Google and Amazon Push to Sell Custom AI Chips Externally, Directly Challenging Nvidia
Claire Weston
Google and Amazon are both moving to sell their in-house AI chips to outside customers, shifting from internal use to a direct challenge against Nvidia's 85%–92% grip on the data-center GPU market.
How is Google selling its chips?
Google is providing $3.2 billion in financing to TeraWulf's Lake Mariner data center, acquiring a 14% stake in the company.
The facility will pivot from crypto mining to AI and high-performance computing, deploying thousands of Google TPUs.
AI startup Anthropic has signed on to lease most of that capacity. This means → Google is backing the TPU's competitiveness against Nvidia GPUs with a real commercial deal, not a press release.
Google had previously said it would sell TPUs to "select" customers. This project is the first large-scale execution of that strategy.
Where does Amazon stand?
Per Bloomberg, Amazon has announced plans to sell its custom AI chips externally — Trainium3 for training and Inferentia2 for inference.
Amazon AI head Peter DeSantis confirmed talks are underway but did not name specific customers.
CEO Andy Jassy flagged this direction in his 2025 shareholder letter: "Demand for our chips is so significant that we will likely sell full racks to third parties."
In plain terms = chips Amazon originally built only for its own cloud are now in such high demand that keeping them internal no longer makes sense.
Can Nvidia's numbers still hold up?
Nvidia's FY2027 Q1 (ending April 26, 2025) revenue hit $81.6 billion, up 85% year-over-year; diluted EPS reached $2.39, up 214%.
Management guided Q2 revenue growth to accelerate further to 95% year-over-year. This means → at least through the current quarter, Nvidia's momentum shows no sign of slowing despite competition fears.
On valuation, Nvidia trades at roughly 23× forward price-to-sales, below Alphabet's 24× and Amazon's 27×. This reflects the market already pricing in some competitive pressure on Nvidia.
What is the real test for this battle?
The key checkpoint: whether top-tier AI customers like Anthropic will add TPUs or Trainium to their procurement alongside Nvidia GPUs for real, large-scale workloads.
In plain terms = the question isn't who announced what — it's whether big buyers split their actual chip orders between Nvidia and its new rivals.
Nvidia's moat — the CUDA software ecosystem and high customer switching costs — remains intact, but the era of "Nvidia is the only option" is starting to crack.
Content is for reference only, not financial advice.