The plan was simple: build your own chips, cut Nvidia out. It isn't working.
According to The Information, Nvidia's share of the AI inference chip market appears to be rising—a counterintuitive finding given how aggressively the rest of the industry has moved to unseat it.
Over the past few years, major AI developers and cloud providers have poured resources into designing their own server chips, explicitly to reduce their dependence on Nvidia's hardware. Analysts and executives at those companies predicted the in-house designs would chip away at Nvidia's dominance, particularly in inference—the process of running a trained AI model to generate responses, which now accounts for the bulk of AI computing demand.
That hasn't happened, according to The Information. Instead, Nvidia's position in the inference market appears to be strengthening rather than eroding.
The finding matters because inference is where the AI economy actually runs. Training a model is a one-time (if expensive) event; inference is the continuous, round-the-clock workload that powers every chatbot query, every image generation, every AI-assisted search. Whoever owns inference owns the recurring revenue stream at the heart of the AI industry.
For years, the conventional wisdom held that Nvidia's training dominance was unassailable but its inference lead was vulnerable—inference is less computationally exotic, the thinking went, making it easier for custom silicon to compete on price and efficiency. The emerging picture, at least according to this report, suggests that advantage has not materialized in the way skeptics hoped.
If Nvidia is genuinely expanding its inference footprint even as rivals invest billions in alternatives, it raises serious questions about whether any competitor can meaningfully dent the company's position—and what that means for the cost and control of AI infrastructure for years to come.