Nvidia and Broadcom: The Compute and Connectivity Rails of the AI S-Curve

The narrative for AI spending has fundamentally shifted. Last year, the story was about the GPU bottleneck. This year, the market is pricing in what's coming: a move down the stack. The initial phase of exponential adoption has passed the compute layer, concentrating demand on the supporting infrastructure that feeds and connects it. This is the new S-curve.

The evidence is clear. As one analyst noted, 2025 AI & Semiconductor Outlook saw memory and optics become the beneficiaries, as bottlenecks moved from GPUs to the things that feed and connect them. The market is now rewarding companies providing these fundamental rails. This shift creates a new wave of demand for semiconductor firms, and they are experiencing significant pricing power as a result.

Nvidia sits at the epicenter of this new paradigm. Its latest results show the engine is accelerating. The company reported record revenue of $57.0 billion for the third quarter, a 62% year-over-year jump. More telling is the source: data center revenue, which powers AI training and inference, grew 66%. CEO Jensen Huang described the dynamic as "the virtuous cycle of AI". The AI ecosystem is scaling fast-with more new foundation model makers, more AI startups, across more industries, and in more countries. AI is going everywhere, doing everything, all at once.

This creates a self-reinforcing loop. As more companies build AI models and deploy them, the demand for Nvidia's compute grows. That growth fuels further investment in the AI stack, which in turn drives even more compute demand. It's a compounding effect where each wave of adoption builds the foundation for the next. The company's ability to maintain gross margins of 73.4% under this pressure underscores its pricing power and operational scale.

The bottom line is that the AI adoption curve has matured. The initial spike in GPU demand has settled into a sustained, accelerating climb, but the focus has shifted. The next phase of growth is being built on the rails of memory, networking, and the foundational compute that Nvidia provides. For investors, the thesis is no longer just about buying a GPU stock. It's about participating in the infrastructure layer of a paradigm shift, where the company that owns the compute stack is compounding its growth on every new wave of adoption.

Nvidia: The Foundational Compute Layer and the Custom Chip Frontier

Nvidia's record quarter confirms its role as the foundational compute layer for the AI paradigm. The numbers are staggering: record revenue of $57.0 billion for the third quarter, with data center sales alone hitting $51.2 billion and growing 66% year-over-year. This isn't just growth; it's the scaling of a new technological substrate. The company is moving down the S-curve, where the initial GPU bottleneck has given way to a demand for the massive, efficient compute that powers every new model and deployment. The strategic shift now is toward higher-margin, differentiated products. This is exemplified by the partnership with OpenAI to co-develop custom AI accelerators. The deal calls for deploying at least 10 gigawatts of NVIDIA systems for OpenAI's next-generation infrastructure. This move is a classic frontier play. By embedding the lessons from training frontier models directly into custom silicon, the goal is to unlock new levels of capability and intelligence. It's a trend where the most advanced users are building their own chips, but they are doing so on the fundamental compute stack that Nvidia provides.

This trend represents a powerful extension of competitive advantage. Custom silicon allows for tighter integration and optimization, which can command premium pricing. For Nvidia, it means shifting from selling standardized GPUs to enabling and co-developing the specialized hardware that will define the next generation of AI. It's a move from a commoditized product to a platform for innovation, where Nvidia's foundational architecture remains the essential rail.

The bottom line is that Nvidia is not just selling compute; it is defining the infrastructure layer. Its massive scale provides the platform, while its strategic partnerships with the most advanced AI builders ensure it remains at the center of the next exponential wave. The company is building the rails not just for today's AI, but for the systems that will follow.

Broadcom: The Strategic Connectivity Layer and Pricing Power

While Nvidia owns the compute stack, Broadcom is building the essential connectivity layer that makes it all work. In the AI paradigm, raw power is useless without the ability to move data at scale. Broadcom's role as a key supplier of networking chips and custom AI accelerators positions it as a foundational partner, not just a vendor. This strategic importance was cemented last quarter with a major multi-year partnership to co-develop systems for OpenAI.

The deal is a direct validation of Broadcom's infrastructure role. The companies will co-develop accelerators and Ethernet solutions, with Broadcom deploying racks of AI systems targeted for the second half of 2026. The goal is to scale up and scale out next-generation AI clusters, meeting surging global demand. By designing its own chips, OpenAI can embed what it's learned from frontier models directly into the hardware. Broadcom's networking solutions are the critical enabler for this custom silicon, ensuring the data flows efficiently across the cluster. This isn't a one-off contract; it's a long-term collaboration that ties Broadcom's growth directly to the expansion of the most advanced AI systems.

This partnership underscores a powerful trend: the most sophisticated AI builders are moving down the stack to design their own chips, but they are doing so on Broadcom's fundamental rails. The company's expertise in high-performance networking and custom ASICs gives it a sustained competitive advantage. This is reflected in its financial trajectory. Analysts expect Broadcom to deliver 51% growth for fiscal year 2026, a figure that demonstrates strong pricing power and a dominant position in the AI supply chain. In a market where demand is outstripping supply, Broadcom is flexing its muscle, much like its foundry partner TSMC, which recently implemented price hikes for advanced nodes.

The bottom line is that Broadcom is compounding its advantage by owning the connectivity layer. As AI adoption accelerates down the S-curve, the need for efficient, scalable systems grows exponentially. Broadcom's strategic partnerships with hyperscalers ensure it is not just a supplier, but a co-builder of the infrastructure that will define the next phase of the AI paradigm. Its growth story is built on the same exponential adoption curve, but on a different, equally critical, layer of the stack.

Catalysts, Risks, and the Path to 2030

The forward view for these infrastructure plays is defined by a clear catalyst and a looming cyclical risk. The primary near-term catalyst is the multi-year deployment of custom AI accelerator racks, with initial shipments targeted for the second half of 2026. This isn't just another contract; it's the validation of a new hardware architecture. As OpenAI designs its own chips, it embeds frontier model learnings directly into the silicon. Broadcom's role is to scale this vision, providing the Ethernet and connectivity solutions that make these custom accelerators work in massive clusters. This partnership, with its 10-gigawatt target, ties Broadcom's growth directly to the expansion of the most advanced AI systems, providing a multi-year revenue visibility that few companies can match.

The key risk, however, is the inherent cyclical nature of the semiconductor industry. While 2026 is expected to be a strong year, with analysts projecting 51% growth for Broadcom, the market may face a downturn in 2027-2028. This would follow a period of massive, front-loaded investment in AI infrastructure. The cycle is a classic capital-intensive boom, where today's surging demand for compute and connectivity fuels a wave of capacity expansion. The ultimate test for any infrastructure play is its ability to capture a durable share of the growing AI compute budget, measured by long-term revenue growth and pricing power. The current setup suggests these companies are well-positioned for the next phase of exponential adoption, but they are not immune to the inevitable market corrections that follow a paradigm shift.

The path to 2030 will be measured by their ability to maintain this infrastructure dominance. For Nvidia, it's about extending its foundational compute layer into new applications and custom silicon partnerships. For Broadcom, it's about deepening its role as the essential connectivity layer for every major AI cluster. The ultimate winners will be those who can navigate the next cycle, turning a period of industry consolidation into an opportunity to further solidify their position on the technological S-curve.