Tenstorrent, a chip startup focused on AI inference, has raised $1.5 billion at a $6.2 billion valuation, according to TechCrunch’s report published on October 19, 2026. The money will go toward scaling chips built for serving open-weight AI models and for video generation. The round is the latest sign that investors are still willing to back challengers to Nvidia, even as the incumbent’s grip on AI hardware remains strong.

What the round looks like

The valuation more than doubled from the $2.6 billion Tenstorrent fetched in late 2024, per TechCrunch. The round was led by Trace Capital, a new firm launched by Quiet Capital co-founder Ben Mahdavi, and by Innovation X. Other backers named in the report include Cadence Ventures, Eclipse, Fidelity, SanDisk and T. Rowe Price. The company is ten years old, which makes it an unusually long-lived entrant in a field where many startups are only a few years into their first product cycle.

The raise also arrives with a notable backstory. Five months earlier, TechCrunch notes, Tenstorrent had reportedly received acquisition interest from Intel and Qualcomm at valuations of as much as $5 billion. The company chose to stay independent. Jim Keller, the veteran microprocessor engineer who serves as Tenstorrent’s CEO, told Bloomberg that the decision followed a demonstration that the company’s Galaxy servers could run DeepSeek models. We have not independently verified the acquisition-interest report, which TechCrunch attributes to earlier reporting.

Why open-weight models are the target

Tenstorrent is explicitly aiming at inference for open-weight models, meaning models whose trained parameters are published so that anyone can download and run them (see our glossary entry on open-weight models). That is a different market from the one served by the largest closed labs, which increasingly build custom silicon or rent huge Nvidia clusters.

The logic is straightforward. Companies that want to host an open model themselves need hardware that is cheap to buy and flexible to deploy, and they are not locked to a single vendor’s software stack. A chipmaker that can show strong results on popular open models gives those buyers an alternative to the default choice. Keller’s reference to DeepSeek models fits that pitch: running a widely used open model well is the most visible proof point a hardware startup can offer.

The second stated target, video generation, is a heavier workload per request than text. Video models need a large amount of memory and compute for each clip, which is why serving costs for them are a recurring concern for providers. Tenstorrent has not, in the coverage we reviewed, published performance numbers tied to this round, so the claim should be read as a product direction rather than a measured result.

A crowded field of Nvidia challengers

TechCrunch frames Tenstorrent as part of a wave of AI hardware startups trying to chip away at Nvidia’s dominance. It points to Etched, which the outlet reported earlier this month was fielding investment offers at valuations of $40 billion or more. That figure is far above Tenstorrent’s and illustrates how unevenly capital is flowing among challengers: investors appear to reward companies with a sharp story about a specific workload.

Nvidia’s advantage is not only silicon. As we explain in why Nvidia dominates AI, the CUDA software ecosystem makes switching costly, and any challenger has to match not just speed but also the convenience developers expect. Big cloud companies have taken a different route by designing their own accelerators, a trend covered in why Big Tech builds its own AI chips. A startup selling to everyone else has to win on price, availability or openness.

What to watch next

  • Whether Tenstorrent publishes independent benchmark results for Galaxy servers on popular open-weight models.
  • How the new capital is split between inference chips and video-generation hardware.
  • Whether other startups, including Etched, close rounds at the valuations being reported.
  • Customer announcements that move Tenstorrent from demonstrations to production deployments.

Key takeaways

  • Tenstorrent raised $1.5 billion at a $6.2 billion valuation, up from $2.6 billion in late 2024.
  • Trace Capital and Innovation X led the round; Fidelity and T. Rowe Price also participated.
  • The company reportedly turned down acquisition interest from Intel and Qualcomm of up to $5 billion.
  • The focus is inference for open-weight models and video generation, not frontier training.
  • No independent performance figures accompanied the announcement, so claims remain unverified.

Frequently asked questions

Who runs Tenstorrent?

Jim Keller, a prominent microprocessor engineer, is the CEO, according to TechCrunch.

Why does inference matter more than training for a startup like this?

Inference, the work of running a finished model for users, happens continuously and at scale, so buyers care about cost per query. Training clusters are concentrated among a few labs, while inference demand is spread across many customers that a new chip vendor can reach.

Is this a threat to Nvidia?

Not immediately. A single funding round does not change market share, and Nvidia’s software ecosystem remains a major barrier. It does show that capital for alternatives is available.