Tenstorrent just raised $1.5 billion at a $6.2 billion valuation, turning down acquisition offers to stay independent and scale its open-weight AI chips. The company, led by legendary chip designer Jim Keller, is now one of the best-funded challengers trying to break Nvidia’s dominance in AI hardware.

The funding comes as the biggest cloud providers make their most aggressive moves yet to weaken Nvidia’s position. Amazon launched TorchNeuron this week, a developer tool designed to make it easier to run PyTorch workloads on non-Nvidia chips. That matters because PyTorch is the foundation for most AI applications today. If Amazon can convince developers to write code that runs just as well on its custom Trainium and Inferentia chips, Nvidia loses one of its biggest moats.

Microsoft is taking a different approach but heading toward the same goal. The company is warning enterprises to treat AI agents like insider threats because “frontier models can’t be trusted,” particularly when those agents interact with systems outside the control of major labs like OpenAI and Anthropic. Reading between the lines: Microsoft wants companies thinking hard about where AI workloads run and who controls the infrastructure.

The pressure isn’t just coming from software. Amazon also announced funding for a new nuclear plant in Indiana that will provide up to 960 megawatts of power for data centers. The facility, developed with X-energy and American Electric Power, will replace a defunct coal plant in Sullivan County. It won’t be online for several years, but the move signals how seriously the hyperscalers are taking long-term infrastructure planning as AI demand keeps climbing.

The chip wars are getting expensive

Tenstorrent’s raise is substantial, but it’s also a defensive move. Building chips that can compete with Nvidia requires massive capital for R&D, manufacturing partnerships, and customer acquisition. Keller has credibility from his work on processors at AMD, Apple, and Tesla, but credibility doesn’t pay for tape-outs.

The company is betting on open-weight models and a more flexible architecture than Nvidia’s tightly integrated hardware-software stack. That’s appealing to customers who want more control and potentially lower costs, but it’s also a harder sell than Nvidia’s plug-and-play ecosystem.

What’s notable is that Tenstorrent had acquisition interest earlier this year and chose to stay independent instead. That suggests either the offers weren’t high enough or the company believes it can build something significantly more valuable on its own. Given the valuation, probably both.

Nvidia isn’t standing still

While competitors push from every angle, Nvidia launched new open models designed to run AI agents in enterprise data centers and on individual PCs. The company is leaning into the trend of enterprises wanting to host their own models rather than rely on API calls to third-party services.

That’s smart positioning. If the market fragments between cloud-based AI and on-premise deployments, Nvidia wants to sell chips for both. The company’s advantage is that it doesn’t just sell hardware. It sells an entire platform that developers already know how to use.

What this means

The AI chip market is splitting into camps. Nvidia still dominates, but the hyperscalers are investing billions to reduce their dependence on a single supplier. Amazon’s TorchNeuron and its nuclear power deal show a company building for a future where it controls more of the stack. Microsoft’s warnings about AI agents suggest similar thinking.

Tenstorrent’s $1.5 billion raise is significant because it shows there’s still investor appetite for challenging Nvidia, even after years of failed attempts by other companies. The valuation implies Tenstorrent is already generating meaningful revenue or has customer commitments that justify the price.

The broader pattern is clear: the companies spending the most on AI infrastructure don’t want to be locked into a single vendor. That creates opportunities for challengers like Tenstorrent, but it also means the competitive bar is extraordinarily high. You’re not just building a chip. You’re building an alternative ecosystem.