Building the AI Infrastructure Stack: A $3,000 Allocation for the Compute S-Curve

The investment story for AI is no longer just about the chips themselves. It has stepped down the stack, where the real exponential growth is now being built. The paradigm shift is clear: the bottleneck has moved from the GPU to the foundational infrastructure that feeds and connects them. This is the fundamental rails for the next computing paradigm, and the spending narrative reflects it. In 2025, the winners were memory and optics, as bottlenecks became the beneficiaries. The market is pricing in what's coming, with the expansion of the Nvidia cycle extending to its suppliers.

This sets the stage for a multi-year S-curve. Data center operators are projected to spend more than $500 billion on capital expenditures in 2026. That's a massive, near-term inflection point. But the real curve is longer. Research from McKinsey shows that data centers may require $7 trillion to meet demand for compute power by 2030. That's not just a forecast; it's a map of the exponential adoption path. The infrastructure layer-power, cooling, networking, and the specialized memory and optics that connect it-is where the growth is scaling.

For investors, this means looking beyond the GPU leaders to the companies building the rails. The story is about who can supply the critical bottlenecks as demand explodes. It's about the companies that will be paid for the watts, the bandwidth, and the memory that keep the AI engines running. This is the infrastructure layer where the paradigm shift is most visible, and where the exponential growth curve is just beginning its steep climb.

Assessing the Core Infrastructure Players: AMD, NVDA, and AVGO

The exponential growth of AI infrastructure is being built by a distinct set of players, each occupying a different rung on the adoption S-curve. Their growth profiles are defined by dominance, strategic partnership, and custom-built expansion. Let's break down the three key companies shaping the rails.

Nvidia remains the undisputed leader at the peak of the curve. Its Q2 2026 revenue surged 56% to $46.74 billion, a record driven by AI infrastructure demand. This isn't just a growth story; it's a dominance story. The company is projecting at least 42% revenue growth over the next year, with CEO Jensen Huang outlining a vision to scale into a $3 trillion to $4 trillion AI infrastructure opportunity over five years. That ambition is backed by a massive $60 billion buyback program, a clear signal of confidence in its own trajectory. Nvidia's position is one of first-mover advantage and architectural leadership, making it the primary beneficiary of the current AI buildout.

AMD is carving out its own powerful niche through deep partnership and custom growth. Its landmark deal with OpenAI is a multi-year commitment to power next-generation AI infrastructure, with OpenAI agreeing to deploy up to 6 gigawatts of AMD GPUs. This isn't a one-off sale; it's a multi-generational agreement that includes warrants for up to 160 million shares, tying AMD's future directly to the scale of OpenAI's deployments. The first phase begins in late 2026, with the total project valued at well above $100 billion. This partnership positions AMD as a critical strategic compute partner, allowing it to scale alongside a major AI hypercaler and capture value from the energy-intensive compute required.

Broadcom represents a different kind of wild card, focused on the networking and integration layer. The company expects 51% growth for fiscal year 2026, a staggering pace fueled by AI. Its unique positioning comes from hypercalers partnering to spec their own computing units, making Broadcom a key enabler in the custom silicon supply chain. While its growth is less visible than Nvidia's or AMD's partnership headlines, its trajectory is equally steep, riding the wave of data center expansion and the need for high-bandwidth connectivity.

Together, these three companies illustrate the layered nature of the AI infrastructure stack. Nvidia leads the compute charge, AMD secures massive, long-term partnerships, and Broadcom provides the essential connectivity. Their combined growth profiles map directly onto the multi-year S-curve of AI adoption, each playing a vital role in building the fundamental rails.

Allocation Strategy: Weighing Dominance, Partnership, and Custom Growth

The exponential growth of AI infrastructure is a multi-year S-curve, and the allocation strategy must reflect the different rungs of that climb. This $3,000 plan is a bet on three distinct growth engines: dominance, partnership, and custom integration. Each company's financial profile and strategic position map directly to a phase in the adoption curve.

First, Nvidia represents the dominant peak. Its Q2 2026 revenue surged 56% to $46.74 billion, a record that underscores its unmatched execution and market leadership. The company's ambitious projection for at least 42% growth over the next year, backed by a massive buyback, signals confidence in its architectural moat. This is the foundational compute layer, and its proven ability to scale is worth a commanding stake. Therefore, 40% ($1,200) is allocated to Nvidia. This is the anchor position, providing exposure to the core of the AI stack.

Next, Advanced Micro Devices embodies the high-growth partnership rung. Its landmark deal with OpenAI is a multi-year commitment to power next-generation infrastructure, with OpenAI agreeing to deploy up to 6 gigawatts of AMD GPUs. This isn't a one-time sale; it's a multi-generational agreement that includes warrants for up to 160 million shares, directly tying AMD's future to OpenAI's scale. The first phase begins in late 2026, with the total project valued at well above $100 billion. This partnership provides a clear, high-visibility growth trajectory. AMD is also projecting revenue growth at a 35% compound annual rate over the next few years. This strategic alignment justifies a significant allocation. Therefore, 30% ($900) is allocated to AMD.

Finally, Broadcom represents the aggressive custom growth layer. The company expects 51% growth for fiscal year 2026, a staggering pace fueled by AI. Its unique positioning comes from hypercalers partnering to spec their own computing units, making Broadcom a key enabler in the custom silicon supply chain. While its growth is less visible than Nvidia's or AMD's partnership headlines, its trajectory is equally steep, riding the wave of data center expansion and the need for high-bandwidth connectivity. This is the integration and networking layer where the custom build-out accelerates. Therefore, 30% ($900) is allocated to Broadcom.

This allocation is a deliberate spread across the S-curve. Nvidia captures the dominant, proven leader. AMD captures the high-growth, partnership-driven phase. Broadcom captures the aggressive, custom integration phase. Together, they provide a diversified bet on the fundamental rails of the AI infrastructure stack, each playing a vital role in building the exponential adoption path.

Catalysts, Scenarios, and Key Risks

The investment thesis for AI infrastructure is a bet on an exponential adoption S-curve. To monitor this thesis, we need to track the leading indicators that could accelerate it and the risks that could flatten it. The framework is simple: watch for announcements of new data center partnerships and capex plans from major tech companies as the clearest signals of demand. The key risk is a slowdown in AI spending acceleration, which would directly pressure the valuations of infrastructure providers.

The near-term catalysts are already materializing. The landmark deal between AMD and OpenAI is a prime example, with OpenAI agreeing to deploy up to 6 gigawatts of AMD GPUs and the total project valued at well above $100 billion. This isn't just a contract; it's a multi-year commitment that sets a benchmark for future partnerships. Similarly, Hut 8's $7 billion deal with Anthropic for data center capacity is a direct play on the watts needed for AI. These are leading indicators. When other major tech companies follow with similar capex plans and partnerships, it confirms the demand is broadening beyond a few leaders, accelerating the S-curve.

The primary risk to this thesis is a deceleration in the pace of AI investment. The market is pricing in a multi-year buildout, but if spending growth slows, the entire infrastructure stack faces pressure. This could flatten the adoption curve and lead to a reassessment of valuations for companies whose growth is tied to future capex. The evidence shows the cycle is still in its early, steep phase, but the risk is that it peaks sooner than expected.

A secondary, but critical, signal to monitor is semiconductor equipment (WFE) spending. While WFE was "tepid" in 2025, the stock market performance of suppliers like memory and optics showed the market was pricing in future demand. A sustained increase in WFE orders would signal that the broader industry cycle is shifting, confirming that the AI buildout is not just a stock market story but a real capital expenditure surge. Conversely, a slowdown in WFE would be an early warning of a broader industry contraction.

The bottom line is that this is a monitoring framework, not a static forecast. The exponential growth story depends on continuous validation of demand through new partnerships and capex announcements. The key risk is that the spending acceleration falters. For now, the evidence points to a powerful, multi-year inflection, but the path will be tracked by the announcements that confirm it.