{
  "id": 6652115,
  "title": "Positron AI Raises $875M to Scale AI Inference Hardware",
  "url": "https://urgent.news/2026/09/10/positron-ai-raises-875m-to-scale-ai-inference-hardware",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-10T22:17:08.000Z",
  "source": {
    "name": "Ventureburn",
    "slug": "ventureburn",
    "url": "https://ventureburn.com/positron-ai-raises-875m-inference-hardware/"
  },
  "original_language": "en",
  "account": "Positron AI has successfully raised $875 million in Series C funding, bringing its valuation to an impressive $5 billion. The innovative company specializes in hardware designed to make AI inference more affordable and energy-efficient. The recent financing came in two tranches: a $375 million Series C round followed by a Series C-1 of up to $500 million. NEA, Atreides Management, and Valor Equity Partners co-led the main round, with co-investors Andra Capital and SemiAnalysis Capital also participating. Industry veteran Jim Clark led the second tranche. The new capital infusion will support Positron's growing inference business and strengthen its board, adding experienced technology investors like NEA's Forest Baskett and Atreides Management's Gavin Baker. Positron's memory-first architecture targets AI inference challenges by focusing on memory capacity and bandwidth, rather than raw compute power. Their systems utilize commodity LPDDR5X memory, which helps reduce reliance on limited high-bandwidth memory supply chains. Achieving over 90% of available memory bandwidth, Positron's systems offer top performance in tokens per dollar and tokens per watt. Their architecture supports both air-cooled and liquid-cooled data centers, allowing customers to deploy systems across various rack densities. Positron already has customers deploying its first-generation Atlas system, with over 50 Atlas racks running at Oracle Cloud Infrastructure. The company is also developing its next-generation Asimov silicon and Titan systems, targeting massive memory capacity and large-scale AI inference workloads. Asimov is expected to tape out on TSMC's N3P process by late 2026, with production planned for the second half of 2027. Each Asimov chip supports between 288GB and 2,304GB of memory, catering to increasingly demanding AI inference workloads. Titan, which combines four to eight Asimov chips, aims to support models exceeding 16 trillion parameters and context windows beyond 10 million tokens. Positron plans to build a 2MW-plus engineering data center and emulation platform to support development, testing, and production readiness. With the new financing, Positron will secure LPDDR5X supply commitments, increase production capacity, and expand its system integration. The company's go-to-market operations will also grow alongside manufacturing, helping it meet the increasing demand for inference infrastructure. CEO Mitesh Agrawal noted that Atlas deployments provided valuable customer insights, influencing the design of Asimov and Titan. The $5 billion valuation reflects investor confidence in the inference market. As Positron moves into the execution phase, the company must complete silicon development while scaling production and customer deployments, all while focusing on the economics of running AI models and reducing both energy consumption and infrastructure costs.",
  "summary": "Positron AI Raises $875M Series C Funding Positron AI has raised $875 million in Series C financing at a $5 billion valuation. The company develops hardware designed to make AI The post Positron AI Raises $875M to Scale AI Inference Hardware appeared first on Ventureburn .",
  "key_points": [
    "Positron AI raises $875M in Series C funding, valuing company at $5B",
    "NEA, Atreides Management, Valor Equity Partners co-lead main round",
    "New capital supports inference business, adds experienced investors"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}