AWS Launches New Nova AI Models And a Service That Gives Customers More Control

Trainium3 Chip Takes Center Stage as Amazon Challenges Nvidia

Amazon’s new Trainium3 chip is already dominating tech conversations, as users search for whether Amazon can finally challenge Nvidia’s grip on the AI hardware market. The company used the AWS re:Invent 2025 stage to unveil a major leap in performance, promising faster speeds, lower power usage, and dramatically cheaper training costs. In the first few minutes of this announcement, one question became clear: Is Amazon now the strongest competitor Nvidia has ever faced? With billions of dollars already tied to its Trainium line, Amazon believes the answer is yes.

Credit: Kevin Winter / Getty Images

Amazon Says Trainium3 Is 4x Faster—and Cheaper—Than Before

During the keynote, Andy Jassy confirmed what industry watchers had anticipated for months: Trainium3 is not just an upgrade, but a significant redesign meant specifically to win high-volume cloud customers. The new chip reportedly delivers 4x the performance of Trainium2 while using less power, a combination that directly targets Nvidia’s lucrative AI GPU market. Jassy emphasized that cost-efficiency has always been Amazon’s differentiator, and Trainium3 applies that same strategy at massive cloud scale. Early reactions suggest cloud developers are already lining up to test whether it really can offer high-end training at a fraction of the price.

Trainium2 Already a Multibillion-Dollar Business, Jassy Reveals

In a rare disclosure posted on X, Jassy shared internal metrics that highlight the momentum behind Amazon’s chip strategy. According to him, Trainium2 has already reached a multibillion-dollar revenue run rate, with more than 1 million chips in production and over 100,000 companies actively using it. For an in-house silicon program launched just a few years ago, those numbers stunned analysts. The scale also suggests Trainium is no longer an experiment—it’s now one of AWS’s most profitable growth engines.

Bedrock Drives Massive Demand for Amazon’s AI Silicon

One key factor behind Trainium’s rapid adoption is Amazon Bedrock, the company’s platform for accessing top-tier foundation models. Bedrock lets customers mix and match AI models based on cost or accuracy, and Jassy noted that most current Bedrock usage is already running on Trainium2. That shift hints at an intentional strategy inside AWS: funnel high-traffic workloads toward custom chips instead of Nvidia GPUs. The more Bedrock grows, the more Trainium grows—creating a feedback loop that strengthens Amazon’s hardware ambitions.

Why Amazon Thinks It Can Beat Nvidia on Price-Performance

For years, Nvidia has been untouchable in AI hardware due to unmatched performance and software dominance. Yet Jassy believes Amazon has found its wedge: price-performance optimization. Amazon claims Trainium delivers better efficiency at significantly lower operating costs, making it attractive for companies unable to secure expensive Nvidia clusters. In cloud computing—where margins are razor-thin and workloads are unpredictable—offering comparable power for less money is a massive advantage. AWS insiders say customers increasingly see Trainium as “good enough” for many large-scale training tasks.

Inside Amazon’s Strategy: Homegrown Tech at Cloud Scale

Amazon’s playbook hasn’t changed since the earliest days of AWS: build its own tools, cut costs, and pass savings to customers to win market share. Trainium mirrors Amazon’s history with S3, EC2, and the company’s own ARM-based Graviton chips. By reducing reliance on external suppliers and handling silicon design in-house, Amazon gains direct control over pricing and availability. In a cloud market strained by GPU shortages, that advantage becomes even more important. Amazon believes its vertical approach could reshape the economics of AI training.

AWS CEO Matt Garman Reveals Anthropic Is Driving Huge Chip Demand

AWS CEO Matt Garman added new details in an interview with CRN, confirming that one customer in particular is responsible for a large portion of Trainium’s booming revenue: Anthropic. The AI safety-focused startup is one of Amazon’s largest cloud partners and a major user of Bedrock. Anthropic’s rapid model development cycles require enormous compute, and AWS sources say Trainium has become central to powering those workloads. With Anthropic scaling globally, Amazon is effectively guaranteed a steady pipeline of high-value demand.

Anthropic’s Reliance on Trainium Signals a Bigger Shift in the Market

Anthropic’s adoption of Trainium doesn’t just represent a big customer win—it signals that major AI labs are now willing to run frontier model training on non-Nvidia hardware. That shift would have been unthinkable two years ago when every leading model—from GPT-4 to Llama 2—relied primarily on Nvidia GPUs. Now, high-profile labs are proving that competing chips can support cutting-edge training. If more AI labs follow suit, Trainium could accelerate the diversification of the entire AI silicon market.

Can Amazon Really Take Nvidia’s Market Share?

Despite Amazon’s momentum, Nvidia remains far ahead in both performance and software dominance, especially with its CUDA ecosystem and accelerating GPU innovation. Yet analysts say Amazon doesn't need to “defeat” Nvidia to win. The AI market is expanding so fast that simply capturing a slice of the demand could generate tens of billions in annual cloud revenue. Trainium’s positioning as a cheaper, cloud-native alternative makes it compelling for customers who don’t need the most advanced GPU features.

The Real Battle: Supply Chain, Not Just Performance

Industry experts point out that the race isn’t just about raw compute power. The real competition is who can deliver chips at scale without shortages. Nvidia’s supply constraints have left customers waiting months for access to new GPUs. Amazon, by contrast, is designing chips specifically for its own data centers and can allocate Trainium nodes instantly to cloud customers. If AWS can guarantee availability while Nvidia struggles with demand bottlenecks, Trainium’s appeal could rise even faster.

Why Trainium3 Could Reshape AI Cloud Economics in 2026

With Trainium3 entering general availability next year, AWS is preparing for a surge in enterprise interest. Many businesses that held off due to performance concerns are now re-evaluating custom silicon due to rising GPU prices and AI budget pressures. Amazon claims Trainium3 will drastically reduce training time for enterprise-scale models while keeping operational costs predictable. If those claims hold, Trainium3 could become a central pillar of AI infrastructure for companies migrating large workloads to the cloud.

Amazon Bets Big: Is This the Future of Cloud AI?

For Amazon, Trainium isn’t just another product line—it’s a long-term bet on owning the infrastructure behind the next generation of AI systems. While Nvidia is still miles ahead in market share and ecosystem strength, Amazon’s strategy reflects a new competitive landscape where multiple chip suppliers can thrive. With Trainium3 now positioned as a cost-efficient, high-performance alternative, AWS is signaling that the future of AI computing won’t be dominated by a single compan