Vaultz Capital sells 47 bitcoin to back an unnamed GPU-memory startup at a £90M valuation
London-listed Vaultz Capital sold 47 bitcoin — about £2.96 million — to fund a £3 million stake in a stealth AI infrastructure startup whose software claims to let a single GPU do the work of a whole cluster. The startup’s identity is being kept secret.
Vaultz Capital plc, a London-listed bitcoin treasury and investment company, announced today that it has completed a £3 million investment in a newly incorporated software company it refuses to name. The funding came primarily from Vaultz’s bitcoin holdings: the company sold 47 bitcoin at an average price of £62,914, leaving it with 86 bitcoin worth about £5.41 million plus £0.4 million in cash, according to the regulatory announcement.
The unnamed startup’s claim is the interesting part. Its patented software layer sits between AI and high-performance-computing workloads and the GPUs they run on, optimising how memory and compute resources are used. Vaultz says the technology has been independently benchmarked and validated on third-party infrastructure spanning consumer and frontier-class systems.
The memory wall#
Every GPU has a hard limit on the size of the problem it can hold in its own memory. Beyond that limit, the industry’s answer has been more GPUs wired together with expensive, specialist networking. The startup says it attacks that constraint in software instead: a single GPU — including older, lower-cost hardware — can process workloads that would otherwise require a large cluster, while groups of GPUs can scale without the same networking constraints as conventional architectures.
The company says the software is hardware-agnostic and designed to slot into standard AI training and serving stacks without modifying the underlying hardware or models. Claimed results from independent benchmarking include more usable compute per GPU, cheaper and faster model fine-tuning, faster response times for long-document inference, and less energy consumed per unit of useful work. It has even been demonstrated on non-AI workloads — the announcement cites a full-resolution operational weather model run on a single desk-side device.
The terms — and the secrecy#
Vaultz subscribed for 449,101 new ordinary shares at £6.68 each, implying a £90 million pre-money and £95 million post-money valuation. The round is £5 million in total, and Vaultz holds roughly 2.3% of the enlarged share capital. Two Vaultz directors, chairman Charlie Wood and Fungai Ndoro, co-invested £140,000 on the same terms — a related-party transaction disclosed under Aquis Growth Market rules.
The one thing Vaultz won’t say is who it just bought into. “The identity of the investee company is not being disclosed at this time as certain information remains commercially sensitive,” the announcement states. So the public filing confirms the money, the valuation, and the technical claims — but not the company behind them.
Why bitcoin money is chasing GPUs#
Vaultz describes the deal as the first deployment of a portion of its treasury into a high-growth technology position that the board believes has “the potential to be transformational for shareholder value.” Chairman Charlie Wood framed the GPU angle directly: “AI requires enormous computing power, with access to GPUs … becoming an increasingly important constraint on its growth. The memory available on each GPU can also limit the size of the workload it can handle, meaning larger workloads may require multiple GPUs working together. This technology is designed to overcome that limitation, allowing larger workloads to run on existing hardware and potentially reducing the number of chips, infrastructure and energy required.”
The company’s stated addressable market is the global installed base of GPU compute across AI training and inference, cloud infrastructure, enterprise and sovereign computing, and scientific and financial modelling. Its modified net asset value per share, including the investment at cost, is now about 3.41 pence.
What to watch#
Read the announcement’s own fine print before treating the claims as fact: Vaultz cautions that the performance figures come from third-party benchmarking on specific hardware configurations and “may not be representative of performance in other environments.” An unnamed company, a £90 million valuation, and a funding source that was literally sold off the balance sheet — the disclosure is the news today. Independent verification of the technology will have to come later, and the identity of the startup will be the first thing everyone watches for.