At Friday’s close, the semiconductor sector experienced turbulence, with Nvidia (NVDA) shares falling about 6% in a single day, pulling its market capitalization back to the $5 trillion mark. The market’s reaction was closely tied to news that three major cloud giants – Amazon (AMZN), Alphabet (GOOG), and Microsoft (MSFT) – are accelerating the deployment of their own self-developed AI chips into their data centers, aiming to reduce their reliance on Nvidia GPUs.
This is not a new story, but 2026 may prove to be an inflection point. Amazon’s custom chip business is already the most mature. Its Graviton processor, Trainium AI chip, and Nitro networking chip reached an annualized revenue run-rate of over $20 billion in the first quarter of 2026. CEO Andy Jassy went a step further on the earnings call, saying that if this business were a standalone operation selling chips to AWS and other third parties – as other leading chip companies do – its annualized revenue run rate would be $50 billion, placing it among the top three data center chip businesses in the world.
However, this has not curbed Amazon’s appetite for Nvidia. The company plans to spend roughly $200 billion on capital expenditures in 2026, with the vast majority of that infrastructure still heavily reliant on Nvidia GPUs to serve AWS customers.
Google is a veteran in self-developed chips. Its tensor processing units (TPUs) have now reached the eighth generation, and its strategy has fundamentally shifted: the chips are no longer for internal use only. In May, Blackstone announced a joint venture with Google to offer TPUs as a rentable cloud service, with an initial commitment of $5 billion and plans to bring 500 megawatts of capacity online in 2027. That followed an agreement to give AI lab Anthropic access to as many as 1 million TPUs, as well as a previously reported leasing deal with Meta Platforms. Even so, reports surfaced this week that Google signed a multi-year SpaceX cloud deal involving access to about 110,000 Nvidia GPUs – meaning self-development and procurement are proceeding in parallel.
Microsoft lags the furthest behind in custom silicon. Its effort centers on the Maia accelerator, and the second generation Maia 200 only recently went live in some data centers to support some of the workloads behind Microsoft 365 Copilot and partner OpenAI’s models. Still, the vast majority of AI workloads inside Microsoft’s Azure cloud run on Nvidia GPUs, so Maia is a way to claw back some of that spending over time, not a wholesale replacement. The company expects to invest roughly $190 billion in capital expenditures during calendar 2026, and even though Azure revenue grew 40% in its fiscal third quarter (ended March 31, 2026), the cloud remains capacity constrained through year end.
Adding it up, the four tech giants – Amazon, Google, Microsoft, and Meta – are on track to spend roughly $725 billion on capital expenditures in 2026, up about 77% from last year. For Nvidia, this forms a clear bear case: a growing share of that spending will flow to chips its largest customers design themselves, and those customers have a strong incentive to reduce their dependence on a single supplier.
Yet the bull case also holds. Nvidia’s latest earnings report showed that for its fiscal first quarter of 2027 (ended April 26, 2026), revenue rose 85% year over year to $81.6 billion, with data center revenue up 92%. Hyperscalers still accounted for about half of that data center business. Founder and CEO Jensen Huang said on the earnings call that “demand has gone parabolic,” pointing to a fast growing tier of buyers – AI start-ups, enterprises, and governments – that “do not build chips, do not design their own chips.”
In the near term, both forces will coexist. Custom silicon is real and growing, and it could erode Nvidia’s pricing power over time. But the overall pool of AI spending is still expanding so quickly that even if Nvidia loses marginal share, its absolute revenue could continue to grow. With a current price-to-earnings ratio of about 32, the bigger risk may not be that in-house chips fail – it is that they succeed slowly while the market continues to price Nvidia for permanent dominance.
