Positron Raises $875 Million at $5 Billion Valuation for Asimov Chip
AI-chip startup Positron has raised $875 million at a reported $5 billion valuation, five times the valuation attached to its February financing. The company has a shipping product rather than a slide deck: roughly 50 Atlas racks have gone to customers including Oracle, Jump Trading and Parasail, according to the Wall Street Journal.
The new money changes the underwriting. Financing is now a weaker explanation for failure. Tape-out, manufacturing, software, customer qualification and the working capital of a 2027 Asimov launch have a much larger capital cushion.
What the headline does not disclose cleanly is whether the $5 billion mark is pre-money or post-money. That is not cosmetic. Assuming the $875 million is entirely primary equity, a $5 billion post-money valuation would give the new money an economic share of about 17.5%; a $5 billion pre-money valuation implies a $5.875 billion post-money value and about 14.9%. Security terms, secondaries or structured rights could change both numbers. The round therefore establishes the capital raised and the valuation reference, not an exact ownership transfer.
Positron says Asimov will carry 864GB to 2.3TB of memory per chip, draw about 400 watts and deliver five times the tokens per dollar and per watt of Nvidia's Rubin generation. Those are vendor claims about a product due in 2027. Atlas deployments prove that customers are willing to operate Positron hardware; they do not yet prove Asimov yields, sustained inference economics or software reliability.
Memory is precisely where Positron is attacking a real bottleneck. Large-model inference can be constrained by memory capacity and bandwidth as much as raw arithmetic throughput. If customers can keep more model state close to compute and reduce the number of accelerators or servers needed for a workload, a memory-heavy architecture can create value without beating Nvidia on every benchmark.
The $5 billion price now embeds a commercialization outcome that the 50-rack Atlas base can only partially validate. A late-stage investor does not need Positron to replace Nvidia; it needs Asimov to win enough high-value inference workloads at margins that support a multibillion-dollar equity value after another generation of manufacturing and software spend.
The first decisive evidence will be customer qualification on Asimov itself: production systems, workload-level cost, utilization and gross margin. Until that appears, the round shows that investors are willing to fund the attempt. It is not evidence that the 2027 economics have already been earned.
Sources
Wall Street Journal - Positron funding and customers · Positron - Asimov · Positron - Atlas