The market narrative around Nvidia has always been about the GPU. The H100, the B200, the relentless cadence of silicon that fuels the world's large language models. But beneath this noise, a quieter, more structural shift is taking place. It is not about teraflops or tensor cores. It is about the unglamorous piece of silicon that feeds the beast: the CPU.
Recent signals from the company's internal forecasts point to a specific target: doubling CPU business revenue by fiscal 2028. On the surface, this is a simple growth metric. But for those of us who have spent years watching the liquidity of tech narratives pool and dissipate, this is a confirmation of something deeper. Nvidia is no longer selling acceleration. It is selling the entire infrastructure of computation. And the CPU is the keystone of that architecture.
To understand this move, you must first abandon the traditional x86 mental model. This is not Intel versus AMD, round four. This is a new category definition. Nvidia's Grace CPU is designed not as a general-purpose master controller, but as a specialized data feeder for its own GPUs. The architectural choice—Arm's Neoverse V2 cores paired with LPDDR5X memory and a proprietary NVLink-C2C interconnect—is a declaration of intent. The CPU is a peripheral to the GPU, but a critical one.
The numbers, based on my own tracking of DGX and HGX system shipments over the past year, suggest a baseline of roughly $40-60 billion in current CPU-attributable revenue. This is a fraction of the total, perhaps 3-5% of the company's top line. But the growth trajectory is what matters. A doubling from that base implies a run rate of $240-320 billion by FY2028. This is not incremental. This is a compound annual growth rate of 60-80%, a pace that would reshape the competitive landscape of the AI server market.
The financial mechanics are deceptively simple. The margin profile of a Grace CPU is lower than a Blackwell GPU. Selling more CPUs will mathematically dilute Nvidia's gross margin. This is the kind of surface-level concern that dominates short-term trading. But the strategic logic is more profound. The CPU enables the system. It locks in the customer. It makes the alternative—a mixed Intel or AMD server environment—a less attractive, higher-friction option.
The real battlefield is not the CPU socket. It is the system-level efficiency ratio. Nvidia's own data and several third-party benchmarks point to a 30-50% advantage in performance-per-watt for a Grace+Blackwell system over a comparable x86+GPU configuration. The NVLink-C2C interconnect, with its 900GB/s bandwidth, creates a communication channel that is seven times faster than PCIe 5.0. This is not an incremental improvement. It is a step change in how data moves between compute units.
This brings me to the contrarian view that most market commentary misses. The threat to Intel and AMD is not that Nvidia will steal their existing market share. The x86 fortress of enterprise general-purpose computing is secure for the next decade. The threat is that Nvidia is redefining the value proposition of the AI server itself. When the core competitive dimension shifts from "CPU single-core performance" to "CPU-GPU integration degree," the incumbents are playing a game they did not design and cannot easily win. AMD is the most realistic challenger, with its EPYC line and improving Instinct GPU integration. But it lacks the software ecosystem moat that CUDA provides.
There is also a geopolitical layer to this that warrants attention, particularly from my vantage point in Manila. The export controls on high-end AI chips to China have created a strange dynamic. They limit Nvidia's addressable market, but they also sever Intel and AMD from a massive customer base. In markets where "de-x86-ification" is a policy consideration, the Arm architecture offers a veneer of neutrality that x86 cannot claim. This is not a primary driver, but it is a tailwind that cannot be ignored.
The critical risk is not competition. It is cyclicality. The AI demand curve is steep, but it is not linear. A reduction in hyperscaler capital expenditure would hit Nvidia's CPU ambitions directly. The mitigating factor is that enterprise AI penetration is still in its infancy. But this is a cyclical industry, and I have seen liquidity illusions before. In the DeFi summer of 2021, we saw billions in TVL evaporate when the incentive structures were removed. The same principle applies to AI capex. It is a flow, not a stock. It can reverse.
Based on my experience auditing the economic sustainability of early DeFi protocols, I see a parallel here. The moat is not the hardware. It is the system integration and the software lock-in. Nvidia is building an economic moat, not just a technical one. The question is whether that moat can withstand a cyclical downturn in sentiment and spending.
Another signal to track is whether Nvidia begins selling Grace CPUs as a standalone product, decoupled from the GPU bundle. This would be a significant strategic pivot, opening a larger addressable market but also exposing the CPU to direct competition on its own merits. For now, the bundling strategy is sound. It maximizes customer stickiness and creates a formidable barrier to entry.
The takeaway for the macro-observant is clear: we are witnessing a transition from a chip company to a systems company. Nvidia's strategy is not about winning a benchmark. It is about controlling the architecture of AI infrastructure. The CPU is the instrument of that control.
When the dust settles, we may look back at this moment as the point where the AI hardware market was no longer defined by the silicon itself, but by the invisible threads of interconnect and software that bind it together. In that world, the CPU is not a commodity. It is a strategic asset. And Nvidia has just placed its bet.