Tenspresso, Three Months Old, Seeks $50 Million at $500 Million Value

A three-month-old AI chip startup is negotiating to raise at least $50 million at a valuation of roughly $500 million, according to people familiar with the matter. The company, Tenspresso, is based in Sunnyvale, California, and has barely shipped a résumé's worth of history — yet investors are already circling.

The talks come only weeks after Tenspresso secured $20 million from Striker Venture Partners. Since that round closed, Altimeter Capital Management has approached the startup about putting in additional money, the people said. Tenspresso is also speaking with strategic investors. The final size of the round could still shift as those conversations develop, so the $500 million figure should be read as a working number, not a signed term sheet.

Who is behind Tenspresso?

The founding team is the story here. Dong Hyuk Woo and Harigit Khaitan — Harshit Khaitan — co-founded the company after spending several years working together on Google's tensor processing units (TPUs, Google's in-house AI accelerator chips). Much of the 15-person team has come over from early AI chip efforts at Alphabet Inc.'s Google.

That pedigree explains the speed. A 15-person outfit landing $20 million, then fielding inbound interest from Altimeter within weeks, doesn't happen without a track record investors can price. Ex-TPU engineers are among the most sought-after hires in semiconductors right now.

What is the company actually building?

Tenspresso is designing silicon for AI workloads that rely on long context windows — the capability that lets models work through larger and more varied pools of data on complex tasks, rather than reasoning over short snippets. The startup expects demand for such specialized chips to build in the years ahead.

The timeline is ambitious but not reckless by chip-startup standards. Tenspresso aims to finish design work on its first product in late 2027 or early 2028. That's roughly two to two-and-a-half years from founding to tape-out-ready design — tight, but plausible for a team that has shipped accelerators before.

Why look outside the standard supply chain?

This is where the plan gets contrarian. Tenspresso is in talks with design and manufacturing partners, and it's looking at options outside the conventional silicon supply chain used by Nvidia and Google. The company hopes that route will help it avoid component backlogs and secure manufacturing capacity at Taiwan Semiconductor Manufacturing Co.

The logic tracks with industry conditions. Analysts expect constraints in three areas to persist through 2027:

• Advanced packaging — the step that stacks and links dies into a single high-performance chip
• High-bandwidth memory (the fast memory stacked next to AI processors)
• Leading-edge production capacity at the biggest foundries

Everyone building AI silicon today competes for the same constrained inputs. Nvidia and Google, with their scale, win those allocations. A startup that structures its design around an unconventional supply route, rather than fighting for scraps of the standard one, is making a survival bet as much as a technical one.

What does this change for the AI chip market?

In the near term, not much. Tenspresso has no product, no tape-out date before late 2027 at the earliest, and 15 employees. The $500 million valuation is a wager on people and architecture, not revenue.

The signal matters more than the sum. Investors are pricing ex-Google TPU teams at half a billion dollars before those teams finish a first design. That tells you capital still sees room in AI silicon beyond Nvidia's dominance — specifically in workload-specialized chips rather than general-purpose GPUs (graphics processing units repurposed for AI).

It also marks Altimeter's interest as worth watching. The firm approaching Tenspresso unprompted, weeks after Stricker Venture Partners led the $20 million round, suggests competitive pressure among investors to get into founding-team deals early. Three months from incorporation to a $500 million valuation talk is fast even by 2023-2024 AI funding standards.

Whether the bet pays off depends on execution risks the money can't buy away: finishing a complex chip design on a two-year clock, locking foundry capacity in a tight market, and betting correctly that long-context workloads grow into a chip category of their own. Tenspresso has the founders and now, likely, the cash to try. The market will judge the design in late 2027.