Key Takeaways:
Google's TPU and Amazon's Trainium are the two most credible alternatives to Nvidia's GPU dominance, with each taking a fundamentally different path to the same goal.
Google's TPU-powered Cloud grew 63% to $20 billion in Q1 while Amazon's Trainium business topped a $20 billion annual run rate, making both the strongest in-house challengers to Nvidia's AI chip monopoly.
"We own frontier models and own the silicon," Sundar Pichai, chief executive officer of Google, told investors on the earnings call, framing the TPU as a vertical weapon tied directly to the Gemini model stack.
Amazon's approach is the mirror image. Andy Jassy, Amazon's CEO, said the chips business grew triple digits year over year with total Trainium commitments exceeding $225 billion. Trainium2 undercuts comparable Nvidia GPUs by 30% on cost per unit of performance, according to Amazon. AWS grew 28%, its fastest pace in 15 quarters, aided by multi-gigawatt training deals from Anthropic and OpenAI.
The philosophical split — Google treating TPU as an internal moat versus Amazon selling Trainium as a merchant chip — carries distinct investment implications. Google's forward P/E of 25 and 22% revenue growth make it the cleaner near-term buy. Amazon's forward P/E near 29 demands patience, but its distribution muscle could make Trainium the dominant non-Nvidia standard by 2027.
TPU 8t vs. Trainium3: Two Silicon Strategies, One Target
Google's new 8th-generation TPU 8t delivers 3x the processing power of its predecessor Ironwood, and Gemini now handles more than 16 billion tokens per minute via direct API. The company's Cloud backlog nearly doubled to $462 billion, though the CFO noted the majority of TPU hardware revenue lands in 2027. Google plans $180 billion to $190 billion in capital expenditures for 2026.
Amazon's Trainium3, which began shipping early this year, offers 30% to 40% better price performance than Trainium2, the company claims. Jassy said the chip is "nearly fully subscribed" and hinted at selling racks directly in the next couple of years. Amazon's 2026 capex guide sits at roughly $200 billion. Meta, Anthropic and OpenAI have signed as anchor Trainium customers.
Why the Market Is Watching 2027 Deliveries
The next test for both companies is converting backlog into shipped silicon. Google must prove it can turn that $462 billion pipeline into delivered TPU capacity. Amazon faces margin pressure if memory costs keep rising before Trainium3 reaches full scale. Both companies are spending at levels that require AI revenue to materialize on schedule.
For Nvidia, the threat is cumulative. Every dollar a hyperscaler spends on its own silicon is a dollar that does not flow to Nvidia's data center segment, which generated $62 billion in the most recent fiscal year. Trainium2's 30% cost advantage and TPU 8t's 3x generational leap each chip away at Nvidia's pricing power from a different angle.
Google shares have gained 18.7% year to date, outperforming Amazon's 10.5% rise. Google's Cloud operating margin nearly doubled year over year to 32.9%, showing that its vertical silicon strategy is already compounding margin. Amazon's broader thesis depends on Trainium becoming the default non-Nvidia cloud standard — a bet that requires years of execution. Morgan Stanley and Mizuho have raised price targets on Amazon this year, citing AWS acceleration and Trainium demand.
This article is for informational purposes only and does not constitute investment advice.
