Amazon is in talks to acquire an AI chip startup focused on chip flexibility, according to a report, a move aimed directly at reducing its reliance on NVIDIA GPUs in the data center.

The report, surfaced via Analytics India Magazine, does not name a purchase price or a closing timeline. But the target's focus — flexibility in AI silicon — signals what Amazon wants: chips it can adapt quickly to shifting AI workloads rather than fixed-function accelerators bought from an external vendor.

Why is Amazon shopping for silicon now?

Amazon already designs its own accelerators. Its Trainium chips train large models and its Inferentia line handles inference, both deployed across AWS infrastructure. Yet the company still buys NVIDIA GPUs at scale to satisfy customers who demand them, and NVIDIA hardware remains the default choice for much of the AI buildout.

An acquisition would bring that design talent and intellectual property in-house rather than licensing or partnering. Buying a startup gives Amazon engineers and architecture immediately, sidestepping the multi-year cadence of organic chip development.

The competitive logic is straightforward. Every dollar of AI compute that runs on Amazon's own silicon carries higher margin for AWS than the same dollar spent on NVIDIA hardware, where NVIDIA captures the majority of the value. Cloud rivals face the same arithmetic, which is why custom silicon programs now run at nearly every major hyperscaler.

What does "flexibility" mean in an AI chip?

Startups in this segment pitch reconfigurable or software-defined architectures: silicon whose dataflow, precision or compute blocks customers can retarget as models change. That matters because AI model architectures evolve faster than chip design cycles. A flexible accelerator can adapt to new attention mechanisms or sparsity tricks without a full tape-out.

For AWS, that flexibility could complement Trainium and Inferentia rather than replace them — one line optimized for known workloads, another able to chase new ones.

How does this fit the NVIDIA rivalry?

NVIDIA's grip on AI training comes from more than chips. Its CUDA software stack locks in developers, and its annual release cadence keeps it ahead on performance. Competing requires matching that ecosystem, not just the hardware.

Amazon's leverage is demand-side: AWS hosts a large share of the world's AI workloads and can steer customers toward its own silicon through pricing and placement. An acquisition strengthens the supply side of that equation — more design capacity, faster iteration.

The talks are at a report stage. No price, no target name confirmed by Amazon, no regulatory filing. Deals of this kind can collapse, and Amazon has walked away from silicon bets before restructuring them.

If the acquisition closes, expect AWS to fold the startup's architecture into its accelerator roadmap and pitch it to customers as a cheaper, more adaptable alternative to rented NVIDIA GPUs.