Infrastructure, Not Models, as the Competitive Moat
Jensen Huang has built NVIDIA into one of the world's most valuable companies by thinking in systems rather than individual products. Under his long-standing leadership, the company has grown far beyond being a chip manufacturer into something far more strategic: a platform company that forces organisations to think about their entire AI stack as an integrated whole. While many observers remain fixated on Large Language Models (LLMs) and chatbots—often mistaking them for the entirety of artificial intelligence—there is a pronounced lack of focus on the vital infrastructure and foundational layers required for this ecosystem to genuinely flourish and deliver measurable business value.
Huang's 5-layer cake hypothesis outlines everything required for organisations and nations to compete effectively in the AI era. While I will not attend NVIDIA GTC 2026 in person, I am following the event closely through multiple channels. In this post, I want to unpack the 5-layer cake framework and explain why it matters for your organisation's AI strategy—particularly if you sit in infrastructure, operations, or strategic planning.
The Prophecy: From Software as a Cost Centre to Intelligence as Infrastructure
When Jensen Huang speaks about the AI 5-Layer Cake, he is delivering more than a technical roadmap or a sales pitch. He is articulating a structural shift in how organisations should think about technology investment and competitive positioning. For decades, Information Technology was treated as a cost centre—a collection of tools to help humans work faster, reduce errors, and manage operations. Huang's framework forces a fundamental pivot in that logic.
AI is not a tool in the traditional sense. It is a utility—akin to electricity or water supply. Much like the historical transition from localized steam engines scattered across factories to a centralized electrical grid that powered entire regions and industries, we are now moving from standalone software programs to what Huang calls intelligence factories. This shift brings a level of clarity and focus that eliminates the noise of individual model benchmarks, vendor positioning, and point solutions. Instead, it concentrates attention on the structural integrity of the entire system—every layer that supports the application layer where business value actually gets realised.
The prophecy Huang is articulating suggests that within the next five years, the competitive moat for an organisation will not be the specific Large Language Model (LLM) they have licensed or built in-house. Instead, competitive advantage will accrue to organisations that have built and optimised an efficient AI Factory—an end-to-end system capable of processing data at scale, refining models continuously, and deploying applications that solve real business problems. If your organisation lacks the energy infrastructure, the physical networking, the compute capacity, or the operational discipline to process data at industrial scale, even the most sophisticated model becomes a stranded asset—expensive to run and unable to deliver ROI at the speed the business demands.
The framework identifies five essential layers that must work in harmony, each dependent on the layer beneath it:
- Energy: The raw fuel of the new economy. Every chip, every data centre, every model inference consumes power. Without abundant, reliable, affordable energy, the layers above cannot scale.
- Chips: The specialised engines—primarily GPUs and increasingly custom silicon—that have replaced general-purpose processors as the engines of AI computation. This is where NVIDIA lives, and where the US currently maintains a multi-generation lead over competitors.
- Infrastructure: The high-speed networking, data centre design, cooling systems, and physical buildout that allows thousands of chips to act as a single coherent computing unit. This layer also includes the construction velocity and permitting frameworks that determine how quickly you can deploy new capacity.
- Models: The logic layer that translates raw data into reasoning, prediction, and generation. This is where most vendor marketing and analyst attention sits, but it is far from the only—or even the most constraining—layer.
- Applications: The interface and integration layer where AI agents and systems execute actual business processes. This is where economic value is realised: optimising supply chains, improving clinical diagnosis, automating financial operations, and enhancing customer experience.
The critical insight—and the one that separates clear thinking from hype—is found in the interdependence and constraint propagation across these layers. A bottleneck at the energy or networking layer renders breakthroughs in the model layer irrelevant. You cannot run a sophisticated model if you do not have the power to run it or the network bandwidth to feed it data. Conversely, brilliant applications cannot be built if the model layer is weak or the infrastructure layer is fragile. For leadership teams and technology strategists, this means that AI strategy must now encompass physical constraints—power generation and distribution, cooling, silicon manufacturing capacity, construction permitting—as much as it does software architecture, data strategy, and algorithmic innovation.
The Shashi.co Perspective: Beyond the Chatbot Horizon
At Shashi.co and through my work at Info-Tech Research Group, we examine how these exponential technology trends collide with organisational friction, legacy systems, and the real constraints of capital allocation. The 5-Layer Cake framework provides essential clarity: we are exiting the experimental phase of AI. When we view AI as infrastructure—functionally equivalent to the internet or the power grid—the strategic focus must shift from ROI on a single use case (a narrow chatbot pilot, a narrow process automation) to the Total Cost of Ownership (TCO) of an intelligence pipeline and the time-to-value across your entire operating model.
This is not theoretical. Organisations that have committed significant capital to AI—whether through in-house model training, third-party model licensing, or inference at scale—are already bumping against energy and infrastructure limits. Data centres are power-constrained. Permitting for new facilities takes years. GPU allocation is still contended. The bottlenecks have moved from "do we have the right algorithm?" to "do we have the physical capacity to run it profitably?"
The real strategic story at the application layer is the rise of Agentic Workflows—autonomous AI systems that operate with minimal human intervention and execute multi-step business processes end-to-end. We are moving toward a future where AI Agents are the primary consumers of the layers beneath them. These agents will not merely answer questions or classify data. They will manage supply chains in real time, optimise energy consumption across facilities, conduct research autonomously, interact with enterprise systems, and make recommendations with business impact. The Prophecy is that the distinction between digital operations and physical operations will progressively dissolve as intelligence becomes a manufactured output of the corporate technology stack, just as electricity is a manufactured utility. The factory floor, the hospital, the financial trading desk, and the customer service centre will all be run in collaboration with autonomous AI agents.
For practitioners at Info-Tech's client base—CIOs, CTOs, and enterprise architects—the implication is direct: your AI investment roadmap must address all five layers, not just the models and applications. If you are evaluating an AI platform or building in-house capabilities, allocate planning and budget across energy resilience, infrastructure modernisation, and organisational readiness for agentic automation, not just model selection.
The 5-Layer Cake in Practice: Real Companies Across the Stack
To make this framework concrete, consider how different organisations across the vendor ecosystem are positioning themselves within Huang's layers:
Energy Layer: This is where policy, regulation, and capital intensity collide. Lumen Technologies, which I covered during their investor day positioning, is increasingly pivoting toward physical AI infrastructure—data centre power delivery and network backbone capacity. The company's edge computing and wavelength services are explicitly designed to address the energy and connectivity bottlenecks that constrain enterprise AI deployment. Similarly, renewable energy companies and grid operators are becoming strategic infrastructure providers in this layer.
Chips Layer: NVIDIA obviously dominates here, but the competition is intensifying. The broader ecosystem includes semiconductor manufacturing partners and architectural innovators. This layer also includes companies like Pinecone, which I profiled as the vector database leader, because vector databases are increasingly specialised silicon optimisation problems as organisations scale AI inference.
Infrastructure Layer: This encompasses cloud providers (AWS, Azure, Google Cloud), data centre operators, and networking specialists. Databricks, which I covered extensively, sits at the bridge between infrastructure and models—their Lakebase approach is fundamentally about building the data pipeline and governance that allows organisations to feed models continuously. Zscaler, which I analysed during their earnings, operates in the security posture of this layer—ensuring that the infrastructure that moves data at scale remains secure and compliant.
Models Layer: This is where most vendor attention sits. IBM's AI strategy, which I examined at length, positions the company as both a model provider (through partnerships and acquisitions) and a trust and governance layer on top of models. The reality is that at this layer, differentiation is increasingly difficult; the gap between frontier models narrows month by month. The edge goes to organisations that can fine-tune, specialise, and integrate models into business workflows faster than competitors.
Applications Layer: This is where the real economic value is realised, and where agentic workflows are emerging. Workday and Five9, which I covered in earnings analysis, exemplify different points in this layer—Workday as an enterprise application platform that is integrating AI into human capital, financial, and supply-chain processes; Five9 as a customer engagement platform where AI agents are beginning to handle interactions autonomously. Creatio, another vendor I profiled, is positioned as a workflow and process automation platform where business users can build agentic applications. Infobip's AgentOS, which I covered, is explicitly building the orchestration layer for AI agents to operate across communications and enterprise systems.
The insight is not that any single vendor owns a layer. It is that your organisation needs visibility and strategy across all five layers. You cannot optimise one layer in isolation. A brilliant application built on Workday or Creatio becomes a stranded asset if you cannot provision the compute infrastructure or the energy to run it. A frontier model from Anthropic or OpenAI becomes an expensive toy if your data pipeline (Databricks), your infrastructure readiness (Zscaler), and your physical capacity constraints are not addressed first.
Huang's 5-layer framework is not a speculative take or a vendor sales argument masquerading as strategy. It is a description of structural reality that is increasingly difficult for organisations and policymakers to ignore. It will likely shape AI investment priorities, infrastructure policy, and competitive positioning for the next 5–7 years.
Key Takeaways for Your Organisation
- Evaluate your AI strategy as a full-stack problem. Do not optimise for the model layer in isolation. Assess energy resilience, infrastructure capacity, and data pipeline maturity across your entire estate.
- Understand your constraint layers. Identify which layers are most likely to limit your AI ambitions in the next 18–24 months. For most enterprises, it is energy and infrastructure, not models.
- Plan for agentic workflows. Begin exploratory work on how autonomous AI agents will operate within your business processes. This is not hype; this is where the application layer is heading.
- Engage on infrastructure policy. If your organisation depends on data-centre capacity, power allocation, or semiconductor supply, monitor and engage with policy discussions on energy, construction permitting, and supply-chain resilience. These will shape your competitive position as much as model quality will.
Sources and References
NVIDIA. "'Largest Infrastructure Buildout in Human History': Jensen Huang on AI's 'Five-Layer Cake' at Davos." NVIDIA Blog, January 2026. https://blogs.nvidia.com/blog/davos-wef-blackrock-ceo-larry-fink-jensen-huang/
Huang, Jensen. "Securing American Leadership on AI: Remarks on the Five-Layer Cake." Center for Strategic and International Studies (CSIS), December 2025. https://www.csis.org/analysis/nvidias-jensen-huang-securing-american-leadership-ai
Quartz. "Jensen Huang Brings a 5-Layer AI Pitch to Davos." Quartz, January 2026. https://qz.com/jensen-huang-nvidia-speech-davos-2026
Your Story / YourStory. "NVIDIA CEO Jensen Huang on AI's 5 Layers: Energy, Chips, Infrastructure, Models, Applications." Your Story, February 2026. https://yourstory.com/2026/02/jensen-huang-ai-five-layer-stack-energy-chips-infrastructure
