AI doesn't follow linear progress-it accelerates along an exponential S-curve, and the layers at the foundation capture disproportionate value. This is the core thesis for AI infrastructure stocks, and recent market dynamics have created a rare entry point.
The evidence is clear: AI development is accelerating at rates that defy traditional tech timelines. BlackRock's Tony Kim, after his annual tech tour of 25 companies across Silicon Valley, observed that change at the chip level is coming at 2x a year, while the model intelligence layer sees 10x annual improvement. This isn't incremental progress-it's exponential scaling that compounds rapidly as the technology moves up the stack. The infrastructure layer-chips, networking, and cloud framework-forms the foundation upon which everything else depends. Without it, the entire AI edifice collapses.
Yet here's the opportunity: the 2026 "anything-but-AI" selloff has left well-capitalized AI infrastructure stocks trading at meaningful discounts. After strong gains in 2025, sentiment shifted dramatically, creating a window where quality infrastructure names are available below their intrinsic value. This isn't a temporary dip-it's a structural repositioning by investors who failed to recognize that the AI adoption curve is still in its early innings.
Nvidia exemplifies both the magnitude of the opportunity and the patience required. The stock's 72.21% rolling annual return demonstrates the sheer scale of compute demand, yet the story remains far from complete. With a market cap of $4.775 trillion and a wide economic moat built on GPU leadership, networking solutions, and software infrastructure, Nvidia sits at the center of the AI supply chain. The stock currently trades at a 32% discount to Morningstar's fair value estimate-a rare valuation gap for a company capturing exponential demand.
The key insight: infrastructure wins because it's the necessary condition for every AI application. Whether AI moves toward superintelligence, edge deployment, or physical-world integration, the compute layer scales with it. The S-curve is just beginning.
Nvidia: The Compute Backbone Still Scaling
Nvidia's 72.21% rolling annual return isn't a bubble-it's the market pricing in exponential demand for AI compute that has further to run. The stock's 52-week range of $95-$212 captures the volatility of a sector in flux, but the current price near $196 signals investors are reasserting conviction after the 2026 selloff. What drives this? The infrastructure buildout is still in its early innings.
The demand signal is unambiguous: enterprise adoption of AI remains robust, with companies spending aggressively to deploy AI-driven productivity gains. Anthropic-a major AI player-has seen its annual revenue run rate balloon to $30 billion from $9 billion in just three months, and it relies on Nvidia's GPUs alongside Google's TPUs to train models and run inference. This isn't niche experimentation; it's enterprise-scale deployment. Anthropic now counts more than 1,000 businesses spending over $1 million annually on its offerings. Every dollar of that infrastructure spend flows through to chip suppliers, and Nvidia sits at the center of the supply chain.
Nvidia's wide economic moat-built on GPU leadership, networking solutions, and software infrastructure-means it captures disproportionate value as the AI S-curve ascends. The stock trades at a 32% discount to Morningstar's $260 fair value estimate, a rare valuation gap for a company that is the necessary condition for every AI application. With a market cap of $4.775 trillion and a PE TTM of 39.77, the multiple isn't cheap-but neither is the growth trajectory. The key insight: infrastructure wins because AI cannot scale without compute. The S-curve is just beginning, and Nvidia is positioned to scale with it.
Microsoft: The Cloud/AI Platform King
If Nvidia owns the compute layer, Microsoft owns the platform where that compute comes alive. The Azure-OpenAI partnership positions MSFT at the unique intersection of cloud infrastructure and advanced AI models-a dual advantage that few companies can match.
Azure is the centerpiece of the new Microsoft, a $75 billion business still growing at approximately 30% annually. That growth rate isn't slowing-it's accelerating as enterprise AI adoption moves from experimentation to deployment. Cloud providers are introducing new services and scaling capacity, capturing the enterprise AI adoption wave. Microsoft is uniquely positioned to win this round.
The secret sauce is hybrid cloud. Azure offers customers a painless way to experiment and move select workloads to the cloud, creating seamless hybrid environments. Since existing customers remain in the same Microsoft ecosystem, applications and data move easily from on-premises to cloud. This isn't just convenience-it's a switching cost fortress. Once enterprises embed their workflows into Azure, the friction of leaving becomes prohibitive. The platform becomes infrastructure, and infrastructure is sticky.
Add the OpenAI partnership to this foundation, and Microsoft gains something competitors cannot easily replicate: direct access to frontier model capabilities that can be embedded into Azure services. This isn't a passive investment-it's a strategic integration that makes Azure more valuable as AI capabilities advance. The S-curve works in Microsoft's favor because every layer of AI development-from model training to inference to application deployment-flows through cloud infrastructure, and Microsoft controls a critical chokepoint.
The market sees the opportunity. Microsoft carries a 5-star Morningstar Rating and a wide economic moat. Yet the stock trades at a 38% discount to our $600 fair value estimate-the same sentiment-driven gap that created the entry point for Nvidia. For investors building a portfolio around exponential AI adoption, Microsoft isn't just a holding-it's a platform play that captures value across the entire AI stack.
Broadcom: The Connectivity Play Behind AI's Exponential Curve
If Nvidia owns the compute layer and Microsoft owns the platform, Broadcom owns the connectivity that makes both possible. As AI scales from experimentation to enterprise deployment, the infrastructure that ties distributed systems together becomes not just important-it becomes the bottleneck. Broadcom sits at that bottleneck.
The company designs custom chips and networking infrastructure for AI data centers, positioning it at the heart of the AI supply chain. While much of the market focuses on GPUs, the reality is that training and inference at scale require thousands of chips working in concert. The networking fabric that connects them is where Broadcom has built its dominance. This isn't peripheral work-it's the central nervous system of AI computation.
Anthropic's explosive growth provides the clearest signal of continued infrastructure demand. The company's annual revenue run rate ballooned to $30 billion from $9 billion in just three months, and that expansion relies on hardware partners to enable its LLMs to do the heavy lifting. Anthropic trains and runs Claude on Nvidia's graphics processing units and Google's Tensor Processing Units-and the networking infrastructure that connects these chips belongs to companies like Broadcom. The AI company now counts more than 1,000 businesses spending over $1 million annually on its offerings. Every dollar of that enterprise adoption flows through the infrastructure layer, and Broadcom captures a meaningful slice.
This is where the S-curve thesis becomes undeniable. As AI moves down the curve toward mass adoption, the compute requirements don't diminish-they compound. Distributed training clusters grow larger, inference demand explodes, and the networking fabric must scale accordingly. Broadcom's role in enabling this scaling creates a defensible moat. The company isn't just selling chips; it's providing the architectural foundation that makes exponential AI growth physically possible.
Morningstar recognizes this positioning. Broadcom earned a spot among the top constituents of the AI index with a 4-star rating-the same recognition given to Nvidia and Microsoft. Yet the 2026 "anything-but-AI" selloff has created the same valuation gap here that opened for its infrastructure peers. The stock trades at a meaningful discount to fair value, offering investors a chance to buy into the connectivity layer at a fraction of its intrinsic worth.
The key insight: AI cannot scale without networking. As the industry moves from proof-of-concept to enterprise deployment, the companies that control the infrastructure fabric win. Broadcom is that company. The S-curve is still ascending, and the connectivity demand is just beginning to accelerate.
Catalysts and Risks: What Could Break the Thesis
The infrastructure thesis rests on a simple premise: AI adoption is still climbing the steep part of the S-curve. But every investment thesis has its guardrails and watchpoints. For this portfolio, the signals are clear.
Watch Anthropic's revenue run rate. It's the leading indicator for infrastructure demand. The company's annual revenue just jumped to $30 billion from $9 billion in three months-a 3x increase that signals enterprise adoption is accelerating, not slowing. Nvidia's CEO put it plainly: these companies are "severely capacity constrained because demand is just incredible." As long as Anthropic's run rate keeps climbing, the infrastructure buildout has room to run. A slowdown there would be the first warning sign that the S-curve is flattening.
Geopolitical and regulatory risk is real. Tech stocks already pulled back weeks ago as Middle East hostilities weighed on sentiment. The recent de-escalation hopes provided a brief rally, but the underlying vulnerability remains. AI infrastructure is a global supply chain-chips designed in California, manufactured in Taiwan, deployed worldwide. Any disruption at scale would ripple through the entire thesis. Regulatory uncertainty adds another layer. OpenAI's CEO recently called for de-escalation around AI rhetoric after attacks on his home, acknowledging the technology's power is creating real-world tension. The market hasn't priced in a major regulatory crackdown or supply chain rupture. That's a risk, not a certainty-but it's material.
Amazon's Globalstar play opens a new frontier. The $11.57 billion acquisition isn't just about satellite internet-it's a signal that edge compute is expanding beyond traditional data centers. Amazon plans to deploy 3,200 satellites in the next three years. This could create new infrastructure demand: distributed compute nodes at the edge, low-latency connectivity for AI applications, and new deployment environments for the chips and networking gear that Broadcom and Nvidia make. It's an opportunity to think about the S-curve extending into new physical spaces.
The guardrail is straightforward: if AI adoption slows or enterprise budgets contract, infrastructure spending is first to be cut. The current thesis assumes the opposite-that productivity gains from AI are driving real ROI and that companies are committing to long-term deployment. Anthropic's 1,000 businesses spending over $1 million annually suggests that commitment exists. But it's fragile. A recession, a sharp pullback in tech spending, or a credible alternative to GPU-centric architecture would all threaten the thesis.
The bottom line: the thesis holds as long as the exponential curve holds. Watch the leading indicators-Anthropic's revenue, enterprise adoption rates, capacity constraints. The risks are manageable but real. The opportunity, however, remains early enough that the curve has further to climb.