Liberty Global Tech Ventures invests in AI chip startup Positron

Liberty Global Tech Ventures, the technology investment arm of Liberty Global, has taken a position in Positron AI as part of an oversubscribed $875 million funding round that values the Palo Alto company at $5 billion.

The round was co-led by NEA, Valor Equity Partners, Atreides Management, Andra Capital, Dylan Patel's SemiAnalysis Capital, and Jim Clark, the founder of Silicon Graphics. Additional backers include the Qatar Investment Authority, Cisco Investments, and Naver Ventures.

The valuation is the headline number. Positron has more than quadrupled its worth in roughly seven months, a pace that says as much about the current appetite for AI silicon as it does about the company itself.

How the Round Is Structured

The financing came in two tranches rather than as a single raise. The first is a $375 million Series C at a $3.5 billion pre-money valuation. The second is a Series C-1 of up to $500 million, led by NEA and Jim Clark.

Proceeds are earmarked for the final design stage of Asimov, Positron's next-generation chip.

Liberty Global has not disclosed how much it put in or what stake it received. The announcement also stops short of confirming that the transaction has closed, which means this is an announced investment rather than a completed one. Anyone trying to gauge the effect on Liberty Global shareholders has very little to work with.

What Positron Actually Builds

Founded in 2023, Positron makes hardware and software designed specifically for AI inference — the stage where a trained model actually processes inputs and generates outputs, as opposed to the training stage that consumes most of the industry's attention and most of Nvidia's revenue.

The technical bet is memory-first architecture using LPDDR5X. That choice matters more than it might sound. The dominant approach relies on high-bandwidth memory and CoWoS advanced packaging, both of which have been supply-constrained for over two years and both of which are effectively rationed by a small number of suppliers. Building around LPDDR5X sidesteps that bottleneck entirely.

There is a second practical advantage. Positron's Atlas systems install into existing data centre architecture without requiring changes to cooling infrastructure. For operators who would otherwise need to retrofit racks for liquid cooling, that removes a significant capital hurdle.

The company says it is currently deploying more than 50 racks of Atlas at Oracle Cloud Infrastructure. Other customers named include Jump Trading and Parasail.

Asimov and Titan

Asimov, the custom silicon this round is funding, is scheduled to tape out in late 2026 with production targeted for the second half of 2027.

It powers Titan, a multi-terabyte-memory inference system aimed at long-context workloads. Positron says Titan will combine four to eight Asimov chips per node, enough to serve models above 16 trillion parameters with context windows beyond 10 million tokens in a single node.

Those figures are forward-looking specifications for hardware that does not yet exist in production, and the gap between tape-out and shipping silicon is where a lot of chip startups have historically run into trouble. Second half of 2027 is a long runway in this market.

Why Liberty Global Is Here

The inference market is the argument. Compute spending on running models rather than training them is projected to reach $1.3 trillion by 2032, driven largely by hyperscaler demand, according to Bloomberg Intelligence estimates cited in the announcement.

Positron joins a growing list of AI companies Liberty Global Tech Ventures has backed, including ElevenLabs, XBOW, Legora, and Higgsfield. For a company whose core business is telecommunications and cable infrastructure across Europe, the venture arm has been building a notably concentrated AI portfolio.

Liberty Global's pitch to portfolio companies is its global technology footprint and experience scaling digital platforms, which reads as standard strategic-investor language but is not meaningless for a hardware company that will eventually need distribution and deployment partners.

Liberty Global shares were up just over 1 percent ahead of the announcement.

The Broader Context

Positron is one of a cohort of startups arguing that inference is a distinct enough workload to justify purpose-built silicon rather than repurposed training hardware. That thesis has attracted substantial capital across several companies, and the sector has yet to produce a decisive answer about whether specialised inference chips can win meaningful share against general-purpose GPUs.

What makes Positron's position interesting is less the chip and more the supply chain reasoning. If HBM and CoWoS remain constrained through the rest of the decade, a competitive architecture that does not depend on either becomes valuable independent of raw performance comparisons.

Whether that holds depends on how Asimov performs when it arrives, which is still more than a year away.