Tenstorrent raises $1.5 billion to scale an open alternative in AI compute

The story

Tenstorrent has raised $1.5 billion in Series E financing to expand production of its AI-computing systems and the processor designs it licenses to other chipmakers. The company announced the round on October 19 at a $6.2 billion post-money valuation, more than twice the valuation attached to its previous financing in December 2024. Trace Capital and Innovation X led the new investment, with participation from institutions and strategic investors including ARK Invest, Fidelity Management & Research, Morgan Stanley, Sandisk, T. Rowe Price and Tether Investments.

The financing is unusually large even by AI-infrastructure standards, but the more consequential detail is where Tenstorrent says the money will go. Its plan includes volume manufacturing, global commercial expansion and large-scale deployment of Galaxy servers. Those machines combine multiple Tenstorrent accelerators to run open-weight language, coding and video-generation models. The company will also advance the AI-core and CPU intellectual property that customers can license when designing their own chips.

That combination places Tenstorrent across several layers of the computing stack. It sells individual accelerator cards and liquid-cooled workstations, packages processors into servers and superclusters, maintains an open-source software toolchain, and licenses underlying designs. Its TT-Ascalon CPUs use the RISC-V instruction-set architecture, while its Tensix cores target AI workloads. Tenstorrent identifies Samsung and LG among the companies using its licensable technology, although the funding announcement does not disclose the scope or revenue value of those agreements.

Reuters independently confirmed the size, valuation and intended production push. It reported that Tenstorrent wants to scale Galaxy servers containing dozens of AI chips for applications such as chatbots, coding assistants and video generation. Reuters also noted that RISC-V lets processor designers use an open instruction set rather than paying for access to a proprietary architecture, potentially reducing dependence on closed ecosystems. Open architecture, however, does not make the resulting chip inexpensive or simple to manufacture; advanced packaging, memory, networking and software support remain difficult capital-intensive problems.

The round follows more than $693 million raised in late 2024 at a $2 billion pre-money valuation. The new post-money figure represents a sharp step up, but it remains far below the market value of Nvidia and below the scale of spending by hyperscale cloud operators. Tenstorrent therefore does not need to displace the market leader to become significant. It needs to win enough customers that value local deployment, architectural control, supplier diversity or the option to customize silicon.

Axios reported that the company had attracted acquisition interest from Intel and Qualcomm earlier in 2026 but chose to remain independent, while declining to address what it called market speculation. The independent path matters because Tenstorrent’s licensing strategy can put its designs inside products sold under other companies’ names. It also increases the burden on the startup: the new capital must support manufacturing, developer tools, systems engineering, sales and long product cycles without the distribution network of an incumbent chip supplier.

Jim Keller, Tenstorrent’s chief executive, gives the company unusual technical credibility. His career includes work on Apple’s A4 and A5 processors, AMD’s Zen architecture and Tesla’s self-driving computer. Yet history also shows that good chip architecture is only one part of building a durable platform. Developers need reliable compilers and libraries; enterprises need supply, support and predictable performance; system builders need efficient interconnects, memory and cooling. The company’s open-source positioning will be tested most severely in those operational layers.

INNOVOX analysis: the Series E is best understood as a bid to make ownership a product feature. Tenstorrent argues that customers should be able to run models locally, deploy them on premises, inspect and modify parts of the software stack, or license compute IP for custom silicon. That proposition could appeal to governments, industrial buyers and regional cloud providers concerned about supply concentration. But openness only becomes a competitive advantage when customers can deploy it without absorbing prohibitive integration costs.

The financing also reflects a wider shift in AI hardware competition. Startups increasingly differentiate through systems and intellectual-property business models rather than by offering a single accelerator benchmark. A card can be compared on throughput and price; a licensable architecture can influence several generations of customer products. Tenstorrent is trying to do both, which expands its opportunity while multiplying execution risk.

What to watch next is evidence of conversion from capital to installed capacity. Tenstorrent has not published a manufacturing target, delivery schedule, revenue figure or customer count for Galaxy in this announcement. Named production partners, repeat orders, software adoption and additional CPU or AI-core licensees will be more informative than valuation alone. The round gives the company time and purchasing power. Whether it creates a durable open alternative will depend on how quickly that money becomes reliable systems in customers’ hands.

INNOVOX analysis

Tenstorrent is financing a model that differs from selling only finished accelerators. It offers systems, open-source software and processor IP that customers can incorporate into their own silicon. If that stack scales, the company could compete not only for accelerator purchases but also for the architecture decisions behind sovereign and customized AI infrastructure.

What to watch

Watch the pace of Galaxy manufacturing and deployments, disclosed production partners, repeat customers, software compatibility, performance on current open-weight models, and new licenses for TT-Ascalon CPUs or Tensix AI cores. Future revenue disclosure will show whether the large round is funding proven demand or an expensive push to create it.