Two numbers from Nvidia describe the AI chip market better than any earnings call. The company’s gross margin reached 74.9% in its latest quarter, and it announced a $5 billion investment in Intel, a company that competes with it in chips and is trying to compete with its supplier.
Read together, the two data points frame the strange economics of the moment. The margin says Nvidia’s pricing power is intact, that customers will pay almost anything for its accelerators, and that demand for AI computing still runs far ahead of supply. The investment says Nvidia is worried about where those chips get made.
The 74.9% gross margin, disclosed with Nvidia’s most recent results, is the kind of number most manufacturers cannot approach. It reflects a market in which Nvidia’s data center GPUs sell out months in advance and customers accept allocations rather than negotiate price. Pricing power at that level comes from a simple condition: there is only one supplier of the chips everyone wants, and its production is effectively sold out.
The $5 billion stake in Intel is harder to read as a pure financial decision. Intel is an x86 server chip rival, and it is spending billions trying to become a credible second advanced foundry, the position Taiwan Semiconductor has held for a generation. Nvidia is simultaneously a customer of Taiwan Semiconductor and a competitor to Intel in several product lines. Investing in a company that wants to take business from your supplier, and from you, is a hedge, not a friendship.
The logic, as executives and analysts describe it, is about capacity. AI accelerator demand has grown so fast that every available wafer at every advanced foundry is spoken for. Taiwan Semiconductor is expanding as quickly as it can, but its fabs are in Taiwan, and the geopolitical concentration of the world’s most advanced manufacturing has become a board-level risk for every company that depends on it. A second source of advanced chips, located in the United States, changes the risk profile even if it costs more.
Intel’s 18A process is the candidate. Nvidia and Intel have held discussions about producing some of Nvidia’s chips at Intel fabs, according to people familiar with the matter, and analysts at Bank of America noted that Intel is working to build chips for both Google and Nvidia. The $5 billion investment gives Nvidia a seat at the table and a stake in Intel’s success without committing Nvidia’s own production plans. Bank of America analysts noted that Intel’s shares have gained roughly 436% over the past twelve months, and the Nvidia investment, announced this month, was followed by further gains.
There is also a competitive dimension. Nvidia’s dominant position has made it the target of every other chip company’s strategy. Google designs its own TPUs and is financing them aggressively. Amazon makes its own accelerators. Broadcom designs custom chips for hyperscalers. A healthy, well-funded Intel, with American government support behind it, complicates the stories of Nvidia’s rivals as much as it complicates Nvidia’s own supply chain.
The margin figure also explains why Nvidia can afford the gesture. At 74.9% gross margin, Nvidia generates enormous cash from every accelerator it sells, and deploying part of that cash into a potential second supplier is cheap insurance. The investment is small relative to Nvidia’s cash position, and the strategic optionality it buys is large.
The two numbers also frame the debate over Nvidia’s durability. Bulls argue that a 74.9% gross margin is the reward for a decade of platform investment, and that the software ecosystem, CUDA and its successors, keeps customers locked in even as rivals ship competitive hardware. Bears argue that a margin that high is an invitation: every hyperscaler with a checkbook is funding an alternative, from Google’s TPUs to Amazon’s custom silicon to AMD’s Instinct line, and the $5 billion in Intel is an admission that even Nvidia sees the concentration risk in its own supply chain. Both readings are visible in the two disclosures, which is precisely why the company chose to make them public.
The risks are real. Intel’s 18A yields are not yet proven at commercial scale, and foundry customers measure success in defect rates and delivery dates, not announcements. The investment could produce nothing more than a passive stake in a struggling turnaround. But the signal is what matters to the market: the company with the most pricing power in the industry believes that having one source for its most important input is a risk worth paying $5 billion to address.
Together, the two numbers describe an industry in which the winner’s advantage is compounding while the winner hedges against the fragility of its own supply chain. Nvidia’s margin is the reward for being the only game in town. The Intel investment is the acknowledgment that the game’s location is the vulnerability.
