From a registered office in Pilgrim Cottage, Chapel Allerton, in the Somerset town of Axbridge, a one-person company is making a large bet: that the next generation of enterprise AI buyers will not want to procure models through a hyperscaler's catalog, and will instead pay for them through a decentralized marketplace where the infrastructure itself is distributed across independent operators [GOV.UK]. The company is Zenithon AI Ltd, incorporated in 2025 under UK company number 16578992, and its founder, Alex Higginbottom, is building it solo [GOV.UK; PitchBook].
The pitch, as laid out on the company's site and in its whitepaper FAQ, is twofold. First, an AI marketplace where buyers can discover, purchase, and deploy models and tools. Second, a decentralized physical infrastructure layer (the industry shorthand is DePIN) that runs the compute and storage behind those models, with privacy and transparency presented as the core selling points [Zenithon AI website; Zenithon AI whitepaper FAQ]. In the company's own framing, the architecture itself is the product feature: "Ensures privacy, security, and transparency in all interactions by leveraging a decentralized architecture" [Zenithon AI website].
The bet
Strip away the framing and Zenithon is trying to occupy a specific seam in the AI stack. The hyperscaler model catalogs (AWS Bedrock, Azure AI Foundry, Google Vertex) are tightly coupled to their underlying clouds. Hugging Face is the dominant open hub for model discovery but routes most serious deployment back to centralized inference providers. Zenithon's wager is that a meaningful subset of buyers, likely those with regulatory exposure, sovereignty requirements, or ideological preference for non-hyperscaler infrastructure, will pay to transact and deploy on a marketplace whose compute layer is not owned by Amazon, Microsoft, or Google [Zenithon AI whitepaper FAQ].
The ideal customer profile this configuration points toward is narrow and worth naming clearly: mid-market European software teams and regulated-industry buyers (financial services, healthcare, public sector adjacent) who already have a privacy or data-residency mandate, who are comfortable holding tokens or transacting on-chain, and whose procurement teams will accept a non-traditional vendor for a model-layer purchase under roughly £50,000. That is a real population in the UK and EU under GDPR and the incoming AI Act, but it is not a mass market on day one.
Why it could be interesting
Two tailwinds give the thesis weight. The first is the steady drumbeat of European regulatory pressure on data residency and model provenance, which has already pushed buyers to ask harder questions about where inference physically happens. The second is the maturation of DePIN as a category, with compute-focused projects giving Zenithon a credible technical pattern to build on rather than having to invent the primitives from scratch.
If execution holds, the upside case is that Zenithon becomes the default transaction layer for a slice of European AI buyers who actively do not want to route through a US hyperscaler. That is a defensible niche, and niches in infrastructure historically compound. A marketplace that successfully matches even a few hundred model publishers with a few thousand buyers, taking a transaction fee on each deployment, has a path to a real business without needing to displace anyone at the top of the stack.
The team and traction
Zenithon is, today, Alex Higginbottom. Higginbottom's LinkedIn lists him as building Zenithon AI after dropping out of a PhD in AI for complex systems at the University of Manchester [LinkedIn]. That academic grounding is relevant to the model-layer thesis. What the public record does not yet show is a prior enterprise sales leadership role or a previous marketplace exit, both of which tend to matter for a company whose success depends on landing the two sides of a two-sided market in parallel.
The company is registered at Pilgrim Cottage, Chapel Allerton, Axbridge, BS26 2PP, with Companies House filings dating to 2025 [GOV.UK; NorthData]. PitchBook lists Zenithon at pre-seed stage [PitchBook]. No funding rounds, customer logos, or revenue figures have been disclosed in the captured sources, and no accelerator affiliation is confirmed.
| Data point | Value | Source |
|---|---|---|
| Founded | 2025 | PitchBook |
| HQ | Axbridge, Somerset, UK | GOV.UK |
| Companies House number | 16578992 | GOV.UK |
| Stage | Pre-seed | PitchBook |
| Founding team size | 1 | LinkedIn; GOV.UK |
The honest counterfactual
What the bears will say is straightforward: a solo founder with no disclosed funding is trying to build a two-sided marketplace in a category (decentralized AI infrastructure) where the realistic competitive set already includes Hugging Face for discovery, Bittensor and Akash for decentralized compute, and the hyperscaler model catalogs for procurement convenience. Each of those alternatives has either deep liquidity, a token economy, or an existing enterprise procurement relationship that Zenithon has to overcome from a standing start. Procurement cycles for AI infrastructure at the kind of regulated buyer Zenithon is best positioned to serve typically run six to twelve months, and the budget owner is usually a CIO or head of platform engineering who is conservative about novel vendor architectures.
What the bulls can answer, drawing on the cited material, is that Zenithon is not trying to win the general AI developer audience. The whitepaper's emphasis on privacy, transparency, and a decentralized infrastructure layer points to a deliberately narrower wedge [Zenithon AI whitepaper FAQ]. If the founder picks one regulated vertical and one geography (UK financial services is the obvious candidate given the registered address) and lands three reference customers before trying to scale the marketplace globally, the solo-founder constraint becomes less of a structural problem and more of a focusing function.
What to watch
The next twelve months will tell most of the story. Three milestones matter. First, a disclosed pre-seed round with a named lead, which would validate that at least one institutional investor underwrote the DePIN-plus-marketplace thesis. Second, the first named model publisher and the first named buyer on the platform, which is the only credible signal that a two-sided market is actually forming. Third, a second hire, ideally a go-to-market lead with prior enterprise sales experience in UK or EU regulated industries, which would address the most obvious gap in the current setup.
The ICP worth watching, again, is the mid-market European regulated buyer with a data-residency mandate and a model-layer budget under £50,000. The realistic competitive set is Hugging Face, the hyperscaler model catalogs (Bedrock, Vertex, Foundry), and the DePIN compute networks (Akash, Bittensor, Render). Zenithon's job over the next year is to show, with a named logo and a renewal, that there is a buyer who genuinely prefers its seam of the stack to any of those alternatives. Until then, the chart to ask for is not growth. It is retention.