Three AI chip companies collected an extraordinary amount of private capital in the same fortnight. On October 19, 2026, Tenstorrent closed a $1.5 billion Series E at a $6.2 billion post-money valuation, more than doubling its December 2024 value. In early October, TechCrunch reported that Etched is fielding funding offers at $40–50 billion — roughly double the $21 billion mark it set just weeks earlier. And Positron AI closed an $875 million Series C and C-1 at a $5 billion valuation for its Asimov inference chip. Three data points on one curve: investors are paying incumbent-scale prices for Nvidia challengers, and the thesis behind it matters for every founder and chip investor.
The three raises, in brief
Tenstorrent (Santa Clara) raised $1.5 billion in a Series E led by Trace Capital and Innovation X, with ARK Invest, Fidelity, T. Rowe Price, Morgan Stanley, SanDisk, Tether Investments, and others participating. Led by chip architect Jim Keller (ex-Apple, AMD, Tesla), it builds RISC-V-based AI processors and Galaxy servers and licenses its processor IP. The raise follows reported June acquisition talks with Qualcomm ($8–10 billion) that ended with Keller choosing independence. Its prior round raised $693M+ in December 2024 at roughly a $2.6B valuation.
Etched (San Jose) is reportedly reviewing offers at $40–50 billion, per TechCrunch's October 5 report — roughly double the $21 billion mark it set just weeks earlier in a $700 million Jane Street-led round. Its chips do exactly one thing — run transformer models — claiming higher sustained throughput per watt and per dollar than general-purpose GPUs. The compounding is startling: about $5 billion in December 2025, $10.3 billion in July 2026 ($300 million, Sequoia-led), $21 billion in September, and now double that. The company says it holds $1 billion in customer orders, with Jane Street both an investor and an early buyer.
Positron AI closed $875 million across a Series C and C-1 at a $5 billion post-money valuation, co-led by NEA, Atreides Management, Valor Equity Partners, Andra Capital, Dylan Patel's SemiAnalysis Capital, and Jim Clark — up from about $1 billion at its Series B in February 2026. The funding covers tapeout of its second-generation Asimov chip on TSMC's N3P process at the end of 2026, with production targeted for H2 2027. Positron's thesis is "memory-first" inference: instead of scarce, expensive high-bandwidth memory, it co-locates cheaper LPDDR5X next to a custom datapath. More than 50 racks of its first-generation Atlas systems are already deployed at Oracle Cloud Infrastructure, where partner Parasail uses the capacity for inference serving.
The thesis: open weights plus inference economics
Two structural shifts underpin the enthusiasm. First, high-quality open-weight models make the intelligence layer portable: when a model can run anywhere, the binding constraint becomes inference — serving tokens to users cheaply at scale. Inference is the growth workload of the decade; training demand is lumpy and concentrated among a few frontier labs, while inference compounds with every deployed application. All three raises target inference, not training.
Second, the open instruction-set movement (RISC-V) is lowering the architectural toll. Tenstorrent builds on the open-source RISC-V architecture, which lets companies design processors without license fees and reduces dependence on proprietary ecosystems such as Arm — the same democratization logic that once made open-source software dominant. Sovereign buyers notice: Tenstorrent cites 120-plus Galaxy systems deployed in Japan as sovereign AI infrastructure. For nations wary of building AI capacity on a single American-controlled hardware stack, open architecture is a feature, not just a cost saving. Investors are pricing these challengers partly as insurance against Nvidia concentration — a call option on supply diversification — and teams do real work in the pricing: Keller's generation-defining record at Apple, AMD, and Tesla, and Nvidia-alumni density (about 15% of Etched's 400 employees). Customer order books are the other input: Etched's claimed $1 billion backlog and Positron's live Atlas racks at Oracle Cloud turn architectural promises into checkable milestones.
What it means for founders choosing compute
The practical takeaway is portability. Architect inference stacks to run on more than one vendor's hardware — abstracting away from CUDA-specific dependencies where feasible — so you can arbitrage the coming price competition. Pilot challengers at the margin while keeping Nvidia capacity as the baseline, and treat compute diversification like multi-region cloud strategy: a hedge you build before you need it.
Diligence notes for chip investors
The skeptics' case deserves airtime. Etched's narrow bet — transformers only — is spectacular if the architecture stays dominant and stranded if a new paradigm displaces it. Positron's Asimov has not yet taped out, and its H2-2027 production date means the bet is on an economics argument two years forward. Investor-customer overlap (Jane Street at Etched) can inflate both order books and valuations; verify that demand is arm's-length. None of these companies publishes comprehensive benchmarks or revenue figures — the rounds are priced on team, architecture, and momentum, a genuine risk in a capital-intensive industry with long cycles.
Angeled.com connects founders with investors.
Sources: Reuters, TechCrunch via Dealroom, Data Center Dynamics, Business Wire, TechTarget (all Oct 2026), company announcements.
Read Next
General Translation Raises $27 Million Series A Led by a16z to Build Agent-First Localization Infrastructure
LangChain Raises $300 Million Led by CapitalG at a $3 Billion Valuation to Build the Agent Stack
Mount Thor Raises $45 Million to Run Cloud Fleets of Macs for AI Agents
OpenAI Reportedly Raises $15 Billion at a $1.25 Trillion Valuation
Q-CTRL Raises $290 Million Series C Led by Salesforce Ventures, Reaching Unicorn Status
SE3 Raises $30 Million Series A Led by Accel



