Positron AI valuation jumps fourfold to $5bn as Liberty Global joins $875m funding round

Liberty Global Ltd. has invested in artificial intelligence chip startup Positron AI as part of an oversubscribed $875 million financing round that values the inference hardware company at $5 billion. The investment gives Liberty Global Tech Ventures exposure to one of the fastest-rising private companies in AI infrastructure, with Positron’s valuation increasing more than fourfold from approximately $1.06 billion during a February financing round. Positron is developing memory-focused processors and server systems designed specifically for AI inference, positioning itself against Nvidia and other established semiconductor suppliers as computing demand shifts increasingly from training models toward operating them at scale. The deal also expands Liberty Global’s growing AI investment portfolio, which already includes ElevenLabs, Legora, XBOW and Higgsfield.

The financing was co-led by New Enterprise Associates, Atreides Management, Valor Equity Partners, Andra Capital, SemiAnalysis Capital and technology entrepreneur Jim Clark, while Qatar Investment Authority and other investors also participated. Positron plans to use the proceeds to complete development of its next-generation Asimov silicon and increase production of its Titan inference systems after deploying more than 50 racks of its existing Atlas product at Oracle Cloud Infrastructure. The scale of the round shows how aggressively investors are funding potential alternatives to conventional GPUs as AI agents, assistants and other applications create a much larger market for inference computing.

Positron AI’s valuation has jumped from about $1 billion to $5 billion in only seven months

Positron AI’s latest financing represents a dramatic revaluation of the company in less than a year. The startup raised $230 million in February at approximately a $1.06 billion valuation, according to PitchBook data cited by Reuters, before completing the new $875 million Series C at a $5 billion post-money valuation in September. That means investors have increased Positron’s implied value by more than four times in roughly seven months while the company remains in the early stages of commercializing its next-generation hardware.

The increase reflects broader enthusiasm around AI inference rather than only Positron’s own execution. Training large models remains extremely compute-intensive, but once those models are deployed, every query, agent action and generated response requires inference hardware. As adoption expands from occasional chatbot use toward persistent AI assistants, autonomous software agents and enterprise applications, inference workloads could eventually represent a larger and more recurring source of computing demand.

Positron argues that the industry’s architecture should change accordingly. Its technology is designed around memory bandwidth and memory capacity rather than maximizing raw theoretical computing performance, based on the view that many large-model inference workloads are increasingly constrained by how quickly and efficiently data can move through memory. The company believes this approach can support larger models and longer context windows while reducing power and infrastructure requirements.

Its next-generation Asimov accelerator is expected to offer between 288 gigabytes and 2.3 terabytes of memory per chip, with a power requirement of approximately 400 watts. Positron claims the architecture can deliver as much as five times the tokens per dollar and five times the tokens per watt of Nvidia’s Rubin platform under the company’s modeled workloads, although those figures are company estimates involving products that have not yet been broadly deployed commercially.

That caveat matters because Positron is competing with considerably larger companies possessing enormous research budgets, mature software ecosystems and established customer relationships. Nvidia remains the dominant supplier of AI accelerators, while Advanced Micro Devices, Google and multiple startups including Cerebras and Etched are also pursuing inference opportunities. Positron therefore needs to prove that its architectural advantages translate from internal benchmarking into reliable large-scale customer deployments.

Oracle deployment gives Positron commercial validation before its next-generation chip arrives

Positron has already moved beyond the purely developmental stage through its Atlas inference system. The company says more than 50 Atlas racks have been deployed at Oracle Cloud Infrastructure, making Oracle one of the startup’s most important early customers and providing real-world operating data that Positron can use to refine its next generation of hardware.

Atlas uses eight Positron Archer transformer accelerators and is available commercially today. Positron claims that Atlas can deliver more than three times the performance per dollar and more than four times the performance per watt of Nvidia’s H200-based DGX system under a specified Llama 3.1 workload, while operating at approximately 2,000 watts compared with 5,900 watts for the referenced Nvidia configuration. Those comparisons come directly from Positron and should be treated as vendor benchmarks rather than independent measurements, but they illustrate the efficiency argument underlying the company’s strategy.

The more important product for the $5 billion valuation is Asimov, which is expected to enter the market in 2027. Positron plans to use four or eight Asimov chips inside its Titan system, supporting as much as 18.4 terabytes of accelerator-attached memory and models containing trillions of parameters. The company says Titan will support extremely long context windows and large numbers of simultaneously active AI models or agents without requiring some of the expensive networking and cooling infrastructure associated with conventional GPU clusters.

The $875 million financing will support the Asimov silicon tapeout and Titan production ramp, meaning a large portion of the new capital is effectively financing the expensive transition from architecture design to semiconductor manufacturing and commercial deployment. Hardware startups require substantially more capital than conventional software companies because chip design, fabrication, packaging, server manufacturing and inventory all demand significant upfront spending before revenue reaches scale.

That capital intensity helps explain why Positron has raised such a large round despite already receiving $230 million earlier in 2026. Investors are effectively betting that the AI inference market will become large enough to support additional semiconductor platforms alongside Nvidia and that customers will prioritize power efficiency and memory capacity as inference volumes increase.

Liberty Global is building an AI investment portfolio alongside its telecom and infrastructure businesses

For Liberty Global, the Positron investment fits into a broader effort to build exposure to fast-growing technology companies through Liberty Global Tech Ventures and the wider Liberty Growth platform. Liberty Growth holds stakes in roughly 70 companies and funds with a portfolio valued at approximately $3.4 billion, spanning technology, media, sports and infrastructure.

Artificial intelligence has become an increasingly visible component of that portfolio. Liberty Global Tech Ventures invested in legal AI company Legora in April, AI cybersecurity company XBOW in May and video-generation company Higgsfield in August, while its earlier investments also include voice AI company ElevenLabs. Positron adds semiconductor and infrastructure exposure, giving the portfolio participation in the physical computing layer rather than concentrating entirely on software applications.

That diversification could prove important because the economics of the AI market remain uncertain across different layers of the technology stack. Model developers are spending heavily to train increasingly capable systems, software companies are competing to commercialize those capabilities and infrastructure providers are investing hundreds of billions of dollars in chips, power and data centers. Liberty Global’s venture strategy allows the company to spread exposure across multiple segments without committing the majority of its balance sheet to any single AI thesis.

The investment also has some strategic connection to Liberty Global’s operating businesses. The company remains heavily involved in broadband, mobile and digital infrastructure across Europe, with its telecom businesses providing approximately 80 million fixed and mobile connections. Growth in AI traffic, edge computing and enterprise applications could increase demand for the networks and data infrastructure underlying those services even if Liberty Global never directly deploys Positron hardware itself.

Liberty Global has also been actively rotating capital out of mature investments. During the first half of 2026, the company reported approximately $1.2 billion of asset monetizations, including around $900 million from disposals within the Growth portfolio and a roughly $340 million asset-backed loan secured against part of its Wyre stake. Management subsequently raised its year-end corporate cash target to approximately $2 billion, giving it additional flexibility to make selective technology investments while restructuring its broader portfolio.

The Positron investment therefore should not be interpreted as a wholesale strategic shift toward semiconductor manufacturing. Liberty Global has not disclosed the amount invested, and the commitment is likely small relative to the company’s overall portfolio and telecom interests. Its significance lies more in the type of exposure being added: a privately held AI infrastructure company whose valuation could increase substantially if inference emerges as the dominant growth segment within AI computing.

Nvidia competition creates enormous upside for Positron but also raises execution risk

Positron’s valuation depends heavily on its ability to capture even a small portion of an AI accelerator market dominated by Nvidia. That creates potentially asymmetric upside because the market is so large that a credible alternative does not need to displace Nvidia entirely to build a multibillion-dollar business. At the same time, Nvidia’s CUDA software ecosystem, customer relationships and rapid product-development cycle create an exceptionally difficult competitive environment for new entrants.

The inference market may offer a more realistic point of entry than training. Customers running established models often care more about cost per token, latency, memory requirements and power consumption than achieving the maximum possible training performance, creating room for specialized architectures optimized for serving workloads.

Power has become particularly important as data-center developers confront grid constraints. If Positron can provide comparable or superior inference throughput while consuming materially less electricity, customers could deploy more computing capacity within an existing power envelope. That could turn energy efficiency from an environmental advantage into a direct economic benefit for cloud providers and AI companies.

Memory requirements create another potential opening. AI models are becoming larger while context windows expand and agentic applications require multiple models or processes to remain active simultaneously. Positron’s decision to place as much as 2.3 terabytes of memory on each Asimov accelerator directly targets that trend, while Titan systems are designed to scale total accelerator memory to 18.4 terabytes.

The challenge is that the product remains largely forward-looking. Asimov and Titan are scheduled for 2027, leaving Positron exposed to manufacturing delays, performance shortfalls and competitive responses before large-scale commercialization. Nvidia and Advanced Micro Devices are also improving memory capacity and inference performance, while hyperscalers increasingly design proprietary accelerators that reduce their dependence on external semiconductor suppliers.

The $5 billion valuation therefore embeds substantial expectations before the most important product reaches customers. The fact that Positron has deployed Atlas systems with Oracle provides meaningful validation, but the company will need much broader adoption to justify an enterprise value that has increased several billion dollars within months.

Liberty Global shares remain driven more by telecom restructuring than the Positron investment

The investment is unlikely to materially change Liberty Global’s near-term financial results because neither the size of its individual stake nor Positron’s current revenue has been disclosed. Liberty Global’s public-market valuation continues to be influenced primarily by its European telecom assets, portfolio monetization and plans surrounding businesses including VodafoneZiggo and Telenet.

Liberty Global’s Class C shares closed September 10 around $10.43 after trading in a relatively narrow range during early September. The stock remained well below its 52-week high, suggesting investors continue to assign greater weight to telecom operating performance, leverage and corporate restructuring than to individual venture investments.

That does not make the Positron investment irrelevant. Liberty Global Growth is already valued in the billions of dollars, and successful venture investments can eventually become meaningful sources of capital through sales or public listings. Management has demonstrated a willingness to monetize portfolio assets when valuations become attractive, making the appreciation of private holdings an important component of longer-term shareholder value.

Positron provides an especially interesting example because its valuation has already increased from around $1 billion to $5 billion during 2026. Liberty Global is entering at the higher valuation rather than capturing that earlier appreciation, so future returns will depend on whether Positron can grow into and beyond its newly established valuation rather than simply benefiting from continued enthusiasm around AI.

For investors, the core thesis is therefore not that one Positron investment transforms Liberty Global. The importance is that Liberty Global is deploying some of the capital generated through portfolio rotation into areas with stronger structural growth, while Positron offers exposure to one of the most capital-intensive and potentially valuable bottlenecks in the AI economy.

Key takeaways from Liberty Global’s Positron AI investment and the $5 billion inference chip valuation

• Liberty Global Tech Ventures invested in Positron AI as part of an oversubscribed $875 million financing round valuing the company at $5 billion.
• Positron’s valuation has increased more than fourfold from approximately $1.06 billion during its February 2026 financing round.
• The startup develops AI inference hardware focused on memory capacity, memory bandwidth and power efficiency rather than maximizing raw computing performance.
• Positron has already deployed more than 50 racks of its Atlas inference system at Oracle Cloud Infrastructure, providing commercial validation ahead of its next-generation hardware.
• The new funding will support the Asimov chip tapeout and production ramp for Titan systems, which are expected to reach the market in 2027.
• Asimov is designed with as much as 2.3 terabytes of memory per chip, while Titan systems could provide up to 18.4 terabytes of accelerator-attached memory.
• Positron competes in a market dominated by Nvidia and also faces competition from Advanced Micro Devices, hyperscaler-designed chips and other AI semiconductor startups.
• Liberty Global’s wider Growth portfolio contains roughly 70 companies and funds valued at approximately $3.4 billion, with AI investments including ElevenLabs, Legora, XBOW and Higgsfield.
• Liberty Global generated approximately $1.2 billion from asset monetizations during the first half of 2026 and increased its year-end corporate cash target to roughly $2 billion.
• The investment gives Liberty Global exposure to the fast-growing AI inference market, but Positron must still prove that its 2027 products can achieve large-scale adoption and justify a valuation that has risen several billion dollars within months.