Here’s a plot twist nobody saw coming: the world’s richest tech companies are quietly staging a coup against Nvidia.
For years, Google, Amazon, Meta, and Microsoft had no choice but to bow to the chip giant. Every quarter, they’d write enormous checks for Nvidia’s H100 GPUs—$40,000 per chip, with Nvidia pocketing roughly $30,000 in pure profit. That’s not a business relationship; that’s a protection racket.
But something shifted. When the AI boom exploded faster than anyone expected, Nvidia couldn’t manufacture chips fast enough. Suddenly, the companies building the future of AI were stuck waiting in line like everyone else. Amazon told investors it had “more demand than it could fulfill” because of chip shortages. Anthropic couldn’t train Claude fast enough. The most sophisticated tech companies on Earth were capacity-constrained by one supplier.
That’s when they did what they always do: they built their way out.
The Custom Chip Revolution Is Already Here
Google started this years ago with its Tensor Processing Unit (TPU) back in 2015—a skunkworks project nobody paid attention to. Then Amazon launched Inferentia and Trainium. Meta built MTIA. Microsoft created Maia. By 2023, every major hyperscaler had proprietary AI chips in production.
The financial press missed it entirely, too busy watching Nvidia’s stock price while its biggest customers were spending billions to reduce their dependence.
Then Amazon dropped the mic with Project Rainier—an internal AI cluster built entirely on its own Trainium2 chips. It contains 500,000 custom processors. For context, OpenAI used about 25,000 Nvidia GPUs to train GPT-4. Project Rainier is 20 times that, all on Amazon’s own silicon.
April 2026: The Dam Broke
In just two weeks this April, the deals started flying:
– Broadcom extended its TPU partnership with Google through 2031
– Meta committed to $2 billion per year in custom chip design fees
– Google started a second design partnership with Marvell for additional custom chips
– Amazon announced a $25 billion investment in Anthropic, with Anthropic committing $100 billion to AWS over 10 years—centered on custom chips
That’s over $200 billion in long-term commitments pointing in one direction: custom silicon.
Nvidia’s data center share has already dropped from 90% in 2022 to around 75% in late 2025. More than half of hyperscaler inference workloads now run on custom chips. Custom chip sales are growing at 45% annually—nearly three times GPU growth.
The Real Money Is in the Toll Roads
Here’s the thing: in any tech revolution, there are destination companies (fighting for dominance) and toll road companies (collecting fees regardless of who wins).
Intel didn’t pick sides in the PC revolution—it supplied chips to everyone. From 1981 to its peak, Intel delivered over 10,000% gains.
The custom AI chip toll roads are forming around companies like Broadcom, Marvell, Arm Holdings, Synopsys, Cadence, Taiwan Semiconductor, and networking players like Arista and Credo. These companies get paid whether Google wins or Amazon wins—they just need the revolution to keep happening.
And as of April 2026, it has $200 billion in signed commitments pushing it forward.
This isn’t speculation about future revenue. It’s already in the earnings reports. The contracts are signed through the end of the decade.
