Tenstorrent raises $1.5B at $6.2B for open-weight inference
Tenstorrent raised $1.5B at a $6.2B valuation to build RISC-V inference servers for open-weight models. What it means for your AI hosting costs.
Tenstorrent raised $1.5 billion at a $6.2 billion valuation, more than double its $2.6 billion mark from late 2024. The Toronto chip company, run by Jim Keller, builds RISC-V inference hardware and says the money goes to servers for open-weight models and video generation. That is a bet on the same thing many small businesses are betting on: that the model you run doesn't have to come from the company that sells you the chip.
What actually happened
Per TechCrunch and SiliconANGLE:
- Round: $1.5B, led by Trace Capital and Innovation X. Fidelity and T. Rowe Price also took part.
- Valuation: $6.2B, up from $2.6B in late 2024.
- Independence: In May, Bloomberg reported takeover interest from Intel and Qualcomm. Tenstorrent stayed independent.
- Products: Galaxy servers (its best seller, per SiliconANGLE), the TT-QuietBox desktop workstation, and Blackhole PCIe cards, all on RISC-V.
- Software: An open-source stack — TT-Forge compiles PyTorch models to its chips, TT-Metalium handles custom kernels.
- Customers: SiliconANGLE names Virtu Financial (on-premises trading agents) and Equinix, which offers Galaxy servers through its data centers.
- Use of funds: More Galaxy production, RISC-V research and hiring.
Neither report gives hardware prices.
Why open-weight inference chips matter for your business
Your AI bill has three layers: the model, the hosting, and the silicon under the hosting. Open-weight models already broke the first layer open — you can run DeepSeek, Qwen or Nemotron without paying a frontier lab. The silicon layer is still mostly one vendor. Every well-funded challenger there pushes the price of an open-weight token down, even if you never touch their hardware.
What we would do with this:
- Keep your workloads model-portable. The cheapest hosting next year may run hardware you haven't heard of. That only helps you if swapping the endpoint is a config change.
- Don't buy boxes on a funding headline. No prices, few named customers. Wait for benchmarks on your workload from a host you can rent by the hour.
- Watch the colocation route. Equinix offering Galaxy servers means you may get to try this hardware without buying it.
Key takeaways
- Tenstorrent raised $1.5B at a $6.2B valuation, led by Trace Capital and Innovation X
- Money goes to RISC-V Galaxy servers for open-weight models and video generation
- Open-source software stack compiles PyTorch to its chips; no hardware prices disclosed
- More inference-chip competition pushes open-weight token prices down for everyone
- Stay model- and host-portable so you can take the cheaper option when it shows up
Running open-weight models, or want to? We build AI systems where the model and the host are swappable, so a cheaper option is a config change, not a rebuild. See our services, or run the numbers on your AI spend.
Sources: TechCrunch, SiliconANGLE.
- #tenstorrent
- #ai-chips
- #open-weight-models
- #inference
- #risc-v
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Mistral Timestral forecasts on a CPU. Your license is probably fine.
Mistral's Timestral time-series models top forecasting benchmarks, run on a CPU, and cap free use at $20M monthly revenue. Most small businesses clear it.
Read itMiniMax M3.1 Flash: $0.15 tokens. Price the whole agent loop.
MiniMax M3.1 Flash costs $0.15 in and $0.60 out per million tokens and claims 70% cheaper agent tasks. Measure cost per finished task, not per token.
Read it