Amazon’s AI Push: Trainium Chips and the Rise of Texas as an AI Hub

Texas is rapidly becoming a central location for artificial intelligence development, and Amazon is at the forefront of this expansion with its custom-designed Trainium chips. The company is working to reduce its reliance on Nvidia and other suppliers by creating its own AI-specific hardware, aiming for both cost savings and increased reliability.

Amazon’s Investment in Custom AI Chips

Tech giant Amazon is investing heavily in custom “Trainium” chips specifically engineered for machine learning. This move comes as billions of dollars are being poured into the AI sector globally. Amazon Web Services (AWS), the company’s cloud computing unit, began designing its own chips after acquiring Israeli startup Annapurna Labs in 2015. Prior to this, Amazon relied on external suppliers for its chip needs.

The first chips developed were Graviton (for general cloud computing) and Inferentia (for powering AI models), both released in 2018. The first Trainium chip followed in 2020, with a second generation offering improved performance. The latest iteration, Trainium 3, became operational in December 2025 and is reported to double the capabilities of its predecessor whereas being smaller than a credit card. Amazon is already developing Trainium 4, with plans to achieve six times the processing performance of Trainium 3. Yahoo Finance

Texas: A New Tech “El Dorado”

Texas is emerging as a prime location for tech investments, attracting companies with its combination of affordable real estate, cheap energy, relaxed regulations, and favorable tax incentives. Amazon’s Annapurna Labs subsidiary in Austin, Texas, is central to the development and testing of these new chips. Purdue Exponent

At the Annapurna Labs facility, UltraServers equipped with 144 Trainium AI-accelerator chips are undergoing rigorous testing before deployment. This testing focuses on ensuring the longevity and reliability of the chips, which is crucial for data centers that require continuous operation.

Cost and Reliability Advantages

Kristopher King, head of the Annapurna lab in Austin, claims that the latest Trainium chips can reduce the cost of developing and running generative AI models by up to 40% compared to using graphics processing units (GPUs), currently considered the industry standard for AI. Purdue Exponent

Reliability is likewise a key focus. Mark Carroll, head of engineering at Annapurna Labs, emphasizes that AI model training requires thousands of chips operating simultaneously for extended periods. Any failure during this process can necessitate restarting the training from the beginning, highlighting the importance of chip stability.

AWS’s Unique Approach

Unlike major AI chip manufacturers, Amazon Web Services does not sell its Trainium processors to third parties. Instead, it utilizes them exclusively within its own cloud infrastructure, which is then rented to customers. This strategy allows AWS to integrate its hardware with its software platform, Bedrock, offering a wide range of AI models developed both internally and by other companies like Anthropic, OpenAI, and Mistral.

Diversifying the AI Supply Chain

In a market facing challenges in meeting the growing demand for computing power for AI, AWS and its Trainium chips contribute to diversifying the supply chain, reducing reliance on Nvidia and AMD. This diversification is crucial for ensuring continued innovation and accessibility in the rapidly evolving AI landscape.