· Investment Strategies · 2 sources
Positron AI’s $875M Series C: Inference Silicon Hits $5B Without HBM
Positron closed $875M at $5B on Sep 10, 2026 — NEA, Atreides, Valor, Andra, SemiAnalysis Capital, and Jim Clark co-led — after Atlas hit 50+ OCI racks and Asimov targets H2 2027 production on LPDDR5X.
By Venture Capital Tracker Editorial
Positron AI closed $875 million on September 10, 2026, at a $5 billion post-money valuation — more than 4× the February mark of about $1.06 billion after a $230 million raise (PitchBook via Reuters). The financing splits into Series C ($375M at $3.5B pre) co-led by NEA, Andra Capital, Atreides Management, Valor Equity Partners, and Dylan Patel’s SemiAnalysis Capital, plus Series C-1 (up to $500M) led by NEA and Jim Clark (company PR).
Training money bought GPUs; inference money is buying memory bandwidth — and Positron is betting LPDDR5X systems can win tokens-per-dollar without waiting in the HBM/CoWoS queue.
To put inference-specialized chips in context with the rest of the AI compute stack, see the AI infrastructure startups and funding overview, which separates equity investment from debt and infrastructure commitments.
Key facts
- Company
- Positron AI (positron.ai), Reno, NV
- Round
- $875M Series C + C-1 at $5B post
- Co-leads
- NEA, Atreides, Valor, Andra, SemiAnalysis Capital, Jim Clark
- Prior
- ~$230M at ~$1.06B (Feb 2026, PitchBook/Reuters)
- Shipping
- Atlas — 50+ racks at Oracle Cloud Infrastructure
- Named customers
- Parasail (partner), Jump Trading, i3d.net
- Next silicon
- Asimov — TSMC N3P, tapeout end-2026, production H2 2027
- System
- Titan — 4–8 Asimov chips; claims models >16T params / >10M token context in one node
NEA, Atreides, Valor, Andra, SemiAnalysis Capital, QIA, Cisco Investments, and DFJ Growth
Who uses the product — and for what job
Users: hyperscalers, neoclouds, and latency-sensitive shops that need cheap, power-efficient serving, not another training cluster.
Job: run agents/assistants at scale when memory capacity, memory bandwidth, and power — not FLOPS slogans — are the bottleneck. Company claims Atlas realizes >90% of available memory bandwidth and can sit in air- or liquid-cooled halls at varying rack densities.
Treat tokens-per-dollar / tokens-per-watt leadership as company / investor framing until third-party benchmarks print.
Why now
- Center of gravity shifted from training to inference — every agent is a serving bill.
- HBM and advanced packaging remain constrained; Positron’s pitch is to sidestep that supply chain with commodity LPDDR5X.
- Atlas is already in a hyperscaler (OCI), which turns the Series C into a manufacturing + next-tapeout raise, not a lab demo.
Why these investors — portfolio fit
| Investor | Likely fit |
|---|---|
| NEA (Forest Baskett to board) | Classic semiconductor growth; memory-first thesis vs HBM scarcity |
| SemiAnalysis Capital (Dylan Patel to board) | Inference economics diligence culture; “what hardware actually delivers” |
| Jim Clark | Systems/semiconductor pedigree (SGI / Netscape) on long-context serving |
| Atreides / Valor / Andra | Crossover + infra capital for multi-year tapeout/production ramps |
| Strategics (Cisco, Naver, HRT, VentureTech) | Distribution / workload adjacency |
Capital is earmarked for Asimov tapeout, a 2 MW+ engineering/emulation facility, and Titan production (LPDDR supply, integration, GTM).
Competitive map
| Approach | Tradeoff |
|---|---|
| Nvidia / AMD GPU serving | Ecosystem + software; HBM/power cost |
| Other inference ASICs (Groq et al.) | Different latency/throughput bets |
| Positron memory-first + LPDDR | Avoids HBM queue; must prove model perf at Titan scale |
What remains undisclosed
Revenue, gross margin, rack ASP, and independent tokens/$ benchmarks. $5B on a shipping-but-next-gen story only works if Asimov/Titan land on the H2 2027 production plan.
Follow Venture Capital Tracker on Google
Add VCT as a preferred source to see more of our venture-capital reporting in Google Search.
Editorial note: AI tools assisted with research, structure, or drafting. Venture Capital Tracker retains human editorial responsibility for factual accuracy, relevance, and source quality before publication.
