ‰์„ ํ™•๋ณดํ•  ์ˆ˜ ์žˆ์„ ๊ฒƒ์œผ๋กœ ๋ณด๊ณ  ์žˆ๋‹ค. ๋‹ค๋งŒ ์ด๋ฅผ ํ˜„์‹คํ™”ํ•˜๋ ค๋ฉด ์ „๋ ฅ ๊ธฐ์—…์ด ํ•ด๋‹น ๊ธฐ์ˆ ์„ ๋„์ž…ํ•˜๋„๋ก ์œ ๋„ํ•  ์ˆ˜ ์žˆ๋Š” ๊ทœ์ œ ํ™˜๊ฒฝ ๋ณ€ํ™”๊ฐ€ ํ•„์š”ํ•˜๋‹ค. ๊ณจ๋“œ๋งŒ์‚ญ์Šค ๋ฆฌ์„œ์น˜๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ ํ™•์žฅ์„ ์ง€์›ํ•˜๊ธฐ ์œ„ํ•ด 2030๋…„๊นŒ์ง€ ์•ฝ 7,200์–ต ๋‹ฌ๋Ÿฌ์— ๋‹ฌํ•˜๋Š” ์ „๋ ฅ๋ง ํˆฌ์ž๊ฐ€ ํ•„์š”ํ•  ๊ฒƒ์œผ๋กœ ์ถ”์‚ฐํ–ˆ๋‹ค.

๊ณ ๊ธฐ์•„๋Š” โ€œํด๋ผ์šฐ๋“œ ์—…์ฒด๋‚˜ ๋ฆฌ์ „์„ ์„ ํƒํ•  ๋•Œ ์ด์ œ๋Š” ๊ธฐ์ˆ ์  ์ ํ•ฉ์„ฑ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ, ํ•ด๋‹น ์‚ฌ์—…์ž๊ฐ€ ํ–ฅํ›„ 5๋…„ ๋™์•ˆ ์›Œํฌ๋กœ๋“œ๋ฅผ ๊ฐ๋‹นํ•  ์ˆ˜ ์žˆ๋Š” ์ „๋ ฅ ์‚ฌ์šฉ ์—ญ๋Ÿ‰์„ ํ™•๋ณดํ•˜๊ณ  ์žˆ๋Š”์ง€ ์—ฌ๋ถ€๋„ ํ™•์ธํ•ด์•ผ ํ•œ๋‹ค. ์ง€๊ธˆ ๊ฐ™์€ ์‹œ์žฅ์—์„œ๋Š” ํ›จ์”ฌ ๋” ์‹ ์ค‘ํ•˜๊ฒŒ ์„ ํƒํ•ด์•ผ ํ•œ๋‹คโ€๋ผ๊ณ  ์กฐ์–ธํ–ˆ๋‹ค.
dl-ciokorea@foundryco.com

โ€œ๋ฉ”๋ชจ๋ฆฌ ๋ถ€์กฑ์ด 2026๋…„ PC ์‹œ์žฅ ํ”๋“ ๋‹คยทยทยท๋‚ด๋…„๊นŒ์ง€ ๊ฐ€๊ฒฉ ์ธ์ƒโ€ IDC ์ „๋ง

IDC์˜ ์ตœ์‹  โ€˜์ „ ์„ธ๊ณ„ ๋ถ„๊ธฐ๋ณ„ ๊ฐœ์ธ์šฉ ์ปดํ“จํŒ… ๊ธฐ๊ธฐ ์ถ”์  ๋ณด๊ณ ์„œโ€™์— ๋”ฐ๋ฅด๋ฉด 2025๋…„ 4๋ถ„๊ธฐ ๊ธ€๋กœ๋ฒŒ PC ์ถœํ•˜๋Ÿ‰์€ ์ „๋…„ ๋™๊ธฐ ๋Œ€๋น„ 9.6% ์ฆ๊ฐ€ํ–ˆ๋‹ค. ๋‹ค๋งŒ ๋ฉ”๋ชจ๋ฆฌ ๋ถ€์กฑ ์‚ฌํƒœ๊ฐ€ ๊ธฐ์—…๊ณผ ์†Œ๋น„์ž ์‹œ์žฅ ๋ชจ๋‘์— ์˜ํ–ฅ์„ ๋ฏธ์น  ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒ๋˜๋ฉด์„œ, ์ถœํ•˜๋Ÿ‰ ์ฆ๊ฐ€์„ธ๊ฐ€ ์ง€์†๋  ๊ฐ€๋Šฅ์„ฑ์€ ํฌ์ง€ ์•Š์€ ๊ฒƒ์œผ๋กœ ๋ถ„์„๋๋‹ค.

IDC๋Š” ๋ฉ”๋ชจ๋ฆฌ ์ˆ˜๊ธ‰ ์ƒํ™ฉ์ด ๋น ๋ฅด๊ฒŒ ๋ณ€ํ™”ํ•˜๊ณ  ์žˆ๋‹ค๋Š” ์ ์„ ๊ณ ๋ คํ•  ๋•Œ 12๊ฐœ์›” ๋’ค PC ์‹œ์žฅ์ด ํฌ๊ฒŒ ๋‹ฌ๋ผ์งˆ ๊ฒƒ์œผ๋กœ ๋‚ด๋‹ค๋ดค๋‹ค. IDC๋Š” โ€œ์ด๋ฒˆ ์‹ค์ ์—๋Š” ์œˆ๋„์šฐ 10 ์ง€์› ์ข…๋ฃŒ๋กœ ์ธํ•œ ์—…๊ทธ๋ ˆ์ด๋“œ ์ˆ˜์š”, ์—ฐ์ดˆ ๊ด€์„ธ์— ๋Œ€ํ•œ ์šฐ๋ ค๋กœ ๋ฒค๋”๊ฐ€ ๋‹น์ดˆ ๊ณ„ํš๋ณด๋‹ค ๋งŽ์€ ์žฌ๊ณ ๋ฅผ ์•ž๋‹น๊ฒจ ํ™•๋ณดํ•œ ์ ์ด ์˜ํ–ฅ์„ ๋ฏธ์ณค๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

IDC ๋ฆฌ์„œ์น˜ ๋งค๋‹ˆ์ € ์ง€ํ…Œ์‹œ ์šฐ๋ธŒ๋ผ๋‹ˆ๋Š” โ€œ๋ฉ”๋ชจ๋ฆฌ ๋ถ€์กฑ์ด ์‚ฐ์—… ์ „๋ฐ˜์— ์˜ํ–ฅ์„ ๋ฏธ์น˜๊ณ  ์žˆ์œผ๋ฉฐ, ๊ทธ ์—ฌํŒŒ๊ฐ€ ํ–ฅํ›„ 2๋…„๊ฐ„ ์‹œ์žฅ์˜ ์—ญํ•™ ๊ตฌ์กฐ๋ฅผ ์žฌํŽธํ•  ๊ฐ€๋Šฅ์„ฑ์ด ํฌ๋‹คโ€๋ผ๊ณ  ์ „๋งํ–ˆ๋‹ค.

IDC๋Š” ์ง€๋‚œ๋‹ฌ ๊ณต๊ฐœํ•œ ๋ณด๊ณ ์„œ์—์„œ โ€œ๊ธ€๋กœ๋ฒŒ ๋ฐ˜๋„์ฒด ์ƒํƒœ๊ณ„๊ฐ€ ์ „๋ก€ ์—†๋Š” ๋ฉ”๋ชจ๋ฆฌ ์นฉ ๋ถ€์กฑ ์ƒํ™ฉ์— ์ง๋ฉดํ•ด ์žˆ์œผ๋ฉฐ, ์ด๋กœ ์ธํ•œ ํŒŒ๊ธ‰ ํšจ๊ณผ๊ฐ€ ๊ธฐ๊ธฐ ์ œ์กฐ์‚ฌ์™€ ์ตœ์ข… ์‚ฌ์šฉ์ž์—๊ฒŒ๊นŒ์ง€ ํ™•์‚ฐ๋ผ 2027๋…„๊นŒ์ง€ ์ด์–ด์งˆ ์ˆ˜ ์žˆ๋‹ค. AI ๋ฐ์ดํ„ฐ์„ผํ„ฐ์—์„œ ๋ฐœ์ƒํ•˜๋Š” ์ˆ˜์š”๊ฐ€ ๊ณต๊ธ‰์„ ๊ณ„์†ํ•ด์„œ ์›ƒ๋Œ๋ฉด์„œ ๋””๋žจ(DRAM) ๊ฐ€๊ฒฉ์ด ํฌ๊ฒŒ ์ƒ์Šนํ–ˆ๊ณ , ์ด๋กœ ์ธํ•ด ์ˆ˜๊ธ‰ ๋ถˆ๊ท ํ˜•์ด ์‹ฌํ™”๋˜๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ์ง„๋‹จํ–ˆ๋‹ค.

์˜ฌํ•ด PC ๊ฐ€๊ฒฉ์€ ๋†’์€ ์ˆ˜์ค€ ์œ ์ง€

IDC ์—ฐ๊ตฌ์ง„์€ ๋ฉ”๋ชจ๋ฆฌ ์‹œ์žฅ์ด โ€œ์ˆ˜์š”๊ฐ€ ๊ณต๊ธ‰์„ ํฌ๊ฒŒ ์›ƒ๋„๋Š” ์ „๋ก€ ์—†๋Š” ๋ณ€๊ณก์ ์— ๋†“์—ฌ ์žˆ๋‹คโ€๋ผ๋ฉฐ, โ€œ์žฅ๊ธฐ๊ฐ„ ํ˜ธํ™ฉ๊ณผ ๋ถˆํ™ฉ์ด ๋ฐ˜๋ณต๋ผ ์˜จ ์‚ฐ์—…์ด์ง€๋งŒ, ์ด๋ฒˆ์—๋Š” ์–‘์ƒ์ด ๋‹ค๋ฅด๋‹ค. AI ์ธํ”„๋ผ์™€ ์›Œํฌ๋กœ๋“œ๊ฐ€ ๋น ๋ฅด๊ฒŒ ํ™•๋Œ€๋˜๋ฉด์„œ ๋ฉ”๋ชจ๋ฆฌ ์ƒํƒœ๊ณ„ ์ „๋ฐ˜์— ์ƒ๋‹นํ•œ ์••๋ฐ•์„ ๊ฐ€ํ•˜๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ๋ถ„์„ํ–ˆ๋‹ค.

๋žจ ๊ฐ€๊ฒฉ ๊ธ‰๋“ฑ์€ ๊ธฐ์—…์šฉ PC ์‹œ์žฅ์—๋„ ๋‹น๋ถ„๊ฐ„ ์˜ํ–ฅ์„ ๋ฏธ์น  ๊ฐ€๋Šฅ์„ฑ์ด ํฌ๋‹ค. ์ธํฌํ…Œํฌ ๋ฆฌ์„œ์น˜ ๊ทธ๋ฃน์˜ ์ˆ˜์„ ์ž๋ฌธ ์ด์‚ฌ ํ…Œ์˜ค ์•ˆํ† ๋‹ˆ์•„๋””์Šค๋Š” 2026๋…„์€ ๋ฌผ๋ก  2027๋…„๊นŒ์ง€๋„ ๊ฐ€๊ฒฉ์ด ๋†’์€ ์ˆ˜์ค€์œผ๋กœ ์œ ์ง€๋  ๊ฒƒ์œผ๋กœ ๋‚ด๋‹ค๋ดค๋‹ค. ๊ทธ๋Š” โ€œ์ด๋กœ ์ธํ•ด PC ๊ฐ€๊ฒฉ์€ ์ตœ๋Œ€ 20%๊นŒ์ง€ ์˜ค๋ฅผ ์ˆ˜ ์žˆ์œผ๋ฉฐ, PC์™€ ์Šค๋งˆํŠธํฐ ์‹œ์žฅ์—์„œ ๋ช…ํ™•ํ•œ ์Šน์ž์™€ ํŒจ์ž๊ฐ€ ๊ฐˆ๋ฆด ๊ฒƒโ€์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

์•ˆํ† ๋‹ˆ์•„๋””์Šค๋Š” โ€œ์žฅ๊ธฐ ๊ณต๊ธ‰ ๊ณ„์•ฝ๊ณผ ๊ฐ™์€ ํ—ค์ง• ์ „๋žต์„ ๊ฐ–์ถ”๊ณ , ์ถœํ•˜ ๋ฌผ๋Ÿ‰์ด ๋งŽ์œผ๋ฉฐ ์ถฉ๋ถ„ํ•œ ํ˜„๊ธˆ ๋ณด์œ ๊ณ ๋ฅผ ํ™•๋ณดํ•œ ๋ฒค๋”๊ฐ€ ์œ ๋ฆฌํ•œ ์œ„์น˜์— ์„ค ๊ฒƒโ€์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ๋ฐ˜๋ฉด โ€œ๊ฐ€๊ฒฉ ๊ฒฝ์Ÿ๋ ฅ์„ ํ•ต์‹ฌ์œผ๋กœ ์‚ผ์•„ ๋‚ฎ์€ ๋งˆ์ง„์— ์˜์กดํ•˜๊ฑฐ๋‚˜ ๊ทœ๋ชจ๊ฐ€ ์ž‘์€ ์—…์ฒด๋Š” ์ด ๊ธฐ๊ฐ„ ๋™์•ˆ ์–ด๋ ค์›€์„ ๊ฒช์„ ๊ฐ€๋Šฅ์„ฑ์ด ํฌ๋‹คโ€๋ผ๊ณ  ์ „๋งํ–ˆ๋‹ค.

๋Œ€๊ธฐ์—…์ด ์—ฌ์ „ํžˆ ๋…ธํŠธ๋ถ์€ 3๋…„, ๋ฐ์Šคํฌํ†ฑ์€ 5๋…„ ์ฃผ๊ธฐ๋กœ ๊ต์ฒดํ•˜๊ณ  ์žˆ๋Š”์ง€, ์•„๋‹ˆ๋ฉด ์‚ฌ์šฉ ๊ธฐ๊ฐ„์„ ๋” ๋Š˜๋ฆฌ๊ณ  ์žˆ๋Š”์ง€๋ฅผ ๋ฌป๋Š” ์งˆ๋ฌธ์— ์•ˆํ† ๋‹ˆ์•„๋””์Šค๋Š” โ€œ์•„์ง ํ˜„์‹ค์„ ๋ถ€์ •ํ•˜๊ณ  ์žˆ๋Š” ์ƒํƒœโ€๋ผ๊ณ  ์ง„๋‹จํ–ˆ๋‹ค. ๊ทธ๋Š” โ€œ2026๋…„ ์˜ˆ์‚ฐ์ด ์ง€๋‚œ ๋ถ„๊ธฐ์ธ 2025๋…„ 4๋ถ„๊ธฐ์— ํ™•์ •๋๊ธฐ ๋•Œ๋ฌธ์—, ์•„์ง ์ด ๋ฌธ์ œ๋ฅผ ๋…ผ์˜ํ•˜๋Š” ๊ณ ๊ฐ์„ ๊ฑฐ์˜ ๋ณด์ง€ ๋ชปํ–ˆ๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

๋‹ค๋งŒ ๊ทธ๋Š” ์‹ค์ œ ๊ต์ฒด ๊ฒฌ์ ์„ ๋ฐ›๋Š” ๊ณผ์ •์—์„œ ๊ฐ€๊ฒฉ ๋ณ€ํ™”๋ฅผ ์ฒด๊ฐํ•˜๊ฒŒ ๋˜๋ฉด ์ผ๋ถ€ ์ „๋žต ์ˆ˜์ •๊ณผ ๊ตฌ๋งค ์ถ•์†Œ๊ฐ€ ๋‚˜ํƒ€๋‚  ๊ฐ€๋Šฅ์„ฑ์ด ์žˆ๋‹ค๊ณ  ๋‚ด๋‹ค๋ดค๋‹ค. ๊ทธ๋Š” โ€œ๊ฐ€๊ฒฉ์ด ๋‚ด๋ ค๊ฐ€์ง€ ์•Š์„ ๊ฒฝ์šฐ 2027๋…„์—๋Š” ์ €์„ฑ๋Šฅ PC๋ฅผ ํ™œ์šฉํ•œ VDI(Virtual Desktop Infrastructure)๊ฐ€ ๋‹ค์‹œ ๋ถ€์ƒํ•˜๊ณ , ๋…ธํ›„ PC๋กœ ์ธํ•œ ์ƒ์‚ฐ์„ฑ ์ €ํ•˜๋ฅผ ์ƒ์‡„ํ•˜๊ธฐ ์œ„ํ•ด ํด๋ผ์šฐ๋“œ ๊ธฐ๋ฐ˜ ์†”๋ฃจ์…˜์„ ๋‹ค์‹œ ํ•œ๋ฒˆ ํ™•๋Œ€ํ•˜๋ ค๋Š” ์›€์ง์ž„์ด ๋‚˜ํƒ€๋‚  ์ˆ˜ ์žˆ๋‹คโ€๋ผ๊ณ  ์ „๋งํ–ˆ๋‹ค.

IDC ์›”๋“œ์™€์ด๋“œ ๋ชจ๋ฐ”์ผ ๋””๋ฐ”์ด์Šค ํŠธ๋ž˜์ปค์˜ ๋ฆฌ์„œ์น˜ ๋ถ€์‚ฌ์žฅ ์žฅ ํ•„๋ฆฌํ”„ ๋ถ€์ƒค๋ฅด๋Š” โ€œ์ผ๋ถ€ ์ œ์กฐ์‚ฌ๊ฐ€ ์ด๋ฏธ ๋ฐœํ‘œํ•œ ์‹œ์Šคํ…œ ๊ฐ€๊ฒฉ ์ธ์ƒ ์™ธ์—๋„, ๋ฉ”๋ชจ๋ฆฌ ์žฌ๊ณ ๋ฅผ ์œ ์ง€ํ•˜๊ธฐ ์œ„ํ•ด ํ‰๊ท ์ ์ธ PC ๋ฉ”๋ชจ๋ฆฌ ์‚ฌ์–‘์ด ๋‚ฎ์•„์งˆ ์ˆ˜๋„ ์žˆ๋‹ค. ์•ž์œผ๋กœ 1๋…„์ด ๋งค์šฐ ๋ณ€๋™์„ฑ์ด ํฐ ์‹œ๊ธฐ๊ฐ€ ๋  ๊ฒƒโ€์ด๋ผ๊ณ  ๋‚ด๋‹ค๋ดค๋‹ค.

IDC๋Š” ์ง€๋‚œ๋‹ฌ ๋ณด๊ณ ์„œ์—์„œ ๋ฉ”๋ชจ๋ฆฌ ๋ถ€์กฑ ์‚ฌํƒœ๊ฐ€ AI PC๋ฅผ ์ค‘์‹ฌ์œผ๋กœ ํ•œ ์—…๊ณ„์˜ ์„ฑ์žฅ ๊ตฌ์กฐ๋ฅผ ํ”๋“ค ์ˆ˜ ์žˆ๋‹ค๊ณ  ์ง€์ ํ–ˆ๋‹ค. IDC๋Š” NPU๋ฅผ ํƒ‘์žฌํ•œ ๋ชจ๋“  PC๋ฅผ AI PC๋กœ ์ •์˜ํ•˜๊ณ  ์žˆ๋Š”๋ฐ, ์ด๋Ÿฌํ•œ ๊ธฐ๊ธฐ๋Š” ๋ณธ์งˆ์ ์œผ๋กœ ๋” ๋งŽ์€ ๋žจ์„ ์š”๊ตฌํ•˜๋Š” ๊ฒฝํ–ฅ์ด ์žˆ๋‹ค. ์‹ค์ œ๋กœ ๋งˆ์ดํฌ๋กœ์†Œํ”„ํŠธ์˜ ์ฝ”ํŒŒ์ผ๋Ÿฟ+ PC๋Š” ์ตœ์†Œ 16GB ๋žจ์„ ํ•„์š”๋กœ ํ•œ๋‹ค.

์—ฐ๊ตฌ์ง„์€ ์†Œ๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ๊ณผ ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ์ด ์ ์ฐจ ๊ธฐ๊ธฐ ๋‚ด๋ถ€์— ํƒ‘์žฌ๋˜๋ฉด์„œ ๋ฉ”๋ชจ๋ฆฌ์˜ ์ค‘์š”์„ฑ์ด ๋”์šฑ ์ปค์ง€๊ณ  ์žˆ๋‹ค๊ณ  ๋ถ„์„ํ–ˆ๋‹ค. ์—ฐ๊ตฌ์ง„์€ โ€œ์ด์— ๋”ฐ๋ผ ๊ณ ๊ธ‰ํ˜• ์‹œ์Šคํ…œ ์ƒ๋‹น์ˆ˜๊ฐ€ 32GB ์ด์ƒ์œผ๋กœ ์ „ํ™˜๋˜๋Š” ํ๋ฆ„์ด ๋‚˜ํƒ€๋‚˜๊ณ  ์žˆ์ง€๋งŒ, ์—…๊ณ„๊ฐ€ ๋žจ ์ฆ์„ค์˜ ํ•„์š”์„ฑ์„ ์ธ์‹ํ•˜๋Š” ์‹œ์ ์— ์˜คํžˆ๋ ค ๋น„์šฉ ๋ถ€๋‹ด์ด ์ง€๋‚˜์น˜๊ฒŒ ์ปค์กŒ๋‹ค๋Š” ์ ์ด ๋ฌธ์ œ๋‹ค. ์„ค๋ น ๊ณต๊ธ‰์„ ํ™•๋ณดํ•˜๋”๋ผ๋„ ์ถ”๊ฐ€ ๋น„์šฉ์ด ๊ณผ๋„ํ•ด์ง„ ์ƒํ™ฉ์ด๋‹ค. ๊ทธ ๊ฒฐ๊ณผ ๊ฐ€๊ฒฉ ์ธ์ƒ์ด๋‚˜ ๋งˆ์ง„ ์ถ•์†Œ, ํ˜น์€ ์‹ ๊ทœ ์‹œ์Šคํ…œ์—์„œ ๋žจ ์šฉ๋Ÿ‰์„ ๋‚ฎ์ถ”๋Š” ์„ ํƒ์ด ๋ถˆ๊ฐ€ํ”ผํ•ด์งˆ ์ˆ˜ ์žˆ๋‹คโ€๋ผ๊ณ  ๋‚ด๋‹ค๋ดค๋‹ค.

์šฐ๋ธŒ๋ผ๋‹ˆ๋Š” ์ง€๋‚œ 1~2๋…„๊ฐ„ ๋ฉ”๋ชจ๋ฆฌ ๋ถ€์กฑ ๋ฌธ์ œ๊ฐ€ ์—†์—ˆ์Œ์—๋„ ๊ธฐ์—…์šฉ AI PC ์ˆ˜์š”๊ฐ€ ๋‹ค์†Œ ์ฃผ์ถคํ•œ ํ๋ฆ„์„ ๋ณด์—ฌ์™”๋‹ค๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ๊ทธ๋Š” โ€œํด๋ผ์šฐ๋“œ ๊ธฐ๋ฐ˜ AI ์˜ต์…˜์˜ ์™„์„ฑ๋„๊ฐ€ ๋งค์šฐ ๋†’์•„์ง€๋ฉด์„œ, ๋งŽ์€ ์ด์šฉ์ž๊ฐ€ ํด๋ผ์šฐ๋“œ AI์™€ ์˜จ๋””๋ฐ”์ด์Šค AI์˜ ์ฐจ์ด๋ฅผ ์ฒด๊ฐํ•˜์ง€ ๋ชปํ•˜๊ณ  ์žˆ๋Š” ์ ์ด ์ฃผ์š” ๋ฐฐ๊ฒฝโ€์ด๋ผ๊ณ  ๋งํ–ˆ๋‹ค.

๊ทธ๋Š” โ€œ์ด๋ฒˆ ๋ฉ”๋ชจ๋ฆฌ ๊ณต๊ธ‰๋‚œ์˜ ์˜ํ–ฅ์œผ๋กœ ๋ฒค๋”๊ฐ€ ํ•˜์ด๋ธŒ๋ฆฌ๋“œ AI๋ฅผ ๋ณธ๊ฒฉ์ ์œผ๋กœ ๋‚ด์„ธ์šฐ๊ธฐ ์‹œ์ž‘ํ•  ์ˆ˜๋„ ์žˆ๋‹ค. ๋ฉ”๋ชจ๋ฆฌ ๋ถ€์กฑ์œผ๋กœ ์ธํ•ด ์ผ๋ถ€ ์‹œ์Šคํ…œ์€ ์ตœ์‹  AI ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์„ ์›ํ™œํ•˜๊ฒŒ ๊ตฌ๋™ํ•˜์ง€ ๋ชปํ•  ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธโ€์ด๋ผ๊ณ  ๋ง๋ถ™์˜€๋‹ค.

๋Œ€๊ธฐ์—…์ด AI PC๋‚˜ ์ฝ”ํŒŒ์ผ๋Ÿฟ PC์— ํ”„๋ฆฌ๋ฏธ์—„์„ ์ง€๋ถˆํ• ์ง€, ์•„๋‹ˆ๋ฉด ํ•ด๋‹น ๋ฐฉํ–ฅ์„ ์„ ํƒํ•  ๊ฒฝ์šฐ ๊ณต๊ธ‰์‚ฌ์— ๋” ์ €๋ ดํ•œ SKU ์ œ๊ณต์„ ์š”๊ตฌํ• ์ง€์— ๋Œ€ํ•œ ์ „๋ง๋„ ์ œ์‹œ๋๋‹ค. ์•ˆํ† ๋‹ˆ์•„๋””์Šค๋Š” โ€œAI PC ๋ถ„์•ผ์—์„œ ์‹œ์žฅ ํ๋ฆ„์— ๋งž์„ค ๋งŒํ•œ ์˜ํ–ฅ๋ ฅ์„ ๊ฐ€์ง„ ๊ธฐ์—…์€ ์—†๋‹ค๊ณ  ๋ณธ๋‹ค. ๊ธฐ์กด์— ์žฅ๊ธฐ ๋ฒค๋” ๊ณ„์•ฝ์„ ์ฒด๊ฒฐํ•ด ๋‘” ๊ธฐ์—…์ด ์ƒ๋Œ€์ ์œผ๋กœ ์œ ๋ฆฌํ•œ ์œ„์น˜์— ์„ค ๊ฒƒโ€์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.
dl-ciokorea@foundryco.com

โ€œ์ „๊ธฐ๋ฃŒ ์ธ์ƒ, ์ฃผ๋ฏผ์— ๋– ๋„˜๊ธฐ์ง€ ์•Š๊ฒ ๋‹คโ€ MS, AI ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ง€์—ญ ์ƒ์ƒ ์›์น™ ์ œ์‹œ


MS์˜ ๋ถ€ํšŒ์žฅ ๋ธŒ๋ž˜๋“œ ์Šค๋ฏธ์Šค ์ด๋ฆ„์œผ๋กœ ๊ณต๊ฐœ๋œ ๊ธ€์— ๋”ฐ๋ฅด๋ฉด, MS๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ๊ฐ€์ ธ์˜ค๋Š” ํ˜œํƒ์ด ์ง€์—ญ์‚ฌํšŒ๊ฐ€ ๊ฐ๋‹นํ•ด์•ผ ํ•  ๋ถ€๋‹ด์„ ์ƒํšŒํ•ด์•ผ๋งŒ ์žฅ๊ธฐ์ ์ธ ์„ฑ๊ณต์ด ๊ฐ€๋Šฅํ•˜๋‹ค๊ณ  ํŒ๋‹จํ–ˆ๋‹ค๋ฉฐ, ์ง€์—ญ๊ณผ ์ƒ์ƒํ•˜๊ธฐ ์œ„ํ•œ ๋‹ค์„ฏ ๊ฐ€์ง€ ์›์น™์„ ์ œ์‹œํ–ˆ๋‹ค.

MS๊ฐ€ ์ œ์‹œํ•œ 5๋Œ€ ์›์น™ ๊ฐ€์šด๋ฐ ์ฒซ ๋ฒˆ์งธ๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ „๋ ฅ ์‚ฌ์šฉ์œผ๋กœ ์ธํ•œ ์ฃผ๋ฏผ ์ „๊ธฐ์š”๊ธˆ ์ธ์ƒ์„ ๋ง‰๊ฒ ๋‹ค๋Š” ์•ฝ์†์ด๋‹ค. MS๋Š” ๊ตญ์ œ์—๋„ˆ์ง€๊ธฐ๊ตฌ(IEA) ์ž๋ฃŒ๋ฅผ ์ธ์šฉํ•ด ๋ฏธ๊ตญ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ „๋ ฅ ์ˆ˜์š”๊ฐ€ 2035๋…„๊นŒ์ง€ ํ˜„์žฌ 200ํ…Œ๋ผ์™€ํŠธ์‹œ(TWh)์—์„œ 640ํ…Œ๋ผ์™€ํŠธ์‹œ๋กœ ์„ธ ๋ฐฐ ์ด์ƒ ์ฆ๊ฐ€ํ•  ๊ฒƒ์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ์ด์— ๋”ฐ๋ผ MS๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ์— ํ•„์š”ํ•œ ๋ฐœ์ „ยท์†ก์ „ยท๋ณ€์ „ ์ธํ”„๋ผ ๋น„์šฉ์„ ๊ธฐ์—…์ด ์ง์ ‘ ๋ถ€๋‹ดํ•˜๊ณ , ์ „๋ ฅํšŒ์‚ฌ ๋ฐ ์ฃผ(ๅทž) ๊ทœ์ œ๊ธฐ๊ด€๊ณผ ํ˜‘๋ ฅํ•ด ๋Œ€๊ทœ๋ชจ ์ „๋ ฅ ์‚ฌ์šฉ์ž์— ์ ํ•ฉํ•œ ์ „๊ธฐ์š”๊ธˆ ์ฒด๊ณ„๋ฅผ ์ ์šฉํ•˜๊ฒ ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค. ํ•ต์‹ฌ ๋ชฉํ‘œ๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์ „๋ ฅ ์ˆ˜์š” ์ฆ๊ฐ€๊ฐ€ ์ฃผ๊ฑฐ์šฉ ์ „๊ธฐ์š”๊ธˆ ์ธ์ƒ์œผ๋กœ ์ด์–ด์ง€์ง€ ์•Š๋„๋ก ํ•˜๋Š” ๊ฒƒ์ด๋‹ค.

๋‘ ๋ฒˆ์งธ ์•ฝ์†์€ ๋ฌผ ์‚ฌ์šฉ ๊ด€๋ฆฌ๋‹ค. ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๋ƒ‰๊ฐ ๊ณผ์ •์—์„œ ์‚ฌ์šฉ๋˜๋Š” ๋ฌผ์— ๋Œ€ํ•œ ์ง€์—ญ์‚ฌํšŒ์˜ ์šฐ๋ ค๋ฅผ ๊ณ ๋ คํ•ด ๋ฌผ ์‚ฌ์šฉ๋Ÿ‰์„ ์ตœ์†Œํ™”ํ•˜๊ณ , ์‚ฌ์šฉํ•œ ์–‘๋ณด๋‹ค ๋” ๋งŽ์€ ๋ฌผ์„ ์ง€์—ญ์— ํ™˜์›ํ•˜๊ฒ ๋‹ค๋Š” ๋ชฉํ‘œ๋ฅผ ์ œ์‹œํ–ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ํ์‡„ํ˜• ๋ƒ‰๊ฐ ์‹œ์Šคํ…œ ๋„์ž…, ์žฌํ™œ์šฉ์ˆ˜ ํ™œ์šฉ, ๋ˆ„์ˆ˜ ํƒ์ง€, ์Šต์ง€ ๋ณต์› ํ”„๋กœ์ ํŠธ ๋“ฑ์ด ์ฃผ์š” ์ˆ˜๋‹จ์œผ๋กœ ์–ธ๊ธ‰๋๋‹ค.

์„ธ ๋ฒˆ์งธ๋Š” ์ง€์—ญ ์ฃผ๋ฏผ์„ ์œ„ํ•œ ์ผ์ž๋ฆฌ ์ฐฝ์ถœ์ด๋‹ค. ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๊ฑด์„ค๊ณผ ์šด์˜ ๊ณผ์ •์—์„œ ๋ฐœ์ƒํ•˜๋Š” ์ผ์ž๋ฆฌ๊ฐ€ ์™ธ๋ถ€ ์ธ๋ ฅ์— ์ง‘์ค‘๋˜์ง€ ์•Š๋„๋ก ์ง€์—ญ ์ฃผ๋ฏผ์„ ๋Œ€์ƒ์œผ๋กœ ํ•œ ์ง์—… ํ›ˆ๋ จ๊ณผ ๊ฒฌ์Šต ํ”„๋กœ๊ทธ๋žจ์„ ํ™•๋Œ€ํ•˜๊ฒ ๋‹ค๋Š” ๊ณ„ํš์ด๋‹ค. ์ปค๋ฎค๋‹ˆํ‹ฐ ์นผ๋ฆฌ์ง€์™€ ์ง์—… ๊ต์œก๊ธฐ๊ด€๊ณผ์˜ ํ˜‘๋ ฅ์„ ํ†ตํ•ด ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์šด์˜์— ํ•„์š”ํ•œ ๊ธฐ์ˆ  ์ธ๋ ฅ์„ ์–‘์„ฑํ•˜๋Š” ๊ฒƒ์ด ํ•ต์‹ฌ์ด๋‹ค.

๋„ค ๋ฒˆ์งธ ์•ฝ์†์€ ์„ธ์ˆ˜ ํ™•๋Œ€๋‹ค. ๋งˆ์ดํฌ๋กœ์†Œํ”„ํŠธ๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๋ถ€์ง€์™€ ์‹œ์„ค์— ๋Œ€ํ•ด ์žฌ์‚ฐ์„ธ ๊ฐ๋ฉด์„ ์š”๊ตฌํ•˜์ง€ ์•Š๊ณ , ์ง€์—ญ ๋ณ‘์›ยทํ•™๊ตยท๋„์„œ๊ด€ยท๊ณต์› ๋“ฑ ๊ณต๊ณต์„œ๋น„์Šค ์žฌ์›์œผ๋กœ ํ™œ์šฉ๋  ์ˆ˜ ์žˆ๋„๋ก ์ •๋‹นํ•œ ์„ธ๊ธˆ์„ ๋‚ฉ๋ถ€ํ•˜๊ฒ ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค. ์ด๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ์ง€์—ญ ์žฌ์ •์— ์‹ค์งˆ์ ์œผ๋กœ ๊ธฐ์—ฌํ•˜๋Š” ๊ตฌ์กฐ๋ฅผ ๋งŒ๋“ค๊ฒ ๋‹ค๋Š” ์ทจ์ง€๋‹ค.

๋งˆ์ง€๋ง‰์€ ์ง€์—ญ AI ์—ญ๋Ÿ‰ ๊ฐ•ํ™”๋‹ค. ๋ฐ์ดํ„ฐ์„ผํ„ฐ๊ฐ€ ์œ„์น˜ํ•œ ์ง€์—ญ ์ฃผ๋ฏผ๋“ค์ด AI ๊ธฐ์ˆ  ํ™•์‚ฐ์˜ ์ˆ˜ํ˜œ์ž๊ฐ€ ๋  ์ˆ˜ ์žˆ๋„๋ก ํ•™๊ต, ๋„์„œ๊ด€, ๋น„์˜๋ฆฌ๋‹จ์ฒด์™€ ํ˜‘๋ ฅํ•ด AI ๊ต์œก๊ณผ ํ›ˆ๋ จ ๊ธฐํšŒ๋ฅผ ์ œ๊ณตํ•˜๊ฒ ๋‹ค๋Š” ๊ณ„ํš์ด๋‹ค. ํ•™์ƒ๊ณผ ์„ฑ์ธ, ์†Œ์ƒ๊ณต์ธ์„ ๋Œ€์ƒ์œผ๋กœ ํ•œ AI ๋ฆฌํ„ฐ๋Ÿฌ์‹œ ๊ต์œก์ด ์ฃผ์š” ๋‚ด์šฉ์ด๋‹ค.

๋งˆ์ดํฌ๋กœ์†Œํ”„ํŠธ๋Š” โ€œAI๊ฐ€ ๊ฐ€์ ธ์˜ฌ ๊ธ์ •์ ์ธ ๋ณ€ํ™”๋ฅผ ๋ฏฟ์ง€๋งŒ, MS ๊ฐ™์€ ๊ธฐ์ˆ  ๊ธฐ์—…์€ ์ง€์—ญ์‚ฌํšŒ๊ฐ€ ์ง๋ฉดํ•œ ๊ณผ์ œ๋ฅผ ์ •๋ฉด์œผ๋กœ ํ•ด๊ฒฐํ•ด์•ผ ํ•  ์ฑ…์ž„๋„ ์žˆ๋‹คโ€๋ผ๋ฉฐ 2026๋…„ ์ƒ๋ฐ˜๊ธฐ ์ค‘ ์ด ๊ฐ™์€ ์•ฝ์†๋“ค์„ ์‹คํ–‰์— ์˜ฎ๊ธฐ๊ฒ ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.
jihyun.lee@foundryco.com

๋ฉ”ํƒ€, AI ์ธํ”„๋ผ ์ด๊ด„ ์กฐ์ง โ€˜๋ฉ”ํƒ€ ์ปดํ“จํŠธโ€™ ์ถœ๋ฒ”ยทยทยทโ€์ดˆ๋Œ€ํ˜• AI ํด๋Ÿฌ์Šคํ„ฐ ๊ตฌ์ถ•ํ•  ๊ฒƒโ€

๋ฉ”ํƒ€๊ฐ€ AI ์ธํ”„๋ผ๋ฅผ ์ตœ์ƒ์œ„ ์ „๋žต ๊ณผ์ œ๋กœ ๊ฒฉ์ƒํ•˜๊ณ , ์‹ ๊ทœ ์กฐ์ง์ธ โ€˜๋ฉ”ํƒ€ ์ปดํ“จํŠธ(Meta Compute)โ€™๋ฅผ ์ถœ๋ฒ”ํ–ˆ๋‹ค. ๋ฉ”ํƒ€ ์ปดํ“จํŠธ๋Š” ๋ฐ์ดํ„ฐ์„ผํ„ฐ์™€ ๋„คํŠธ์›Œํฌ์˜ ๊ตฌ์ถ•ยท์šด์˜ ์ฑ…์ž„์„ ๋‹จ์ผ ๋ฆฌ๋”์‹ญ ์ฒด๊ณ„๋กœ ํ†ตํ•ฉํ•œ๋‹ค.

๋ฉ”ํƒ€์˜ CEO ๋งˆํฌ ์ €์ปค๋ฒ„๊ทธ๋Š” ์†Œ์…œ๋ฏธ๋””์–ด๋ฅผ ํ†ตํ•ด โ€œ๋ฉ”ํƒ€๋Š” ํ–ฅํ›„ 10๋…„ ๋™์•ˆ ์ˆ˜์‹ญ ๊ธฐ๊ฐ€์™€ํŠธ(GW) ๊ทœ๋ชจ์˜ ์ „๋ ฅ์„ ์‚ฌ์šฉํ•˜๋Š” AI ์ธํ”„๋ผ๋ฅผ ๊ตฌ์ถ•ํ•  ๊ณ„ํš์ด๋ฉฐ, ์žฅ๊ธฐ์ ์œผ๋กœ๋Š” ๊ทธ ๊ทœ๋ชจ๋ฅผ ์ˆ˜๋ฐฑ ๊ธฐ๊ฐ€์™€ํŠธ ์ด์ƒ์œผ๋กœ ํ™•๋Œ€ํ•  ๊ฒƒโ€์ด๋ผ๋ฉฐ โ€œ์ด ์ธํ”„๋ผ๋ฅผ ์–ด๋–ป๊ฒŒ ์„ค๊ณ„ํ•˜๊ณ  ํˆฌ์žํ•˜๋ฉฐ ํŒŒํŠธ๋„ˆ์‹ญ์„ ๋งบ๋А๋ƒ๊ฐ€ ์ค‘์š”ํ•œ ์ „๋žต์  ๊ฒฝ์Ÿ๋ ฅ์ด ๋  ๊ฒƒโ€์ด๋ผ๊ณ  ๋ฐํ˜”๋‹ค.

ํ•ด๋‹น ์กฐ์ง์€ ์‚ฐํ† ์‹œ ์ž๋‚˜๋ฅด๋‹จ๊ณผ ๋‹ค๋‹ˆ์—˜ ๊ทธ๋กœ์Šค๊ฐ€ ๊ณต๋™์œผ๋กœ ์ด๋ˆ๋‹ค. ์ž๋‚˜๋ฅด๋‹จ์€ ๊ธฐ์กด๊ณผ ๊ฐ™์ด ๋ฉ”ํƒ€์˜ ๋ฐ์ดํ„ฐ์„ผํ„ฐ์™€ ๋„คํŠธ์›Œํฌ ๊ธฐ๋ฐ˜ ์ „๋ฐ˜์„ ์ด๊ด„ํ•˜๊ณ , ๊ทธ๋กœ์Šค๋Š” AI ์ธํ”„๋ผ์˜ ์žฅ๊ธฐ ์šฉ๋Ÿ‰ ๊ณ„ํš๊ณผ ๋ฒค๋” ์ „๋žต, ๋น„์ฆˆ๋‹ˆ์Šค ๋ชจ๋ธ๋ง์„ ๋‹ด๋‹นํ•œ๋‹ค.

์ €์ปค๋ฒ„๊ทธ๋Š” โ€œ๋‘ ์‚ฌ๋žŒ์€ ์ตœ๊ทผ ๋ฉ”ํƒ€์— ์‚ฌ์žฅ ๊ฒธ ๋ถ€ํšŒ์žฅ์œผ๋กœ ํ•ฉ๋ฅ˜ํ•œ ๋””๋‚˜ ํŒŒ์›” ๋งฅ์ฝ”๋ฏน๊ณผ ๊ธด๋ฐ€ํžˆ ํ˜‘๋ ฅํ•ด, ์ •๋ถ€์™€ ๊ตญ๊ฐ€ ๋‹จ์œ„ ํŒŒํŠธ๋„ˆ์™€์˜ ํ˜‘์—…์„ ํ†ตํ•ด ๋ฉ”ํƒ€ ์ธํ”„๋ผ์˜ ๊ตฌ์ถ•๊ณผ ๋ฐฐ์น˜, ํˆฌ์ž, ์žฌ์› ์กฐ๋‹ฌ์„ ์ถ”์ง„ํ•  ๊ฒƒโ€์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ๋งฅ์ฝ”๋ฏน์€ ๋ฏธ๊ตญ ๋ฐฑ์•…๊ด€์—์„œ ๊ตญ๊ฐ€์•ˆ๋ณดํšŒ์˜(NSC) ์ „๋žต ๋‹ด๋‹น ๋ถ€๋ณด์ขŒ๊ด€์„ ์ง€๋‚ธ ์ธ๋ฌผ๋กœ, ๋ฐฐ์šฐ์ž์ธ ๋ฐ์ด๋ธŒ ๋งฅ์ฝ”๋ฏน์€ ํŽœ์‹ค๋ฒ ์ด๋‹ˆ์•„์ฃผ ์ถœ์‹  ์—ฐ๋ฐฉ ์ƒ์›์˜์›์ด์ž ์ƒ์› ์—๋„ˆ์ง€ ์†Œ์œ„์›ํšŒ ์œ„์›์žฅ์„ ๋งก๊ณ  ์žˆ๋‹ค.

์ด๋ฒˆ ์กฐ์น˜๋Š” ์ฃผ์š” ํด๋ผ์šฐ๋“œ ์—…์ฒด๊ฐ€ ๋„คํŠธ์›Œํฌ ์„ฑ๋Šฅ๊ณผ ์ „๋ ฅ ๊ณต๊ธ‰์— ๊ทน๋‹จ์ ์ธ ๋ถ€๋‹ด์„ ์ฃผ๋Š” ์ดˆ๋Œ€ํ˜• AI ํด๋Ÿฌ์Šคํ„ฐ ๊ตฌ์ถ• ๊ฒฝ์Ÿ์— ๋‚˜์„œ๊ณ  ์žˆ๋Š” ์ƒํ™ฉ์—์„œ ๋‚˜์™”๋‹ค. AI ์ธํ”„๋ผ ๊ทœ๋ชจ๊ฐ€ ์ปค์งˆ์ˆ˜๋ก ๋ฐ์ดํ„ฐ์„ผํ„ฐ์™€ ๋„คํŠธ์›Œํฌ, ์ „๋ ฅ ์„ค๋น„ ์ „๋ฐ˜์— ๋Œ€ํ•œ ๊ณ„ํš์„ ๋”์šฑ ๊ธด๋ฐ€ํ•˜๊ฒŒ ์กฐ์œจํ•  ํ•„์š”๊ฐ€ ์žˆ๋‹ค๋Š” ์ด์œ ๋‹ค.

์ด ๊ฐ™์€ ๊ทœ๋ชจ์—์„œ๋Š” ์ธํ”„๋ผ ์ œ์•ฝ์ด AI ํ™•์žฅ์˜ ๊ฒฐ์ •์ ์ธ ํ•œ๊ณ„ ์š”์ธ์œผ๋กœ ์ž‘์šฉํ•œ๋‹ค. ์‹ ๊ทœ ๋ฐ์ดํ„ฐ์„ผํ„ฐ๋ฅผ ์–ด๋””์— ๊ฑด์„คํ•  ์ˆ˜ ์žˆ๋Š”์ง€, ๊ฐ ์‹œ์„ค์„ ์–ด๋–ค ๋ฐฉ์‹์œผ๋กœ ์ƒํ˜ธ ์—ฐ๊ฒฐํ• ์ง€์™€ ๊ฐ™์€ ํŒ๋‹จ์ด AI ์ „๋žต ์ „๋ฐ˜์— ์ง์ ‘์ ์ธ ์˜ํ–ฅ์„ ๋ฏธ์น  ์ˆ˜ ์žˆ๋‹ค.

๋ฉ”ํƒ€๋Š” ์ตœ๊ทผ ๋น„์ŠคํŠธ๋ผ(Vistra), ํ…Œ๋ผํŒŒ์›Œ(TerraPower), ์˜คํด๋กœ(Oklo) ๋“ฑ ์ „๋ ฅ ๊ธฐ์—…๊ณผ ๋Œ€๊ทœ๋ชจ ๊ณ„์•ฝ์„ ์ฒด๊ฒฐํ•œ ๋ฐ” ์žˆ๋‹ค. ํ•ด๋‹น ๊ณ„์•ฝ์€ ์˜คํ•˜์ด์˜ค์™€ ํŽœ์‹ค๋ฒ ์ด๋‹ˆ์•„ ์ง€์—ญ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ํด๋Ÿฌ์Šคํ„ฐ์— ์ „๋ ฅ์„ ๊ณต๊ธ‰ํ•˜๊ธฐ ์œ„ํ•ด ์ตœ๋Œ€ 6.6GW, ์ฆ‰ ์›์ž๋ ฅ ๋ฐœ์ „์†Œ ์—ฌ๋Ÿฌ ๊ธฐ์— ํ•ด๋‹นํ•˜๋Š” ๊ทœ๋ชจ์˜ ์—๋„ˆ์ง€๋ฅผ ํ™•๋ณดํ•˜๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•˜๊ณ  ์žˆ๋‹ค.

๋ฐ์ดํ„ฐ์„ผํ„ฐ ๋„คํŠธ์›Œํฌ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ

์—…๊ณ„ ๋ถ„์„๊ฐ€๋“ค์€ AI ๊ฒฝ์Ÿ์—์„œ ๋„คํŠธ์›Œํฌ์™€ ๊ณ ์„ฑ๋Šฅ ์—ฐ๊ฒฐ ๊ธฐ์ˆ ์ด ๋” ์ด์ƒ ๋ณด์กฐ ์š”์†Œ๊ฐ€ ์•„๋‹ˆ๋ผ, AI ์„ฑ๋Šฅ๊ณผ ํ™•์žฅ์„ฑ์„ ์ขŒ์šฐํ•˜๋Š” ํ•ต์‹ฌ ๊ฒฝ์Ÿ ์š”์†Œ๋กœ ์ž๋ฆฌ ์žก๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ค€๋‹ค๊ณ  ๋ถ„์„ํ–ˆ๋‹ค.

์นด๋˜์Šค ์ธํ„ฐ๋‚ด์…”๋„์˜ ์ˆ˜์„๋ถ€์‚ฌ์žฅ ํˆด๋ฆฌ์นด ์‹ค์€ ๋ฉ”ํƒ€์˜ ์ด๋ฒˆ ์ด๋‹ˆ์…”ํ‹ฐ๋ธŒ๊ฐ€ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ๋„คํŠธ์›Œํฌ์˜ ๋ณ€ํ™” ํ•„์š”์„ฑ์„ ๋ถ„๋ช…ํžˆ ๋ณด์—ฌ์ค€๋‹ค๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. AI ์ธํ”„๋ผ ๋‚ด๋ถ€์—์„œ ๋ฐœ์ƒํ•˜๋Š” ๋ง‰๋Œ€ํ•œ ๋ฐ์ดํ„ฐ ํ๋ฆ„์„ ๊ฐ๋‹นํ•˜๋ ค๋ฉด, ๋„คํŠธ์›Œํฌ๊ฐ€ ๊ณ ๋Œ€์—ญํญ๊ณผ ์ดˆ์ €์ง€์—ฐ์„ ๋™์‹œ์— ์ถฉ์กฑํ•˜๋Š” ๋ฐฉํ–ฅ์œผ๋กœ ๋น ๋ฅด๊ฒŒ ์ง„ํ™”ํ•ด์•ผ ํ•œ๋‹ค๋Š” ๊ฒƒ์ด๋‹ค.

์‹ค์€ โ€œ๋ฐ์ดํ„ฐ์„ผํ„ฐ์˜ ๊ทœ๋ชจ์™€ GPU ์ง‘์ ๋„๊ฐ€ ์ปค์งˆ์ˆ˜๋ก ๋„คํŠธ์›Œํฌ์™€ ๊ด‘ํ•™ ๋ถ€ํ’ˆ ๊ณต๊ธ‰๋ง์— ๊ฐ€ํ•ด์ง€๋Š” ์••๋ฐ•์ด ๋”์šฑ ์ปค์งˆ ๊ฒƒโ€์ด๋ผ๋ฉฐ โ€œ์ด์— ๋”ฐ๋ผ ๋ณด๋‹ค ๊ณ ๋„ํ™”๋œ ์—ฐ๊ฒฐ ๊ธฐ์ˆ ๊ณผ ๋” ๋น ๋ฅธ ๊ด‘์„ฌ์œ ์— ๋Œ€ํ•œ ์ˆ˜์š”๊ฐ€ ์ฆ๊ฐ€ํ•˜๊ฒŒ ๋  ๊ฒƒโ€์ด๋ผ๊ณ  ์ง„๋‹จํ–ˆ๋‹ค.

์ผ๋ถ€ ์ „๋ฌธ๊ฐ€๋Š” ์ธํ”„๋ผ ์•„ํ‚คํ…์ฒ˜ ์ „๋ฐ˜์˜ ๋ณ€ํ™”๋กœ ์ด์–ด์งˆ ๊ฐ€๋Šฅ์„ฑ์—๋„ ์ฃผ๋ชฉํ–ˆ๋‹ค.

ํฌ๋ ˆ์Šคํ„ฐ์˜ ์ˆ˜์„ ์• ๋„๋ฆฌ์ŠคํŠธ ๋น„์Šค์™€์ง€ํŠธ ๋งˆํ•˜ํŒŒํŠธ๋ผ๋Š” โ€œ๋ฉ”ํƒ€๋Š” DSF(Disaggregated Scheduled Fabric)์™€ ๋น„์Šค์ผ€์ค„๋ง ํŒจ๋ธŒ๋ฆญ์„ ์ƒˆ๋กœ์šด 51๋ฐ๋ผ๋น„ํŠธ(Tbps)๊ธ‰ ์Šค์œ„์น˜ ๋ฐ ์Šค์ผ€์ผ์—… ๋„คํŠธ์›Œํ‚น์šฉ ์ด๋”๋„ท๊ณผ ํ•จ๊ป˜ ์‚ฌ์šฉํ•˜๊ณ  ์žˆ๋‹ค. ์ด๋กœ ์ธํ•ด ์Šค์œ„์น˜์šฉ ๋ฐ˜๋„์ฒด, ๊ด‘ ๋ชจ๋“ˆ, ์˜คํ”ˆ ๋ž™ ํ‘œ์ค€ ์ „๋ฐ˜์— ๊ฑธ์ณ ๋ถ€๋‹ด์ด ์ปค์ง€๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ์ด์–ด ๊ทธ๋Š” โ€œ๋ฉ”ํƒ€๊ฐ€ ๋ฐฑ๋ณธ ๋„คํŠธ์›Œํฌ ํ™•์žฅ๊ณผ AI ํด๋Ÿฌ์Šคํ„ฐ ํ™•๋Œ€๋ฅผ ์ถ”์ง„ํ•˜๋ฉด์„œ, ์ƒํƒœ๊ณ„ ์ „๋ฐ˜์— ๋” ๋น ๋ฅธ ๊ด‘ ์—ฐ๊ฒฐ ๊ธฐ์ˆ ๊ณผ ๋” ํฐ ๊ด‘์„ฌ์œ  ์šฉ๋Ÿ‰์„ ์š”๊ตฌํ•˜๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ๋ถ„์„ํ–ˆ๋‹ค.

์ด์ œ ๋„คํŠธ์›Œํฌ๋Š” ๋‹จ์ˆœํ•œ ๋ณด์กฐ ์ˆ˜๋‹จ์ด ์•„๋‹ˆ๋ผ AI ํ™•์žฅ์„ ๊ฐ€๋กœ๋ง‰๋Š” ํ•ต์‹ฌ ์ œ์•ฝ ์š”์†Œ๋กœ ๋ถ€์ƒํ•˜๊ณ  ์žˆ๋‹ค. ์‹ค์€ ๋Œ€๊ทœ๋ชจ AI ํ™˜๊ฒฝ์—์„œ ๋„คํŠธ์›Œํฌ ๋ณ‘๋ชฉ์„ ํ”ผํ•˜๋ ค๋Š” ์ฃผ์š” ํด๋ผ์šฐ๋“œ ์—…์ฒด์—๊ฒŒ ์ฐจ์„ธ๋Œ€ ์—ฐ๊ฒฐ ๊ธฐ์ˆ ์ด ์—ฐ์‚ฐ ์ž์› ์ ‘๊ทผ์„ฑ๋งŒํผ์ด๋‚˜ ์ค‘์š”ํ•œ ์š”์†Œ๊ฐ€ ๋˜๊ณ  ์žˆ๋‹ค๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

๋„คํŠธ์›Œํฌ ์•„ํ‚คํ…ํŠธ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ

์ˆ˜์‹ญ ๊ธฐ๊ฐ€์™€ํŠธ ๊ทœ๋ชจ์˜ AI ์ธํ”„๋ผ๋ฅผ ๊ณ„ํšํ•˜๋ ค๋ฉด ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์„ค๊ณ„์ž์™€ ๋„คํŠธ์›Œํฌ ์•„ํ‚คํ…ํŠธ๋Š” ๊ณผ๊ฑฐ๋ณด๋‹ค ํ›จ์”ฌ ๊ธด๋ฐ€ํ•˜๊ฒŒ ์ „๋ ฅ๊ณผ ๋„คํŠธ์›Œํฌ๋ฅผ ํ†ตํ•ฉํ•ด์•ผ ํ•œ๋‹ค.

์‹ค์€ โ€œ์•„ํ‚คํ…ํŠธ๋Š” ์—๋„ˆ์ง€ ์†Œ๋น„์™€ ์—ด ๋ถ„์‚ฐ, ์›Œํฌ๋กœ๋“œ ๋ฐฐ์น˜๋ฅผ ๊ท ํ˜• ์žˆ๊ฒŒ ์„ค๊ณ„ํ•˜๋Š” ๋™์‹œ์— ์ค‘๋ณต ๊ตฌ์„ฑ๊ณผ ์ง€๋Šฅํ˜• ๋ผ์šฐํŒ…์„ ํ†ตํ•ด ์•ˆ์ •์„ฑ์„ ํ™•๋ณดํ•ด์•ผ ํ•œ๋‹คโ€๋ผ๋ฉด์„œ, โ€œ์ด ์ •๋„ ๊ทœ๋ชจ์—์„œ๋Š” ์„ฑ๋Šฅ๊ณผ ์‹ ๋ขฐ์„ฑ์„ ์œ ์ง€ํ•˜๊ธฐ ์œ„ํ•ด ์ „๋ ฅ ์‚ฌ์šฉ์„ ๊ณ ๋ คํ•œ ์„ค๊ณ„์™€ ์ง€์—ฐ์„ ์ตœ์†Œํ™”ํ•œ ๋„คํŠธ์›Œํฌ๊ฐ€ ํ•„์ˆ˜์ โ€์ด๋ผ๊ณ  ๋ถ„์„ํ–ˆ๋‹ค.

๋งˆํ•˜ํŒŒํŠธ๋ผ๋Š” ํ”„๋กœ๋ฉ”ํ…Œ์šฐ์Šค์™€ ํ•˜์ดํŽ˜๋ฆฌ์˜จ๊ณผ ๊ฐ™์€ ๋Œ€ํ˜• AI ์Šˆํผํด๋Ÿฌ์Šคํ„ฐ๋ฅผ ์˜ˆ๋กœ ๋“ค๋ฉฐ, ์ด๋Ÿฌํ•œ ํ™˜๊ฒฝ์—์„œ๋Š” ์ง€์—ญ ๋‹จ์œ„๋กœ ๊ฒฌ๊ณ ํ•˜๊ฒŒ ์—ฐ๊ฒฐ๋œ ๋„คํŠธ์›Œํฌ์™€ ์œ ์—ฐํ•œ ์ธํ”„๋ผ ๊ตฌ์„ฑ, ๊ทธ๋ฆฌ๊ณ  ์ง€์†์ ์ธ ํ™•์žฅ์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•˜๋Š” ์ž„์‹œ ๋ฐฐ์น˜ ๊ตฌ์กฐ๊ฐ€ ์š”๊ตฌ๋œ๋‹ค๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ๊ทธ๋Š” ๋ฏธ๋ž˜ AI ์ˆ˜์š”๊ฐ€ ๋ถˆํ™•์‹คํ•œ ์ƒํ™ฉ์—์„œ, ์—ฌ๋Ÿฌ ์‹œ์„ค์— ์›Œํฌ๋กœ๋“œ๋ฅผ ๋ถ„์‚ฐํ•  ์ˆ˜ ์žˆ๋Š” ์„ค๊ณ„๊ฐ€ ์ค‘์š”ํ•ด์ง€๊ณ  ์žˆ๋‹ค๊ณ  ํ‰๊ฐ€ํ–ˆ๋‹ค.
dl-ciokorea@foundryco.com

5 essential skills every project manager needs during a data center transformation to the cloud

As organizations accelerate their shift from traditional data center environments to hybrid and multi-cloud architectures, the scale and complexity of these initiatives demand a new caliber of project leadership. Having recently led a multi-year enterprise-wide data center transformation with global stakeholders, Iโ€™ve seen firsthand that technology alone is not what ensures success. Leadership is the key.

Even the most advanced platforms and tools can fall short without a project manager who brings the right mindset, adaptability and technical fluency. These programs are simultaneously technical undertakings and organizational-change journeys.

Based on lessons learned from managing one of the most ambitious transformations in my organization, here are the five skills essential for any project manager responsible for navigating cloud and data center modernization.

1. Systems thinking & architectural awareness

Data center transformations operate at an enterprise scale, where no system exists in isolation. Every application, integration point and data flow is part of a wider ecosystem and understanding that ecosystem is critical from day one. Systems thinking means looking beyond servers and environments to examine business processes, downstream dependencies, data protection needs and operational realities.

This requires asking targeted questions such as:

  • What is the business impact if this application is down for four hours or more?
  • How many teams, processes or users rely on it?
  • What are its recovery objectives and how does it interact with upstream and downstream systems?

With these insights, project managers can make informed decisions about cutover sequencing and avoid grouping applications solely by physical infrastructure โ€” an approach that often leads to outages or misplaced dependencies. Indeed, a recent empirical study of migrating legacy systems to cloud platforms identified a lack of architectural mapping and understanding of interdependencies as a key risk factor in migration failures.

Takeaway

Architectural awareness isnโ€™t memorizing components; itโ€™s understanding how a single change reverberates across the entire enterprise system.

2. Elastic governance & proactive risk anticipation

Large-scale migrations rarely follow a predictable or linear path. They unfold in iterative phases, each introducing new variables, technical constraints and lessons learned. Because of this, a traditional waterfall approach quickly becomes a liability. What teams need instead is an elastic governance framework that provides structure while adapting to shifting realities.

Elastic governance means adjusting processes, decision models and approval flows as new insights surface. Each application and business unit often carries its own architecture, dependencies and constraints, so a one-size-fits-all model simply doesnโ€™t work. During our migration, daily interactions with implementation teams, developers and product owners gave me real-time visibility into emerging issues and allowed us to refine our approach continuously.

This approach mirrors trends highlighted in the ISACA Journalโ€™s 2023 article, โ€œRedefining Enterprise Cloud Technology Governance.โ€ ISACA argues that traditional governance frameworks are far too rigid for modern cloud environments. Instead, they advocate for adaptive, decentralized models that empower teams to respond quickly as new constraints and dependencies emerge.

Vendor-related challenges were especially common with aging legacy systems. Proactive engagement โ€” rather than reactive firefighting โ€” helped us avoid failures and maintain momentum.

Takeaway

Governance should guide, not grind. Flexibility is essential for managing uncertainty and sustaining progress in complex transformations.

3. Stakeholder coordination and strategic communication

In enterprise-wide transformation programs, stakeholder alignment is often the difference between controlled progress and project derailment. Every migration window, firewall rule adjustment, environment change or sequence shift requires close coordination across security, networking, infrastructure, operations, product teams and business leadership โ€” all operating with their own priorities and pressures.

Research shows that stakeholders often have different โ€œframesโ€ of a digital transformation and successful programs actively manage these perspectives to create shared understanding and alignment. Similarly, a 2023 KPMG report highlights that building trust among stakeholders โ€” particularly around risk, security and compliance โ€” is essential for successful cloud adoption.

A critical part of this role is translation. The project manager must convert technical constraints into clear, business-friendly updates while also translating business expectations into actionable direction for engineering teams. This dual fluency reduces misunderstanding and accelerates decision-making.

To maintain alignment, structured communication becomes essential. I established predictable rhythms โ€” daily standups, weekly product syncs, monthly executive briefings and shared dashboards โ€” to ensure transparency, quick escalation and consistent visibility into progress and risks.

Takeaway

The stronger and more structured the communication, the smoother and more predictable the migration.

4. Technical fluency & decision facilitation

Modernization initiatives involve ongoing decisions about whether to re-host, re-platform or re-architect applications. While a project manager doesnโ€™t need to be the most technical person in the room, they must understand the implications of each option well enough to facilitate informed decision-making.

Technical fluency builds credibility with developers, architects, vendors and deployment teams. It also enables the project manager to ask the right questions, challenge assumptions and guide discussions toward solutions. This is especially important given the โ€œ6 Rsโ€ of cloud migration โ€” re-host, re-platform, refactor (re-architect) and others โ€” which are commonly used to rationalize workloads based on business goals and technical fit.

Takeaway

Technical fluency enables clarity, connection and better decisions.

5. Resilience & change leadership

Data center transformations are long, complex and filled with uncertainties. Unexpected technical issues, compliance demands and shifting business priorities can slow down momentum and strain teams. According to the KPMG report mentioned earlier, many organizations struggle with operational resilience โ€” more than half experienced outages or compliance issues in their cloud operations over the past year. This reinforces the importance of proactive governance and risk management. In such environments, a resilient project manager provides clarity, maintains stability and ensures the team keeps moving forward.

During our project, an unexpected compliance mandate required rapid reprioritization and additional resources. With leadership support, we realigned the plan and still met the migration deadline. Maintaining team morale during such periods is just as important as technical delivery.

Takeaway

Resilient teams donโ€™t resist change; they stay confident through it.

Integrating the 5 skills: The project manager as transformation leader

A data center transformation is more than a technical project โ€” it reshapes processes, roles and behaviors across the organization. When these five skills come together, the project manager transitions from a delivery role into a true transformation leader.

  • Systems thinking eliminates hidden dependencies.
  • Elastic governance adapts to evolving needs.
  • Stakeholder coordination maintains across-the-board alignment.
  • Technical fluency builds trust and accelerates decision-making.
  • Resilience keeps teams focused during disruption.

The most effective transformation leaders balance discipline with flexibility.

Measuring success beyond migration

Traditional success metrics such as reduced downtime, regulatory compliance and cost optimization are important. But true success becomes clear only when the organization demonstrates improved adaptability and stronger collaboration between IT and the business.

When a project manager embeds adaptability deep into the organization, the transformation continues long after the final cutover.

The future-ready project manager

Looking ahead, managing a data center transformation a decade from now will be fundamentally different. The next generation of migrations will involve greater complexity, including advanced automation, AI-driven orchestration, multi-cloud environments and more sophisticated compliance and security requirements. Without continuous upskilling, project managers will struggle to lead confidently in this evolving landscape.

Future-ready leaders must be both technologically fluent and human-centered. They need to leverage data effectively, make decisions at the pace of AI and automation and understand emerging tools and methodologies. At the same time, they must maintain essential human leadership qualities โ€” trust, accountability, resilience and the ability to inspire teams under pressure.

By balancing these technical and human skills, project managers remain indispensable. They not only ensure that migrations succeed technically but also guide teams and organizations with purpose, clarity and adaptability, enabling sustainable transformation that goes beyond the immediate project and strengthens the organizationโ€™s long-term capabilities.

Closing thoughts

Data center transformation was not an easy migration, as it was a complicated and most ambitious undertaking by the organization. Orchestrating more than a hundred stakeholders was not an easy feat and we accomplished it with meticulous planning and risk management. Hence, a project manager with those five skills doesnโ€™t just lead, they become the transformation agents for the organization. As the saying goes: Real transformation happens when leadership turns complexity into clarity and uncertainty into forward motion.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?

์‚ผ์„ฑ์ „์ž, 2026๋…„ ๋ฉ”๋ชจ๋ฆฌ ๊ณต๊ธ‰ ๋ถ€์กฑ ๊ฒฝ๊ณ ยทยทยท์ „ ์‚ฐ์—… ๊ฐ€๊ฒฉ ๊ธ‰๋“ฑ ๋ถˆ๊ฐ€ํ”ผ

์‚ผ์„ฑ์ „์ž๋Š” 2026๋…„ ๋ฉ”๋ชจ๋ฆฌ ๋ฐ˜๋„์ฒด ๊ณต๊ธ‰ ๋ถ€์กฑ์œผ๋กœ ์ „์ž ์‚ฐ์—… ์ „๋ฐ˜์— ๊ฑธ์ณ ๊ฐ€๊ฒฉ ์ธ์ƒ์ด ๋ฐœ์ƒํ•  ๊ฒƒ์ด๋ผ๊ณ  ๊ฒฝ๊ณ ํ–ˆ๋‹ค. ์ตœ๋Œ€ ๊ทœ๋ชจ ์ƒ์‚ฐ ์—ญ๋Ÿ‰์„ ๊ฐ–์ถ˜ ์‚ผ์„ฑ์ „์ž์กฐ์ฐจ ์ž์‚ฌ ์ œํ’ˆ์ด ๊ฐ€๊ฒฉ ๊ธ‰๋“ฑ์—์„œ ์ž์œ ๋กœ์šธ ์ˆ˜ ์—†์Œ์„ ์ธ์ •ํ•œ ๋งŒํผ, ๊ธฐ์—… IT ๊ตฌ๋งค์ž๋Š” ์–ด๋А ๋ฒค๋”๋ฅผ ์„ ํƒํ•˜๋“  ๋น„์šฉ ์ƒ์Šน์„ ํ”ผํ•˜๊ธฐ ์–ด๋ ค์›Œ์ง€๊ณ  ์žˆ๋‹ค.

์‚ผ์„ฑ์ „์ž ๊ธ€๋กœ๋ฒŒ ๋งˆ์ผ€ํŒ… ์ด๊ด„ ์ด์›์ง„ ์‚ฌ์žฅ์€ ๋ธ”๋ฃธ๋ฒ„๊ทธ์™€์˜ ์ธํ„ฐ๋ทฐ์—์„œ ๋ฉ”๋ชจ๋ฆฌ ๋ฐ˜๋„์ฒด ๊ณต๊ธ‰ ๋ถ€์กฑ์ด ์‚ฐ์—… ์ „๋ฐ˜์˜ ๊ฐ€๊ฒฉ์— ์˜ํ–ฅ์„ ๋ฏธ์น  ๊ฒƒ์œผ๋กœ ๋‚ด๋‹ค๋ดค๋‹ค. ์ด ์‚ฌ์žฅ์€ โ€œ2026๋…„์—๋Š” ๋ฐ˜๋„์ฒด ๊ณต๊ธ‰๊ณผ ๊ด€๋ จํ•œ ๋ฌธ์ œ๊ฐ€ ๋ฐœ์ƒํ•  ๊ฒƒ์ด๋ฉฐ, ์ด๋Š” ์‚ผ์„ฑ์ „์ž๋งŒ์˜ ๋ฌธ์ œ๊ฐ€ ์•„๋‹ˆ๋ผ ๋ชจ๋“  ๊ธฐ์—…์— ์˜ํ–ฅ์„ ๋ฏธ์น  ๊ฒƒ์ด๋‹ค. ๊ณต๊ธ‰ ๋ฌธ์ œ๋Š” ์‚ฐ์—… ์ „๋ฐ˜์˜ ํ˜„์‹คโ€์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

ํ…Œํฌ์ธ์‚ฌ์ด์ธ ์˜ ๋ฐ˜๋„์ฒด ์• ๋„๋ฆฌ์ŠคํŠธ ๋งˆ๋‹ˆ์‹œ ๋ผ์™€ํŠธ๋Š” ์‚ผ์„ฑ์ „์ž ์ž์ฒด ์ œํ’ˆ๊ตฐ์กฐ์ฐจ ๊ฐ€๊ฒฉ ์••๋ฐ•์—์„œ ๋ณดํ˜ธ๋ฐ›์ง€ ๋ชปํ•  ์ •๋„๋กœ ์‹œ์žฅ ์—ญํ•™์ด ๋น ๋ฅด๊ฒŒ ๋ณ€ํ™”ํ•˜๊ณ  ์žˆ๋‹ค๊ณ  ๋ถ„์„ํ–ˆ๋‹ค. ๋ผ์™€ํŠธ๋Š” โ€œ๋ฉ”๋ชจ๋ฆฌ ์ œ์กฐ์‚ฌ๋Š” ๊ทœ๋ชจ์˜ ๊ฒฝ์ œ์™€ ์žฌ๊ณ  ๊ด€๋ฆฌ, ์žฅ๊ธฐ ๊ณ„์•ฝ์„ ํ†ตํ•ด ๊ธฐ์ˆ  ์ƒํƒœ๊ณ„์˜ ์™„์ถฉ ์—ญํ• ์„ ํ•ด์™”๋‹ค. ์‚ผ์„ฑ์ „์ž๊ฐ€ ์••๋„์ ์ธ ์ƒ์‚ฐ ๋Šฅ๋ ฅ์„ ๋ณด์œ ํ•˜๊ณ  ์žˆ์Œ์—๋„ ๋ณ€๋™์„ฑ์„ ์™„ํ™”ํ•˜์ง€ ๋ชปํ•œ๋‹ค๋Š” ์ ์€ ์‹œ์žฅ์ด ๋ถˆ๊ท ํ˜• ์ƒํƒœ์— ๋†“์—ฌ ์žˆ์Œ์„ ์‹œ์‚ฌํ•œ๋‹คโ€๋ผ๊ณ  ํ‰๊ฐ€ํ–ˆ๋‹ค.

AI ์ธํ”„๋ผ๊ฐ€ ๊ธฐ์กด ๋ฉ”๋ชจ๋ฆฌ ๊ณต๊ธ‰์„ ์ž ์‹

์ด ๊ฐ™์€ ์‹œ์žฅ ๋ถˆ๊ท ํ˜•์€ ๋ฉ”๋ชจ๋ฆฌ ์ œ์กฐ์‚ฌ๊ฐ€ AI ๋ฐ์ดํ„ฐ์„ผํ„ฐ์šฉ HBM์— ์ƒ์‚ฐ ์—ญ๋Ÿ‰์„ ์ง‘์ค‘ํ•˜๋ฉด์„œ ๋ฐœ์ƒํ•˜๊ณ  ์žˆ๋‹ค. HBM์ด ๊ธฐ์กด ๋””๋žจ๋ณด๋‹ค ๋†’์€ ์ˆ˜์ต์„ฑ์„ ์ œ๊ณตํ•˜๋Š” ๋งŒํผ, ์‚ผ์„ฑ์ „์ž์™€ SKํ•˜์ด๋‹‰์Šค, ๋งˆ์ดํฌ๋ก  ๋“ฑ์€ ์ „ํ†ต์ ์ธ ๊ธฐ์—…ยท์†Œ๋น„์ž์šฉ ๋ฉ”๋ชจ๋ฆฌ ์ œํ’ˆ์—์„œ ์ƒ์‚ฐ ๋Šฅ๋ ฅ์„ ์ค„์ด๊ณ  HBM์œผ๋กœ ์ „ํ™˜ํ•˜๊ณ  ์žˆ๋‹ค.

๊ฐ€ํŠธ๋„ˆ์˜ ์ˆ˜์„ ์• ๋„๋ฆฌ์ŠคํŠธ ์นด๋‹ˆ์‹œ์นด ์ฐจ์šฐํ•œ์€ โ€œ์ง€์†์ ์ธ AI ํˆฌ์ž๋กœ ๋ฉ”๋ชจ๋ฆฌ ์ˆ˜์š”๋Š” ๊ฐ•์„ธ๋ฅผ ๋ณด์ด๊ณ  ์žˆ๋‹ค. AI ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์— ์‚ฌ์šฉ๋˜๋Š” HBM์€ ๊ธฐ์กด ๋ฉ”๋ชจ๋ฆฌ๋ณด๋‹ค ๋†’์€ ๊ฐ€๊ฒฉ์ด ํ˜•์„ฑ๋˜๊ธฐ ๋•Œ๋ฌธ์—, ๋ฒค๋”๋“ค์ด ๋” ๋งŽ์€ ์ƒ์‚ฐ ์—ญ๋Ÿ‰์„ HBM์— ๋ฐฐ๋ถ„ํ•˜๊ณ  ์žˆ๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

๋งˆ์ดํฌ๋ก  ์ž„์›์— ๋”ฐ๋ฅด๋ฉด AI ๊ฐ€์†๊ธฐ์šฉ HBM ์ƒ์‚ฐ์€ ๊ธฐ๊ฐ€๋ฐ”์ดํŠธ๋‹น ํ‘œ์ค€ ๋””๋žจ ๋Œ€๋น„ ์•ฝ 3๋ฐฐ์˜ ์›จ์ดํผ ์šฉ๋Ÿ‰์„ ์†Œ๋ชจํ•œ๋‹ค. ์ด๋กœ ์ธํ•ด ๋ฉ”๋ชจ๋ฆฌ ์ œ์กฐ์‚ฌ๋Š” ์†Œ๋น„์ž ๋ฐ ๊ธฐ์—…์šฉ ์ œํ’ˆ์—์„œ ์ƒ์‚ฐ์„ ์žฌ์กฐ์ •ํ•  ์ˆ˜๋ฐ–์— ์—†์œผ๋ฉฐ, ๊ทธ ๊ฒฐ๊ณผ ๋ฌผ๋Ÿ‰์ด ์ ์€ ์‚ฐ์—… ๋ฐ PC ๊ณ ๊ฐ์„ ์ค‘์‹ฌ์œผ๋กœ ๊ธฐ์กด ๋””๋žจ ๊ณต๊ธ‰์ด ๊ฐ์†Œํ•˜๊ณ  ์žˆ๋‹ค๊ณ  ์ฐจ์šฐํ•œ์€ ๋ถ„์„ํ–ˆ๋‹ค.

SKํ•˜์ด๋‹‰์Šค๋Š” ์ง€๋‚œํ•ด 10์›” ์‹ค์  ๋ฐœํ‘œ์—์„œ 2026๋…„ HBM, ๋””๋žจ, ๋‚ธ๋“œ ์ƒ์‚ฐ ์—ญ๋Ÿ‰์ด โ€œ์‚ฌ์‹ค์ƒ ์™„ํŒ ์ƒํƒœโ€๋ผ๊ณ  ๋ฐํ˜”๋‹ค. ๋งˆ์ดํฌ๋ก  ์—ญ์‹œ ์ตœ๊ทผ ์†Œ๋น„์ž ๋ฉ”๋ชจ๋ฆฌ ์‹œ์žฅ์—์„œ ์™„์ „ํžˆ ์ฒ ์ˆ˜ํ•˜๊ณ , ๊ธฐ์—… ๋ฐ AI ๊ณ ๊ฐ์— ์—ญ๋Ÿ‰์„ ์ง‘์ค‘ํ•˜๋Š” ์ „๋žต์œผ๋กœ ๋ฐฉํ–ฅ์„ ์ „ํ™˜ํ–ˆ๋‹ค.

๊ธฐ์—…์šฉ ํ•˜๋“œ์›จ์–ด ๋น„์šฉ ๊ธ‰๋“ฑ

์ด ๊ฐ™์€ ๊ณต๊ธ‰ ์ œ์•ฝ์œผ๋กœ ์ธํ•ด ๊ธฐ์—…์šฉ ํ•˜๋“œ์›จ์–ด ์ „๋ฐ˜์˜ ๊ฐ€๊ฒฉ์ด ๊ธ‰๋“ฑํ•˜๊ณ  ์žˆ๋‹ค. ์‚ผ์„ฑ์ „์ž๋Š” ์ง€๋‚œํ•ด 9์›” 32GB DDR5 ๋ชจ๋“ˆ ๊ฐ€๊ฒฉ์„ 149๋‹ฌ๋Ÿฌ์—์„œ 239๋‹ฌ๋Ÿฌ๋กœ ์•ฝ 60% ์ธ์ƒํ–ˆ์œผ๋ฉฐ, DDR5 ๊ณ„์•ฝ ๊ฐ€๊ฒฉ์€ 2025๋…„ ์ดˆ ๊ฐœ๋‹น ์•ฝ 7๋‹ฌ๋Ÿฌ์—์„œ 19.50๋‹ฌ๋Ÿฌ๋กœ 100% ์ด์ƒ ๊ธ‰๋“ฑํ–ˆ๋‹ค.

์นด์šดํ„ฐํฌ์ธํŠธ๋ฆฌ์„œ์น˜์— ๋”ฐ๋ฅด๋ฉด ๋””๋žจ ๊ฐ€๊ฒฉ์€ ์˜ฌํ•ด ๋“ค์–ด ์ด๋ฏธ ์•ฝ 50% ์ƒ์Šนํ–ˆ์œผ๋ฉฐ, 2025๋…„ 4๋ถ„๊ธฐ์— ์ถ”๊ฐ€๋กœ 30% ์˜ค๋ฅด๊ณ  2026๋…„ ์ดˆ์—๋Š” ๋‹ค์‹œ 20% ์ƒ์Šนํ•  ๊ฒƒ์œผ๋กœ ์ „๋ง๋œ๋‹ค. ์ด ๊ธฐ๊ด€์€ ๊ธฐ์—… ๋ฐ์ดํ„ฐ์„ผํ„ฐ์—์„œ ๋„๋ฆฌ ์‚ฌ์šฉ๋˜๋Š” DDR5 64GB RDIMM ๋ชจ๋“ˆ ๊ฐ€๊ฒฉ์ด 2026๋…„ ๋ง์—๋Š” 2025๋…„ ์ดˆ ๋Œ€๋น„ 2๋ฐฐ ์ˆ˜์ค€์— ์ด๋ฅผ ์ˆ˜ ์žˆ๋‹ค๊ณ  ๋‚ด๋‹ค๋ดค๋‹ค.

์ฐจ์šฐํ•œ์€ ๊ฐ€ํŠธ๋„ˆ๊ฐ€ ๊ธฐ์กด ๋””๋žจ ์‹œ์žฅ์—์„œ ์‹ฌ๊ฐํ•œ ๊ณต๊ธ‰ ๋ถ€์กฑ์ด ๋ฐœ์ƒํ•˜๋ฉด์„œ 2026๋…„ ๋””๋žจ ๊ฐ€๊ฒฉ์ด 47% ์ƒ์Šนํ•  ๊ฒƒ์œผ๋กœ ์ „๋งํ–ˆ๋‹ค๊ณ  ์ „ํ–ˆ๋‹ค.

์ฃผ์š” ํด๋ผ์šฐ๋“œ ์—…์ฒด์˜ ๊ตฌ๋งค ํ˜‘์ƒ๋ ฅ์€ ์ฆ๊ฐ€

์ด๋Ÿฐ ๊ฐ€๊ฒฉ ์••๋ฐ•๊ณผ ๊ณต๊ธ‰ ์ œ์•ฝ์ด ๊ธฐ์—… IT ๊ตฌ๋งค ์‹œ์žฅ์˜ ํž˜์˜ ๊ท ํ˜•์„ ์žฌํŽธํ•˜๊ณ  ์žˆ๋‹ค. ๊ธฐ์—… ์กฐ๋‹ฌ ๊ด€์ ์—์„œ๋Š” ๋ฉ”๋ชจ๋ฆฌ ์ œ์กฐ์‚ฌ์˜ ๊ทœ๋ชจ๊ฐ€ ํฌ๋‹ค๊ณ  ํ•ด์„œ ์•ˆ์ •์„ฑ์„ ๋ณด์žฅ๋ฐ›๊ธด ์–ด๋ ค์šด ์ƒํ™ฉ์ด๋‹ค. ๋ผ์™€ํŠธ๋Š” โ€œ2026๋…„์œผ๋กœ ๊ฐˆ์ˆ˜๋ก ๊ณต๊ธ‰ ๊ฒฝ์Ÿ์ด ์‹ฌํ™”๋˜๋ฉด์„œ ๋ฌผ๋Ÿ‰๋ณด๋‹ค๋Š” ์ œ์กฐ์‚ฌ์™€์˜ ์ „๋žต์  ๊ด€๊ณ„๊ฐ€ ๊ตฌ๋งค ํ˜‘์ƒ๋ ฅ์— ๋” ํฐ ์˜ํ–ฅ์„ ๋ผ์น  ๊ฒƒโ€์ด๋ผ๊ณ  ๋ถ„์„ํ–ˆ๋‹ค.

์ด๋ฏธ ์ฃผ์š” ํด๋ผ์šฐ๋“œ ์—…์ฒด๋Š” ์žฅ๊ธฐ ๊ณต๊ธ‰ ๊ณ„์•ฝ๊ณผ ์ƒ์‚ฐ ๋Šฅ๋ ฅ ์‚ฌ์ „ ํ™•๋ณด, ์ง์ ‘์ ์ธ ํŒน ํˆฌ์ž ๋“ฑ์„ ํ†ตํ•ด ๋‚ฎ์€ ๋น„์šฉ๊ณผ ์•ˆ์ •์ ์ธ ๋ฌผ๋Ÿ‰์„ ๋™์‹œ์— ํ™•๋ณดํ•˜๊ณ  ์žˆ๋‹ค. ๋ฐ˜๋ฉด ์ค‘๊ฒฌ ๊ธฐ์—…์€ ์ƒ๋Œ€์ ์œผ๋กœ ์งง์€ ๊ณ„์•ฝ๊ณผ ์ŠคํŒŸ ๊ตฌ๋งค์— ์˜์กดํ•˜๋ฉฐ, ๋Œ€๊ทœ๋ชจ ๊ตฌ๋งค์ž๊ฐ€ ์šฐ์„ ์ ์œผ๋กœ ๋ฌผ๋Ÿ‰์„ ํ™•๋ณดํ•œ ์ดํ›„ ๋‚จ์€ ์ƒ์‚ฐ ๋Šฅ๋ ฅ์„ ๋‘๊ณ  ๊ฒฝ์Ÿํ•ด์•ผ ํ•˜๋Š” ๊ตฌ์กฐ์— ๋†“์˜€๋‹ค.

๋ผ์™€ํŠธ๋Š” โ€œ์ด ๊ฐ™์€ ๋ถˆ๊ท ํ˜•์€ ์ค‘๊ฒฌ ๊ธฐ์—…์— ์ด์ค‘์˜ ์ œ์•ฝ์„ ๊ฐ€ํ•œ๋‹คโ€๋ผ๋ฉฐ โ€œํˆฌ์ž… ๋น„์šฉ ์ƒ์Šน๊ณผ ๋‚ฉ๊ธฐ ์ง€์—ฐ์ด ๋™์‹œ์— ๋ฐœ์ƒํ•˜๋ฉด์„œ ์ธํ”„๋ผ ํ™•์žฅ์ด๋‚˜ ์‹ ๊ทœ ์›Œํฌ๋กœ๋“œ ๋ฐฐํฌ, ๋น ๋ฅธ ํ˜์‹ ์„ ์ถ”์ง„ํ•˜๋Š” ๋ฐ ์ง์ ‘์ ์ธ ์ œ์•ฝ์ด ๋œ๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

์‚ผ์„ฑ์ „์ž๋Š” ํ‰ํƒ ์‚ฌ์—…์žฅ์— ์‹ ๊ทœ ๋ฉ”๋ชจ๋ฆฌ ์ƒ์‚ฐ ๋ผ์ธ์„ ๊ตฌ์ถ•ํ•  ๊ณ„ํš์„ ๋ฐœํ‘œํ–ˆ์ง€๋งŒ, ๋ณธ๊ฒฉ์ ์ธ ์–‘์‚ฐ์€ 2028๋…„์—์•ผ ์‹œ์ž‘๋  ์˜ˆ์ •์ด๋‹ค.

๊ธฐ์—… ์ž…์žฅ์—์„œ๋Š” 2026๋…„ ์ดํ›„ ๋ฉ”๋ชจ๋ฆฌ ์ˆ˜๊ธ‰ ์ œ์•ฝ์ด IT ์กฐ๋‹ฌ ์ „๋žต ์ „๋ฐ˜์— ์˜ํ–ฅ์„ ๋ฏธ์น  ๊ฐ€๋Šฅ์„ฑ์ด ๋†’๋‹ค. ์ด์— ๋”ฐ๋ผ ํ•˜๋“œ์›จ์–ด ๊ณ„ํš ๋‹จ๊ณ„์—์„œ๋ถ€ํ„ฐ ๊ณต๊ธ‰ ๊ฐ€์šฉ์„ฑ๊ณผ ๋น„์šฉ ๋ณ€๋™์„ฑ์„ ํ•จ๊ป˜ ๊ณ ๋ คํ•ด์•ผ ํ•  ํ•„์š”์„ฑ์ด ์ปค์ง€๊ณ  ์žˆ๋‹ค.
dl-ciokorea@foundryco.com

์นผ๋Ÿผ | ๋ฒ”์šฉ ์ปดํ“จํŒ… ์ดํ›„์˜ ์‹œ๋Œ€, AI๊ฐ€ ์Šคํ† ๋ฆฌ์ง€๋ฅผ ์žฌํŽธํ•œ๋‹ค

์—”๋น„๋””์•„ CEO์ธ ์  ์Šจ ํ™ฉ์˜ ์ฃผ์žฅ๋Œ€๋กœ ๋ฒ”์šฉ ์ปดํ“จํŒ… ์‹œ๋Œ€๊ฐ€ ์ €๋ฌผ๊ณ  ์žˆ๋‹ค๋ฉด, ์ง€๊ธˆ์€ 200์—ฌ ๋…„ ์ „ ํ•ต์‹ฌ ๋™๋ ฅ์›์ด ๋งˆ๋ ฅ์—์„œ ์ฆ๊ธฐ๋กœ ์ „ํ™˜๋˜๋˜ ์‹œ๊ธฐ์— ๋ฒ„๊ธˆ๊ฐ€๋Š” ๊ทผ๋ณธ์ ์ธ ๋ณ€ํ™”๊ฐ€ ์ด๋ค„์ง€๊ณ  ์žˆ๋‹ค๊ณ  ๋ณผ ์ˆ˜ ์žˆ๋‹ค.

์ „ํ™˜์˜ ์ค‘์‹ฌ์—๋Š” AI ํ™•์‚ฐ๊ณผ ๋งž๋ฌผ๋ ค ์ƒˆ๋กญ๊ฒŒ ์žฌ๊ตฌ์„ฑ๋˜๊ณ  ์žˆ๋Š” ๋ฐ์ดํ„ฐ ์ธํ”„๋ผ ์ „๋ฐ˜์ด ์ž๋ฆฌํ•˜๊ณ  ์žˆ๋‹ค. ์ „๋ก€ ์—†๋Š” ์ˆ˜์ค€์˜ ์—ฐ์‚ฐ ์„ฑ๋Šฅ์„ ๋’ท๋ฐ›์นจํ•˜๊ธฐ ์œ„ํ•ด ๋ฐ์ดํ„ฐ ์ €์žฅ๊ณผ ์ ‘๊ทผ ๋ฐฉ์‹์—์„œ๋„ ์†๋„์™€ ์‹ ๋ขฐ์„ฑ, ํ™•์žฅ์„ฑ์ด ๋™์‹œ์— ์š”๊ตฌ๋˜๊ณ  ์žˆ๋‹ค.

AI๋Š” ์ง€๊ธˆ๊นŒ์ง€ ๋ณผ ์ˆ˜ ์—†์—ˆ๋˜ ์ˆ˜์ค€์˜ ๋ฐ์ดํ„ฐ ์ง‘์•ฝ์  ์›Œํฌ๋กœ๋“œ๋ฅผ ๋งŒ๋“ค์–ด๋‚ด๋ฉฐ ๊ธฐ์—… ์ธํ”„๋ผ์˜ ๊ตฌ์กฐ ์ž์ฒด๋ฅผ ๋น ๋ฅด๊ฒŒ ๋ฐ”๊พธ๊ณ  ์žˆ๋‹ค. ์ „ ์„ธ๊ณ„ ๋ฐ์ดํ„ฐ์„ผํ„ฐ ์šฉ๋Ÿ‰ ํ™•์žฅ์— ํˆฌ์ž…๋˜๋Š” ๋ง‰๋Œ€ํ•œ ํˆฌ์ž ๊ทœ๋ชจ๊ฐ€ ์ด๋ฅผ ๋ณด์—ฌ์ค€๋‹ค. ์ตœ๊ทผ ์ž‡๋”ฐ๋ฅธ ๋ฐœํ‘œ ๊ฐ€์šด๋ฐ ๋ฉ”ํƒ€์˜ 6,000์–ต ๋‹ฌ๋Ÿฌ ํˆฌ์ž ๊ณ„ํš๋„ ๊ทธ์ค‘ ํ•˜๋‚˜๋‹ค. ์ง€๋‚œํ•ด 4์›” ๋งฅํ‚จ์ง€๋Š” โ€œ์—ฐ์‚ฐ ์ˆ˜์š” ์ฆ๊ฐ€ ์†๋„๋ฅผ ๋”ฐ๋ผ์žก๊ธฐ ์œ„ํ•ด ํ•„์š”ํ•œ ๋น„์šฉ์ด 7์กฐ ๋‹ฌ๋Ÿฌ์— ๋‹ฌํ•  ๊ฒƒโ€์ด๋ผ๊ณ  ์ง„๋‹จํ•œ ๋ฐ” ์žˆ๋‹ค. AI ํ™•์‚ฐ์˜ ํ๋ฆ„์ด ์œ ์ง€๋œ๋‹ค๋ฉด ์ด ์ถ”์‚ฐ์น˜๋Š” ๋” ์ปค์งˆ ๊ฐ€๋Šฅ์„ฑ๋„ ์žˆ๋‹ค.

์ด ๊ฐ™์€ ํ๋ฆ„์€ ๋ฐ์ดํ„ฐ ์Šคํ† ๋ฆฌ์ง€์—๋„ ๊ทผ๋ณธ์ ์ธ ๋ณ€ํ™”๋ฅผ ์š”๊ตฌํ•˜๊ณ  ์žˆ๋‹ค. ๊ธฐ์กด ์Šคํ† ๋ฆฌ์ง€๋Š” ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋‚˜ ๊ฐ€์ƒํ™”์ฒ˜๋Ÿผ ์˜ˆ์ธก ๊ฐ€๋Šฅํ•˜๊ณ  ์ˆœ์ฐจ์ ์ธ ์›Œํฌ๋กœ๋“œ๋ฅผ ์ „์ œ๋กœ ์„ค๊ณ„๋๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ AI ํ™˜๊ฒฝ์—์„œ๋Š” ์ˆ˜์ฒœ ๊ฐœ์˜ GPU ์Šค๋ ˆ๋“œ๊ฐ€ ๋ณ‘๋ ฌ ๋ฐ ๋ฌด์ž‘์œ„, ๊ณ ๋Œ€์—ญํญ ์ ‘๊ทผ์„ ๋™์‹œ์— ์š”๊ตฌํ•˜๋ฉด์„œ ๊ธฐ์กด ๋ชจ๋ธ์„ ๋’คํ”๋“ค๊ณ  ์žˆ๋‹ค.

์ด๋กœ ์ธํ•ด ๋ฐœ์ƒํ•˜๋Š” ์„ฑ๋Šฅ ๋ฌธ์ œ๋Š” ์ธํ”„๋ผ ์ „๋ฐ˜์— ์—ฐ์‡„์ ์œผ๋กœ ํ™•์‚ฐ๋œ๋‹ค. ์Šคํ† ๋ฆฌ์ง€๊ฐ€ ์ด๋ฅผ ๊ฐ๋‹นํ•˜์ง€ ๋ชปํ•˜๋ฉด GPU๋Š” ๋ฐ์ดํ„ฐ๋ฅผ ๊ธฐ๋‹ค๋ฆฌ๋ฉฐ ์œ ํœด ์ƒํƒœ์— ๋†“์ด๊ณ , ํ•™์Šต ์ฃผ๊ธฐ๋Š” ์ง€์—ฐ๋˜๋ฉฐ, ์ „์ฒด ๋น„์šฉ์€ ๊ธ‰๊ฒฉํžˆ ์ฆ๊ฐ€ํ•œ๋‹ค. ์ถฉ๋ถ„ํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ๊ณต๊ธ‰๋ฐ›์ง€ ๋ชปํ•œ GPU๊ฐ€ ํ•œ ์‹œ๊ฐ„์ด๋ผ๋„ ์ง€์—ฐ๋˜๋ฉด ROI๋ฅผ ํ™•๋ณดํ•˜๋Š” ์‹œ๊ธฐ๋„ ๋Šฆ์ถฐ์ง„๋‹ค. ํ•™์Šต ์ž์ฒด๊ฐ€ ํˆฌ์ž์ด๊ธฐ ๋•Œ๋ฌธ์—, ์ค‘๋‹จ๋˜๊ฑฐ๋‚˜ ๋น„ํšจ์œจ์ ์ธ ํ•™์Šต ์ฃผ๊ธฐ๋Š” ๊ฐ€์น˜ ์‹คํ˜„๊นŒ์ง€ ๊ฑธ๋ฆฌ๋Š” ์‹œ๊ฐ„์„ ๋Š˜๋ฆฐ๋‹ค. ์•„์šธ๋Ÿฌ ๋ฐ์ดํ„ฐ๊ฐ€ ์†์ƒ๋˜๊ฑฐ๋‚˜ ์œ ์‹ค๋˜๋ฉด ๋ชจ๋ธ ์ „์ฒด๋ฅผ ๋‹ค์‹œ ํ•™์Šตํ•ด์•ผ ํ•˜๋Š” ๊ฒฝ์šฐ๋„ ๋งŽ์•„, ์˜ˆ์ƒ์น˜ ๋ชปํ•œ ๋ง‰๋Œ€ํ•œ ๋น„์šฉ์ด ๋ฐœ์ƒํ•  ์ˆ˜๋„ ์žˆ๋‹ค.

์ด ๋ฌธ์ œ์˜ ์˜ํ–ฅ์€ ํ•™์Šต ๋‹จ๊ณ„์—๋งŒ ๊ตญํ•œ๋˜์ง€ ์•Š๋Š”๋‹ค. AI์—์„œ ์‹ค์ œ ์ˆ˜์ต์„ ์ฐฝ์ถœํ•˜๋Š” ๋‹จ๊ณ„๋Š” ์ถ”๋ก ์ด๋ฉฐ, ๋А๋ฆฌ๊ฑฐ๋‚˜ ๋ถˆ์•ˆ์ •ํ•œ ๋ฐ์ดํ„ฐ ํŒŒ์ดํ”„๋ผ์ธ์€ AI ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์ด ์ฐฝ์ถœํ•  ์ˆ˜ ์žˆ๋Š” ๋งค์ถœ๊ณผ ์‚ฌ์—… ์„ฑ๊ณผ๋ฅผ ์ง์ ‘์ ์œผ๋กœ ๋–จ์–ด๋œจ๋ฆฐ๋‹ค. ์ด์— ๋Œ€์‘ํ•ด ๊ธฐ์กด ์Šคํ† ๋ฆฌ์ง€ ๋ฒค๋”๋Š” ์•„ํ‚คํ…์ฒ˜๋ฅผ AI ์ˆ˜์š”์— ๋งž๊ฒŒ ๊ฐœ์กฐํ•˜๋ ค ํ•˜๊ณ  ์žˆ์ง€๋งŒ, ์ƒ๋‹น์ˆ˜๋Š” ์—ฌ์ „ํžˆ ์„ฑ๋Šฅ๊ณผ ํ™•์žฅ์„ฑ ์ธก๋ฉด์—์„œ ํ•œ๊ณ„๋ฅผ ๋ฒ—์–ด๋‚˜์ง€ ๋ชปํ•˜๊ณ  ์žˆ๋‹ค.

๊ฒฐ๊ตญ ๋ณ€ํ™”๋Š” ๋ถˆ๊ฐ€ํ”ผํ•˜๋‹ค. ์ถœ๋ฐœ์ ์€ AI๊ฐ€ ๋ชฉ์ ์— ๋งž๊ฒŒ ์„ค๊ณ„๋œ, ๋ณธ์งˆ์ ์œผ๋กœ ๊ณ ์„ฑ๋Šฅ์„ ์ „์ œ๋กœ ํ•œ ์Šคํ† ๋ฆฌ์ง€ ์‹œ์Šคํ…œ์„ ํ•„์š”๋กœ ํ•œ๋‹ค๋Š” ์ธ์‹์ด๋‹ค.

์‹ ๋ขฐ์˜ ๊ธฐ๋ณธ ์š”์†Œ

์ด ๊ฐ™์€ ์„ฑ๋Šฅ ์••๋ฐ•์€ โ€˜์‹ ๋ขฐโ€™๋ผ๋Š” ๋” ๊ทผ๋ณธ์ ์ธ ๋ฌธ์ œ๋ฅผ ๋“œ๋Ÿฌ๋‚ธ๋‹ค. ๋Œ€๊ทœ๋ชจ AI ๋ชจ๋ธ์€ ํ•™์Šต ๋ฐ์ดํ„ฐ์— ์ค‘๋‹จ ์—†์ด ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ์ „์ œ๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ํ•œ๋‹ค. ๋ฉ”ํƒ€๋ฐ์ดํ„ฐ ์„œ๋ฒ„ ์žฅ์• ๋‚˜ ๋ฐ์ดํ„ฐ ์†์ƒ ๋“ฑ ์–ด๋–ค ํ˜•ํƒœ๋กœ๋ผ๋„ ์ค‘๋‹จ์ด ๋‚˜ํƒ€๋‚˜๋ฉด ์ƒ์‚ฐ์„ฑ์€ ํฌ๊ฒŒ ๋–จ์–ด์ง€๊ณ  ๊ฒฐ๊ณผ์˜ ์‹ ๋ขฐ์„ฑ์ด ํ›ผ์†๋  ์ˆ˜ ์žˆ๋‹ค.

์ด ๋งฅ๋ฝ์—์„œ ์‹ ๋ขฐ์„ฑ์€ ํ•˜๋‚˜์˜ ์ง€ํ‘œ๋กœ๋งŒ ์„ค๋ช…๋˜์ง€ ์•Š๋Š”๋‹ค. ์—ฌ๊ธฐ์—” ๋‚ด๊ตฌ์„ฑ, ๊ฐ€์šฉ์„ฑ, ๋ณต๊ตฌ ๊ฐ€๋Šฅ์„ฑ๋„ ํฌํ•จ๋œ๋‹ค. ๋‹จ์ˆœํ•œ ๊ธฐ์ˆ ์  ์•ˆ์ „์žฅ์น˜์˜ ๋ฌธ์ œ๊ฐ€ ์•„๋‹ˆ๋ผ, ์‹œ์Šคํ…œ์„ ์ค‘๋‹จ ์—†์ด ์šด์˜ํ•˜๊ณ  ๋ฐ์ดํ„ฐ ๋ฌด๊ฒฐ์„ฑ์„ ์ง€์ผœ๋‚ผ ์ˆ˜ ์žˆ๋Š”์ง€๊ฐ€ ๊ณง AI ํˆฌ์ž๋ฅผ ์‹ค์ œ ์„ฑ๊ณผ๋กœ ์ด์–ด์ง€๊ฒŒ ํ•˜๋Š” ํ•ต์‹ฌ ์š”์†Œ์ด๊ธฐ ๋•Œ๋ฌธ์ด๋‹ค.

๋ฌธ์ œ๋Š” ์—ฌ์ „ํžˆ ๋งŽ์€ ์‹œ์Šคํ…œ์ด ์—ฌ๋Ÿฌ ๊ฐœ์˜ ๋ฌผ๋ฆฌ์  ๋””์Šคํฌ๋ฅผ ๋…ผ๋ฆฌ์ ์œผ๋กœ ํ•˜๋‚˜๋กœ ๋ฌถ๋Š” ๋กœ์ปฌ RAID(Redundant Array of Independent Disks)๋‚˜ ๊ณ ๊ฐ€์šฉ์„ฑ(HA) ํŽ˜์–ด ์•„ํ‚คํ…์ฒ˜์— ์˜์กดํ•˜๊ณ  ์žˆ๋‹ค๋Š” ์ ์ด๋‹ค. ์ด๋Ÿฐ ๊ตฌ์กฐ๋Š” ์†Œ๊ทœ๋ชจ ์žฅ์• ์—๋Š” ๋Œ€์‘ํ•  ์ˆ˜ ์žˆ์ง€๋งŒ, AI ํ™˜๊ฒฝ์ฒ˜๋Ÿผ ์ˆ˜์‹ญ, ์ˆ˜๋ฐฑ ๋Œ€ ๋…ธ๋“œ๊ฐ€ ๋™์‹œ์— ๋™์ž‘ํ•˜๋Š” ์ƒํ™ฉ์—์„œ๋Š” ํ•œ๊ณ„๋ฅผ ๋“œ๋Ÿฌ๋‚ธ๋‹ค. ๋ฐ˜๋ฉด ์ตœ์‹  ์„ค๊ณ„๋Š” ๋‹ค์ค‘ ๋‹จ๊ณ„ ์†Œ๊ฑฐ ์ฝ”๋”ฉ(MLEC)๊ณผ ๋น„๊ณต์œ (shared-nothing) ์•„ํ‚คํ…์ฒ˜๋ฅผ ํ™œ์šฉํ•ด ํด๋Ÿฌ์Šคํ„ฐ ์ „์ฒด ์ฐจ์›์˜ ๋ณต์›๋ ฅ์„ ํ™•๋ณดํ•œ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ์—ฌ๋Ÿฌ ๊ตฌ์„ฑ ์š”์†Œ์—์„œ ๋™์‹œ์— ์žฅ์• ๊ฐ€ ๋ฐœ์ƒํ•˜๋”๋ผ๋„ ๊ฐ€๋™ ์‹œ๊ฐ„์„ ์•ˆ์ •์ ์œผ๋กœ ์œ ์ง€ํ•  ์ˆ˜ ์žˆ๋‹ค.

๊ธฐ์กด ์‹œ์Šคํ…œ ๋ฌธ์ œ๋กœ ์ธํ•œ ํŒŒ๊ธ‰ ํšจ๊ณผ๋Š” ์ƒ๋‹นํ•  ์ˆ˜ ์žˆ๋‹ค. ๊ฐ€ํŠธ๋„ˆ๋Š” โ€œ2026๋…„๊นŒ์ง€ AI์— ์ ํ•ฉํ•œ ๋ฐ์ดํ„ฐ๊ฐ€ ๋’ท๋ฐ›์นจ๋˜์ง€ ์•Š๋Š” AI ํ”„๋กœ์ ํŠธ์˜ 60%๋ฅผ ์กฐ์ง์ด ํฌ๊ธฐํ•˜๊ฒŒ ๋  ๊ฒƒโ€์ด๋ผ๊ณ  ๊ฒฝ๊ณ ํ–ˆ๋‹ค. ์‹ค์ œ๋กœ ๋ฐ์ดํ„ฐ ํ’ˆ์งˆ ์ €ํ•˜๋Š” ๊ธฐ์—…๋‹น ์—ฐ๊ฐ„ 1,290๋งŒ~1,500๋งŒ ๋‹ฌ๋Ÿฌ์˜ ์†์‹ค์„ ๋ฐœ์ƒ์‹œํ‚ค๊ณ  ์žˆ์œผ๋ฉฐ, ๋ฐ์ดํ„ฐ ํŒŒ์ดํ”„๋ผ์ธ ์žฅ์• ๋กœ ์ธํ•œ ์ธ์‚ฌ์ดํŠธ ์†์‹ค๊ณผ ์„œ๋น„์Šค ์ˆ˜์ค€ ๊ณ„์•ฝ(SLA) ๋ฏธ์ค€์ˆ˜ ๋น„์šฉ์€ ์‹œ๊ฐ„๋‹น ์•ฝ 30๋งŒ ๋‹ฌ๋Ÿฌ์— ์ด๋ฅด๊ณ  ์žˆ๋‹ค.

AI ์†๋„์— ๋งž๋Š” ์Šคํ† ๋ฆฌ์ง€

AI ์‹œ์Šคํ…œ์ด ์š”๊ตฌํ•˜๋Š” ์ˆ˜์ค€์˜ ์‹ ๋ขฐ์„ฑ์„ ํ™•๋ณดํ•˜๋ ค๋ฉด ๊ธฐ์ˆ ์ ยท์šด์˜์  ์•„ํ‚คํ…์ฒ˜ ์ „๋ฐ˜์„ ๋‹ค์‹œ ์„ค๊ณ„ํ•ด์•ผ ํ•œ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด ๋ณต์›๋ ฅ์€ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜ ๋ณ€ํ™”์— ๋”ฐ๋ผ ๊ธฐ์กด ์Šคํ† ๋ฆฌ์ง€ ์ œํ’ˆ์— ์‚ฌํ›„์ ์œผ๋กœ ๋ง๋ถ™์ด๋Š” ์š”์†Œ๊ฐ€ ์•„๋‹ˆ๋ผ, ์ฒ˜์Œ๋ถ€ํ„ฐ ๊ตฌ์กฐ ์•ˆ์— ๋‚ด์žฌ๋ผ์•ผ ํ•œ๋‹ค.

๊ธฐ์ˆ  ์ธก๋ฉด์—์„œ๋Š” ๋‹ค์ค‘ ๋‹จ๊ณ„ ์†Œ๊ฑฐ ์ฝ”๋”ฉ๊ณผ ๊ฐ™์€ ํ˜„๋Œ€์ ์ธ ๋ถ„์‚ฐ ๋ฐ์ดํ„ฐ ๋ณดํ˜ธ ๋ฐฉ์‹์ด ํ•ต์‹ฌ์œผ๋กœ ๋– ์˜ค๋ฅด๊ณ  ์žˆ๋‹ค. ์ด๋Š” ์ „ํ†ต์ ์ธ RAID์˜ ์ œํ•œ์ ์ธ ์žฅ์•  ๋Œ€์‘ ๋ฒ”์œ„๋ฅผ ๋„˜์–ด, ์—ฌ๋Ÿฌ ๋…ธ๋“œ์— ๊ฑธ์นœ ๋ณดํ˜ธ๋ฅผ ์ œ๊ณตํ•จ์œผ๋กœ์จ ๋‹ค์ˆ˜์˜ ๊ตฌ์„ฑ ์š”์†Œ๊ฐ€ ๋™์‹œ์— ์žฅ์• ๋ฅผ ์ผ์œผ์ผœ๋„ ๋ฐ์ดํ„ฐ ๋ฌด๊ฒฐ์„ฑ์„ ์œ ์ง€ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•œ๋‹ค.

๋™์‹œ์— ํ”Œ๋ž˜์‹œ์™€ ๋””์Šคํฌ๋ฅผ ๊ฒฐํ•ฉํ•œ ํ•˜์ด๋ธŒ๋ฆฌ๋“œ ์•„ํ‚คํ…์ฒ˜๋Š” ๋น„์šฉ ํšจ์œจ์„ฑ์„ ๋†’์ด๋Š” ๋ฐ ๊ธฐ์—ฌํ•œ๋‹ค. ๊ณ ์„ฑ๋Šฅ์ด ์š”๊ตฌ๋˜๋Š” ๋ฐ์ดํ„ฐ๋Š” ํ”Œ๋ž˜์‹œ์— ๋‘๊ณ , ์ƒ๋Œ€์ ์œผ๋กœ ์ค‘์š”๋„๊ฐ€ ๋‚ฎ์€ ์ •๋ณด๋Š” ๋น„์šฉ์ด ๋‚ฎ์€ ์ €์žฅ ๋งค์ฒด๋กœ ๊ณ„์ธตํ™”ํ•˜๋Š” ๋ฐฉ์‹์ด๋‹ค. ์—ฌ๊ธฐ์— ๋ชจ๋“ˆํ˜• ๋น„๊ณต์œ  ๊ตฌ์กฐ๋ฅผ ์ ์šฉํ•˜๋ฉด ๋‹จ์ผ ์žฅ์•  ์ง€์ ์„ ์ œ๊ฑฐํ•  ์ˆ˜ ์žˆ๊ณ , ์ „์šฉ ํ•˜๋“œ์›จ์–ด ์—†์ด ํ‘œ์ค€ ์„œ๋ฒ„ ๋…ธ๋“œ๋ฅผ ์ถ”๊ฐ€ํ•˜๋Š” ๊ฒƒ๋งŒ์œผ๋กœ ์„ฑ๋Šฅ ํ™•์žฅ์ด ๊ฐ€๋Šฅํ•ด์ง„๋‹ค.

์šด์˜ ์ธก๋ฉด์—์„œ๋„ ์ถฉ์กฑํ•ด์•ผ ํ•  ์š”๊ตฌ ์‚ฌํ•ญ์ด ์žˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด ์ž๋™ํ™”๋œ ๋ฐ์ดํ„ฐ ๋ฌด๊ฒฐ์„ฑ ๊ฒ€์‚ฌ๋Š” ๋ฐ์ดํ„ฐ ์†์ƒ์ด AI ํŒŒ์ดํ”„๋ผ์ธ์œผ๋กœ ์œ ์ž…๋˜๊ธฐ ์ „์— ์ด๋ฅผ ๊ฐ์ง€ํ•˜๊ณ  ๊ฒฉ๋ฆฌํ•  ์ˆ˜ ์žˆ๋‹ค. ๋˜ํ•œ ์ •๊ธฐ์ ์ธ ๋ณต๊ตฌ ํ›ˆ๋ จ์€ AI ์šด์˜ ํ™˜๊ฒฝ์ด ์š”๊ตฌํ•˜๋Š” ์ด‰๋ฐ•ํ•œ ์‹œ๊ฐ„ ๋‚ด์— ๋ณต์› ์ ˆ์ฐจ๊ฐ€ ์‹ค์ œ๋กœ ์ž‘๋™ํ•˜๋Š”์ง€๋ฅผ ๊ฒ€์ฆํ•˜๋Š” ์—ญํ• ์„ ํ•œ๋‹ค. ์ด๋Ÿฐ ๊ธฐ์ˆ ์ ยท์šด์˜์  ์š”์†Œ๋ฅผ ๊ฑฐ๋ฒ„๋„Œ์Šค์™€ ๊ทœ์ œ ์ค€์ˆ˜ ์ฒด๊ณ„์™€ ์ •๋ ฌํ•˜๋ฉด ๊ธฐ์ˆ ์  ์œ„ํ—˜๊ณผ ๊ทœ์ œ ๋ฆฌ์Šคํฌ๋ฅผ ๋™์‹œ์— ์ค„์ผ ์ˆ˜ ์žˆ๋‹ค.

๋ถ„๋ช…ํ•œ ์ ์€ ์ด๋Ÿฌํ•œ ์—ญ๋Ÿ‰์ด ๋” ์ด์ƒ ์„ ํƒ ์‚ฌํ•ญ์ด ์•„๋‹ˆ๋ผ๋Š” ๊ฒƒ์ด๋‹ค. ์ด์ œ๋Š” AI ์ธํ”„๋ผ๊ฐ€ ์„ค๊ณ„๋˜๋Š” ๋ฐฉ์‹์˜ ํ•ต์‹ฌ ์š”์†Œ๊ฐ€ ๋๋‹ค. ์•ž์œผ๋กœ AI ์›Œํฌ๋กœ๋“œ์™€ ๋ฐ์ดํ„ฐ์…‹์€ ๊ณ„์† ํ™•์žฅํ•  ์ˆ˜๋ฐ–์— ์—†์œผ๋ฉฐ, ์ด์— ๋”ฐ๋ผ ์Šคํ† ๋ฆฌ์ง€ ์•„ํ‚คํ…์ฒ˜ ์—ญ์‹œ ์ „์ฒด ๊ต์ฒด ์—†์ด๋„ ์šฉ๋Ÿ‰๊ณผ ์„ฑ๋Šฅ์„ ํ™•์žฅํ•  ์ˆ˜ ์žˆ๋„๋ก ๋ชจ๋“ˆํ˜•์ด๋ฉด์„œ ๋ฒค๋” ์ค‘๋ฆฝ์ ์ธ ๊ตฌ์กฐ๋กœ ์ง„ํ™”ํ•ด์•ผ ํ•œ๋‹ค.
dl-ciokorea@foundryco.com

How AI is reshaping the foundations of computing and storage

If Jensen Huang is right that the era of general-purpose computing is coming to an end, then we are witnessing a transformation as profound as the shift from horsepower to steam power two centuries ago.

At the heart of this new revolution are the converging developments across AI and data infrastructure, where unprecedented computational power is aligning (or at least attempting to) with an equally demanding need for speed, reliability and scale in how information is stored and accessed.

By creating the most data-intensive workloads ever seen, AI is radically reshaping enterprise infrastructure. The eye-watering sums being spent on expanding global datacenter capacity bear this out, with Metaโ€™s $600 billion plan among the most recent in a slew of announcements. Back in April this year, McKinsey put a $7 trillion price tag on what they thought would be required โ€œto keep pace with the demand for compute power.โ€ If the momentum behind AI continues unabated, that figure may need to be revised upwards.

The situation also has fundamental implications for data storage. Traditional storage was built for predictable, sequential workloads like databases and virtualization. AI upends that model, with thousands of GPU threads hammering existing systems with parallel, random, high-throughput access.

The performance problems this can create cascade across infrastructure components. When storage cannot keep up, GPUs sit idle, training cycles stall and overall costs soar. Every hour of underfed GPUs delays ROI because training is an investment and stalled or inefficient epochs push out time to value. The risks extend even further. If data is corrupted or lost, entire models often need to be retrained, creating enormous and unexpected costs. The impact goes beyond training inefficiency. Inference is the revenue-generating component, and slow or unstable data pipelines directly reduce the commercial return of AI applications. In response, legacy vendors are trying to retrofit existing architectures to meet AI demand, but despite their best efforts, most of these designs still limit performance and scalability.

Something has to give, starting with the recognition that AI requires purpose-built, natively high-performance storage systems.

Reliability 101

These performance pressures also expose a deeper problem โ€” reliability. Large-scale AI models rely on uninterrupted access to training data, and any disruption, whether itโ€™s a metadata server failure, data corruption or a myriad of other issues, can significantly impact productivity and compromise results.

Indeed, reliability in this context is not a single metric; itโ€™s the product of durability, availability and recoverability. These are crucial issues because the ability to maintain continuous operations and data integrity isnโ€™t just a technical safeguard; itโ€™s what determines whether AI investments actually deliver value.

The problem today is that many legacy systems still rely on local RAID or HA-pair architectures, which protect against small-scale failures but falter at AI scale. In contrast, modern designs utilize multi-level erasure coding and shared-nothing architectures to deliver cluster-wide resilience, ensuring sustained uptime even under multiple simultaneous failures.

The knock-on effect of legacy shortcomings is enormous, with Gartner warning that โ€œthrough 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data.โ€ If that wasnโ€™t bad enough, poor data quality already drains $12.9โ€“$15 million per enterprise annually, and pipeline failures cost around $300,000 per hour in lost insight and missed SLAs.

Storage at the speed of AI

Building the level of reliability AI systems need requires rethinking how systems are technologically and operationally architected. For instance, resilience must be embedded from the outset, rather than being retrofitted to legacy storage products as applications change around them.

At a technological level, capabilities such as multi-level erasure coding (MLEC), a modern distributed data protection mechanism, will replace traditional RAIDโ€™s limited fault tolerance with protection that spans multiple nodes, ensuring data remains intact even if several components fail simultaneously.

At the same time, hybrid flash-and-disk architectures help control cost by keeping high-performance data on flash while tiering less critical information to lower-cost media. Meanwhile, modular, shared-nothing designs eliminate single points of failure and allow performance to scale simply by adding standard server nodes with no proprietary hardware required.

Then there are operational requirements to address. For example, automated data integrity checks can detect and isolate corruption before it enters AI pipelines, while regular recovery drills ensure restoration processes work within the tight timeframes AI production demands. Aligning these technical and operational layers with governance and compliance frameworks minimizes both technical and regulatory risk.

Make no mistake, these capabilities are not just nice-to-haves; they are now fundamental to the way AI infrastructure should be designed. Inevitably, AI workloads and datasets will continue to expand, and storage architectures will need to be modular and vendor-neutral, allowing capacity and performance upgrades without wholesale replacement.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?

๊ณต๊ฐ„ยท์ „๋ ฅ ํ•œ๊ณ„ ๊ทน๋ณตํ•œ๋‹คยทยทยทDDR5 ๋žจ ๋Œ€์•ˆ์œผ๋กœ ๋– ์˜ค๋ฅธ โ€˜์†Œ์บ โ€™

์‚ผ์„ฑ์ „์ž๋Š” ์ง€๋‚œ๋‹ฌ AI ๋ฐ์ดํ„ฐ์„ผํ„ฐ ํ”Œ๋žซํผ์„ ์œ„ํ•ด ์„ค๊ณ„๋œ ์†Œ์บ 2 LPDDR5 ๊ธฐ๋ฐ˜ ๋ฉ”๋ชจ๋ฆฌ ๋ชจ๋“ˆ์„ ๊ณต๊ฐœํ–ˆ๋‹ค.

์†Œ์บ 2๋Š” ๊ธฐ์กด ๋Œ€๋น„ ์„ฑ๋Šฅ์„ ๋Œ์–ด์˜ฌ๋ฆฐ ์ƒˆ๋กœ์šด ๋ฉ”๋ชจ๋ฆฌ ํผํŒฉํ„ฐ๋‹ค. ํ•ด๋‹น ๋ชจ๋“ˆ์ธ CAMM(Compression Attached Memory Module)์€ ๋ธ์ด ๋…ธํŠธ๋ถ์šฉ ๋ฉ”๋ชจ๋ฆฌ ๊ธฐ์ˆ ๋กœ ์ฒ˜์Œ ๊ฐœ๋ฐœํ–ˆ์œผ๋ฉฐ, ์ดํ›„ ์—…๊ณ„ ์ „๋ฐ˜์˜ ์ฑ„ํƒ์„ ์ด‰์ง„ํ•˜๊ธฐ ์œ„ํ•ด ํ‘œ์ค€ํ™” ๊ธฐ๊ตฌ์— ์ด๊ด€๋๋‹ค. CAMM2๋Š” ์ด ๊ธฐ์ˆ ์„ ์‚ฐ์—… ํ‘œ์ค€์œผ๋กœ ๊ฐœ๋ฐœํ•œ ์ฒซ ๋ฒˆ์งธ ์„ธ๋Œ€๋‹ค.

์†Œ์บ 2์˜ ํŠน์ง•์€ ์Šค๋งˆํŠธํฐ๊ณผ ํƒœ๋ธ”๋ฆฟ์— ์‚ฌ์šฉ๋˜๋Š” LPDDR5 ๋ฉ”๋ชจ๋ฆฌ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ๋‹ค๋Š” ์ ์ด๋‹ค. ์ด๋ฅผ ํ†ตํ•ด DDR ๋ฉ”๋ชจ๋ฆฌ ์ˆ˜์ค€์˜ ๊ณ ์„ฑ๋Šฅยท๊ณ ๋Œ€์—ญํญ์„ ์ œ๊ณตํ•˜๋ฉด์„œ๋„ ์ „๋ ฅ ์†Œ๋ชจ๋ฅผ ๋‚ฎ์ถœ ์ˆ˜ ์žˆ๋‹ค.

์‚ผ์„ฑ์ „์ž๋Š” ์†Œ์บ 2๊ฐ€ ์„œ๋ฒ„์— ์‚ฌ์šฉ๋˜๋Š” ํ‘œ์ค€ DDR5 RDIMM ๋Œ€๋น„ 2๋ฐฐ์˜ ๋Œ€์—ญํญ์„ ์ œ๊ณตํ•˜๋ฉด์„œ๋„ ์†Œ๋น„ ์ „๋ ฅ์€ ๋” ๋‚ฎ๋‹ค๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ๋‹ค๋ฅธ ๋ถ„์„์— ๋”ฐ๋ฅด๋ฉด ์†Œ์บ 2๋Š” ํ‘œ์ค€ DDR5 ๋ฉ”๋ชจ๋ฆฌ ๋Œ€๋น„ 1.5๋ฐฐ์—์„œ 2๋ฐฐ ์ˆ˜์ค€์˜ ์„ฑ๋Šฅ์„ ๋‚ด๋ฉด์„œ๋„ ์ „๋ ฅ ์†Œ๋ชจ๋Š” ์•ฝ 55% ์ˆ˜์ค€์ธ ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ๋‹ค.

์†Œ์บ 2 ๋ชจ๋“ˆ์€ ํ•˜๋‚˜์˜ ์นฉ์— ์—ฌ๋Ÿฌ ์ธต์˜ ๋ฉ”๋ชจ๋ฆฌ๋ฅผ ์Œ“๋Š” ์ ์ธต ๊ธฐ์ˆ ์„ ์ ์šฉํ•ด ์ง‘์ ๋„๋ฅผ ๋†’์˜€๋‹ค. ์ด๋กœ ์ธํ•ด ๊ธฐ์กด DDR5 ๋ฉ”๋ชจ๋ฆฌ ์Šคํ‹ฑ๋ณด๋‹ค ํฌ๊ธฐ๊ฐ€ ์ž‘์œผ๋ฉฐ, ๋™์ผํ•œ ์šฉ๋Ÿ‰์˜ DRAM์„ ๊ธฐ์ค€์œผ๋กœ ํ•  ๋•Œ ๋ฉ”์ธ๋ณด๋“œ์—์„œ ์ฐจ์ง€ํ•˜๋Š” ๊ณต๊ฐ„๋„ ์ค„์–ด๋“ ๋‹ค. ์†Œ์บ 2๋Š” DDR ๋ฉ”๋ชจ๋ฆฌ์™€ ํ•จ๊ป˜ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ ์‹œ์Šคํ…œ ๋ฉ”๋ชจ๋ฆฌ๋กœ ๋‹จ๋… ๊ตฌ์„ฑํ•˜๋Š” ๊ฒƒ๋„ ๊ฐ€๋Šฅํ•˜๋‹ค.

๋ธ์€ CAMM ๋ฉ”๋ชจ๋ฆฌ ์‚ฌ์–‘์„ ๊ณต๋™ ์„ค๊ณ„ํ•˜๊ธฐ ์œ„ํ•ด ์—ฌ๋Ÿฌ ํŒŒํŠธ๋„ˆ๋ฅผ ์ฐธ์—ฌ์‹œํ‚จ ๋’ค ์ด๋ฅผ JEDEC ํ‘œ์ค€ํ™” ๊ธฐ๊ตฌ์— ์ด๊ด€ํ–ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์†Œ์บ 2๋Š” ํŠน์ • ๋ฒค๋”์˜ ๋…์ž์ ์ธ ํ•ด๋ฒ•์ด ์•„๋‹Œ ์—…๊ณ„ ํ‘œ์ค€์œผ๋กœ ์ธ์‹๋˜๊ณ  ์žˆ๋‹ค. JEDEC์€ ์ดํ›„ CAMM ์‚ฌ์–‘์— ECC๋ฅผ ๋น„๋กฏํ•œ ๋‹ค์–‘ํ•œ ์˜ค๋ฅ˜ ์ •์ • ๊ธฐ๋Šฅ์„ ์ถ”๊ฐ€ํ•˜๋ฉฐ, ์—”ํ„ฐํ”„๋ผ์ด์ฆˆ ํ™˜๊ฒฝ์„ ๊ณ ๋ คํ•œ ์‚ฌ์–‘์œผ๋กœ ๋ณด์™„ํ–ˆ๋‹ค.

์˜ค๋ธŒ์ ํ‹ฐ๋ธŒ ์• ๋„๋ฆฌ์‹œ์Šค์˜ ๋Œ€ํ‘œ ์ง ํ•ธ๋””๋Š” ์†Œ์บ ์ด ๊ธฐ์กด ํ•˜๋“œ์›จ์–ด๋ฅผ ๋‹จ์ˆœํžˆ ์žฌํฌ์žฅํ•œ ๊ธฐ์ˆ ์ด๋‚˜ ์–ต์ง€ ๋ฌธ์ œ๋ฅผ ๋งŒ๋“ค์–ด๋‚ธ ํ•ด๋ฒ•์€ ์•„๋‹ˆ๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค. ๊ทธ๋Š” ์—…๊ณ„๊ฐ€ ์‹ค์ œ๋กœ ์ง๋ฉดํ•œ ๊ณผ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•œ ๊ธฐ์ˆ ์ด๋ฉฐ, ํ•„์š”์„ฑ๊ณผ ์ˆ˜์š”๊ฐ€ ๋ถ„๋ช…ํ•˜๋‹ค๊ณ  ํ‰๊ฐ€ํ–ˆ๋‹ค.

ํ•ธ๋””๋Š” โ€œ์„œ๋ฒ„ ํ”„๋กœ์„ธ์„œ ์ œ์กฐ์‚ฌ์™€ ์—”๋น„๋””์•„๊ฐ€ ์†Œ์บ ์„ ์ ๊ทน์ ์œผ๋กœ ์ง€์ง€ํ•˜๋Š” ์ด์œ ๋Š” ๋” ๋น ๋ฅธ ์ธํ„ฐํŽ˜์ด์Šค๋ฅผ ํ™•๋ณดํ•  ์ˆ˜ ์žˆ๊ณ , ์ƒ๋Œ€์ ์œผ๋กœ ๋‚ฎ์€ ์ „๋ ฅ ์†Œ๋ชจ๋กœ ์ข์€ ๊ณต๊ฐ„์— ๋งŽ์€ ๋ฉ”๋ชจ๋ฆฌ๋ฅผ ์ง‘์ ํ•  ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธโ€์ด๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

์†Œ์บ 2๊ฐ€ ์ ์ธต ๋ฉ”๋ชจ๋ฆฌ๋ฅผ ์‚ฌ์šฉํ•œ๋‹ค๋Š” ์ ์ด ์ œ์กฐ ๋น„์šฉ์„ ๋†’์ผ ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ด€์ธก๋„ ๋‚˜์˜จ๋‹ค. ํ•˜์ง€๋งŒ ํ•ธ๋””๋Š” ์‹ค์ œ๋กœ๋Š” ๊ทธ๋ ‡์ง€ ์•Š๋‹ค๊ณ  ๋ดค๋‹ค. ๊ทธ๋Š” โ€œ๋ฉ”๋ชจ๋ฆฌ ๋ฒค๋”๋“ค์€ ๊ธฐ์กด DRAM๊ณผ ๋™์ผํ•œ ๊ฐ€๊ฒฉ๋Œ€๋กœ ๋‹ค์–‘ํ•œ ์ ์ธต ๊ตฌ์„ฑ์„ ํŒ๋งคํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ํŒจํ‚ค์ง• ๊ธฐ์ˆ  ์—ญ์‹œ ๋‚ธ๋“œ ํ”Œ๋ž˜์‹œ ์—…์ฒด๊ฐ€ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ์‹๊ณผ ๊ฐ™๋‹ค. ๋ˆˆ์— ๋„๋Š” ๊ฐ€๊ฒฉ ์ฐจ์ด๊ฐ€ ๋ฐœ์ƒํ•  ๊ฐ€๋Šฅ์„ฑ์€ ํฌ์ง€ ์•Š๋‹คโ€๋ผ๊ณ  ์„ค๋ช…ํ–ˆ๋‹ค.

SKํ•˜์ด๋‹‰์Šค๋„ ์†Œ์บ 2 ๋ฉ”๋ชจ๋ฆฌ๋ฅผ ์ง€์›ํ•  ๊ณ„ํš์„ ๋ฐœํ‘œํ–ˆ์ง€๋งŒ, ๊ตฌ์ฒด์ ์ธ ์ถœ์‹œ ์ผ์ •์€ ๊ณต๊ฐœํ•˜์ง€ ์•Š์•˜๋‹ค. ์—…๊ณ„์—์„œ๋Š” ๋งˆ์ดํฌ๋ก ๊ณผ ์‚ผ์„ฑ์ „์ž์— ๋น„ํ•ด ๊ฐœ๋ฐœ ์ผ์ •์ด ๋‹ค์†Œ ๋’ค์ฒ˜์ง„ ๊ฒƒ์œผ๋กœ ๋ณด๊ณ  ์žˆ๋‹ค. ์†Œ์บ 2๋Š” ์—”๋น„๋””์•„๊ฐ€ ๋ฒ ๋ผ ๋ฃจ๋นˆ ํ”Œ๋žซํผ์„ ์ถœ์‹œํ•˜๋Š” ์‹œ์ ์ธ 2026๋…„ 2๋ถ„๊ธฐ ์ „ํ›„๋กœ ์ƒ์šฉํ™”๋  ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒ๋˜๊ณ  ์žˆ๋‹ค.
dl-ciokorea@foundryco.com

โŒ