Amazon and Google’s custom AI chips take on Nvidia: AWS Trainium and Google TPU roadmaps point to a $200 billion battle for AI computing dominance by 2027
Amazon and Google are both betting that custom silicon will let them capture a larger share of the AI computing boom, and both companies have recently signaled that their in-house chips are central to their long-term cloud strategies. Amazon Web Services has been expanding its Trainium and Inferentia chip families, while Google has been deploying its Tensor Processing Units (TPUs) for both internal AI workloads and external cloud customers. The two companies are positioning themselves as alternatives to Nvidia, whose GPUs have dominated the AI training market and helped drive the chipmaker’s data center revenue to record levels.
On October 30, 2026, both companies provided updates on their custom chip roadmaps during their respective earnings calls, according to sources familiar with the matter. Amazon said its Trainium2 chips are now being used by Anthropic for training large language models, and that Trainium3 is on track for launch in late 2026. Google said its Trillium TPU, the sixth generation of its custom chip, is now generally available and delivers a 4.7x improvement in compute performance over the previous generation.
Why AWS Is Betting Big on Trainium2 and Trainium3
Amazon’s custom chip strategy is centered on reducing its dependence on Nvidia GPUs and offering customers a lower-cost alternative for AI workloads. The company has said that Trainium2 offers up to 30% better price-performance than comparable GPU instances, and that it is being used by Anthropic, one of the leading AI model developers, to train its Claude models. Amazon has invested $8 billion in Anthropic, and the two companies have a strategic partnership that includes using AWS as Anthropic’s primary cloud provider and Trainium chips for training and inference.
The Trainium3 chip, which is expected to launch in late 2026, is built on a 3-nanometer process and is designed to deliver even greater performance and energy efficiency. Amazon has not disclosed specific performance figures for Trainium3, but analysts expect it to be a significant step up from Trainium2 and to help AWS compete more effectively with Nvidia’s latest GPUs. The company has also been expanding its Inferentia chips for inference workloads, which are typically less compute-intensive than training but still require specialized hardware.
Google’s TPU Strategy: Trillium and the Path to External Customers
Google’s TPU strategy is different from Amazon’s in that Google has been using TPUs internally for years to power its own AI services, such as search, translation, and YouTube recommendations. The company began offering TPUs to external customers through Google Cloud in 2018, and has been steadily expanding that business. The Trillium TPU, the sixth generation of Google’s custom chip, is now generally available and is designed to deliver significant improvements in performance and energy efficiency.
Google has said that Trillium offers a 4.7x improvement in compute performance over the previous generation, TPU v5e, and that it is being used by customers such as Anthropic, Salesforce, and Midjourney. The company has also been developing its own AI models, such as Gemini, using TPUs, which gives it a unique advantage in optimizing its hardware and software together. Google has not disclosed the revenue from its TPU business, but analysts estimate that it is a small but growing part of Google Cloud’s overall revenue.
The $200 Billion AI Chip Market and Nvidia’s Dominance
The AI chip market is expected to grow to more than $200 billion by 2027, according to estimates from industry analysts. Nvidia currently dominates this market with an estimated 80% share of the AI accelerator market, according to research firm IDC. The company’s data center revenue has been driven by demand for its H100 and H200 GPUs.
Amazon and Google are hoping to capture a larger share of this market by offering custom chips that are optimized for specific workloads and are more cost-effective than Nvidia’s GPUs. However, both companies face significant challenges. Nvidia’s CUDA software platform is widely used by AI developers and has become a de facto standard, which creates a strong network effect that favors Nvidia. Amazon and Google have been developing their own software tools to make it easier for developers to use their chips, but they are still playing catch-up.
What to Watch: Trainium3 Launch and Trillium Adoption
Investors should watch for the launch of Amazon’s Trainium3 chip in late 2026, which will be a key test of whether AWS can offer a competitive alternative to Nvidia’s latest GPUs. Google’s ability to attract external customers to its Trillium TPUs will also be closely watched, as will any announcements about new TPU generations. The companies’ capital expenditure guidance for 2027 will provide clues about how much they are investing in custom silicon. A significant increase in custom chip revenue at either company would signal that the strategy is gaining traction, while a slowdown would suggest that Nvidia’s dominance remains intact.
Sources: rss_autopost · AMZN Newsroom · Yahoo Finance · CNBC · Bloomberg