Positron Raises $875 Million to Fund Its Next AI Inference Chip

The financing funds Positron’s move from deployed Atlas systems to a planned custom chip. Its central bet is that LPDDR5X memory can reduce dependence on scarce AI-chip components.

By 2 min read
Positron Raises $875 Million to Fund Its Next AI Inference Chip
Positron Raises $875 Million to Fund Its Next AI Inference Chip

Listen to this story

The audio brief

About 1:37
0:001:37
Read transcript
Positron has raised $875 million at a $5 billion valuation to take its next AI inference processor, Asimov, from design toward manufacturing. The round was oversubscribed and co-led by NEA, Atreides Management, Valor Equity Partners, Andra Capital, SemiAnalysis Capital, and Jim Clark, with Liberty Global Tech Ventures also participating. The central bet is on memory. Instead of relying on high-bandwidth memory, or HBM, and CoWoS advanced packaging, Positron is designing around LPDDR5X, a type of memory the company says could reduce exposure to constrained AI-chip components. Positron also claims this approach can support air- or liquid-cooled data centers, and says its existing Atlas systems can be installed without cooling changes. Those bandwidth and deployment advantages have not been independently established. The financing will fund Asimov’s tapeout on TSMC’s N3P process, planned for late 2026, plus an engineering data center and emulation platform exceeding two megawatts. It will also support production, integration, and go-to-market work for Titan, Positron’s current platform. More than 50 Atlas racks are being deployed at Oracle Cloud Infrastructure, with Parasail using that capacity for inference services. Asimov is planned to offer 288 to 2,304 gigabytes of memory per chip, with Titan combining four to eight chips. The key constraint is simple: tapeout is still a design milestone, and production is not targeted until the second half of 2027.

Story brief

3 key points

Positron’s oversubscribed financing gives it capital to develop and industrialize Asimov, but the company remains years from a shipping product. The processor is scheduled for tapeout on TSMC’s N3P process in late 2026 and production in the second half of 2027. Positron is betting on LPDDR5X instead of HBM, while using the round to expand engineering infrastructure and ramp its existing Titan platform. Atlas...

  1. 01

    The round values Positron at $5 billion and was co-led by NEA, Atreides Management, Valor Equity Partners, Andra Capital, SemiAnalysis Capital, and Jim Clark.

  2. 02

    Asimov’s planned memory capacity ranges from 288GB to 2,304GB per chip; Titan would combine four to eight chips.

  3. 03

    Funding includes a more-than-2-megawatt engineering data center, an emulation platform, and Titan production, integration, and go-to-market work.

Positron has raised $875 million to carry its next AI inference chip from design toward manufacturing. The financing values the company at $5 billion and backs Asimov, a planned processor that sits at the center of Positron’s attempt to use more readily available memory for running AI models.

The round was co-led by NEA, Atreides Management, Valor Equity Partners, Andra Capital, SemiAnalysis Capital and Jim Clark. Liberty Global Tech Ventures also said it joined the oversubscribed financing as an investor.

The money is for the step before a chip ships

Positron says the capital will fund Asimov’s tapeout, an engineering data center and emulation platform exceeding 2 megawatts, and Titan’s production ramp. Tapeout is when a chip design is sent for manufacturing; it is a major milestone, but it is not yet a production processor.

Positron’s stated spending plan

  • Tape out Asimov on TSMC’s N3P manufacturing process.
  • Build an engineering data center and emulation platform.
  • Ramp Titan production, system integration and go-to-market work.

A different memory supply-chain wager

Positron’s approach is built around LPDDR5X memory rather than high-bandwidth memory, known as HBM, and CoWoS advanced packaging. The company says its systems can operate in air-cooled or liquid-cooled data centers, while its Atlas systems can be installed without cooling-system changes. Its stated bandwidth and deployment characteristics remain company claims.

The company says more than 50 racks of its first-generation Atlas system are being deployed at Oracle Cloud Infrastructure, with Parasail using that capacity for inference services. Positron also identifies Jump Trading and i3d.net as Atlas production customers. Those deployments give the company operating experience, but Asimov’s proposed specifications and schedule are still future targets.

The next system carries the larger promise

Positron says Asimov will support 288GB to 2,304GB of memory per chip. Titan is planned to combine four to eight Asimov chips into one system. The new financing gives Positron resources to pursue that design; its next material test is meeting the stated tapeout schedule and turning the planned chip into production hardware.

Sources

  1. prnewswire.comPositron AI Raises $875 Million at a $5 Billion Valuation to Bring Its Next-Generation Inference Silicon to Market
  2. libertyglobal.gcs-web.comLiberty Global Tech Ventures Invests in AI Inference Hardware and Software Company, Positron AI | Liberty Global Ltd.

Loading discussion...