SkyBiometry, a subsidiary of the Lithuanian deep-learning firm Neurotechnology, announced on Monday that it is launching a portfolio of AI infrastructure services, covering hardware design for AI workloads, private AI cloud operations on bare metal, and production deployment of models on managed Kubernetes. The services are aimed at enterprises that need specialized, high-performance environments to deploy AI systems without going through a hyperscaler.

Neurotechnology, the parent, is not a new entrant to anything. The company was founded in Vilnius in 1990, which means it has been building neural networks since neural networks were a research oddity. Neurotechnology is one of three biometric providers to India’s Aadhaar program, the world’s largest biometric ID system, ran voter deduplication for Ghana’s general election and the 2023 Liberian general elections, and has deployed systems across more than 140 countries. SkyBiometry is, in effect, one of Europe’s longer-running deep-learning firms repositioning its accumulated HPC, on-prem, and high-stakes-deployment expertise as an AI infrastructure services business.

“Our focus is on removing the infrastructure bottlenecks that hinder AI innovation,” SkyBiometry CEO Mantas Kundrotas said in the announcement. This is what most enterprise AI buyers actually want from a specialist vendor, which is someone who will handle the GPU scheduling, the storage tier, the private networking, and the compliance paperwork while the customer focuses on its own models.

SkyBiometry is one of a growing number of firms that, looked at together, suggest the AI infrastructure market is specializing in a specific and interesting direction.

The broader move

Start with what is now well documented. The AI capex flood has been pulling firms out of adjacent industries for at least two years, and the biggest movements are in crypto mining and telecom.

Consider the position of a US bitcoin miner in 2024. You own land. You own a large industrial building on it. You have, probably over years, negotiated a power agreement at low-single-digit cents per kilowatt-hour, because you chose the site for exactly that reason. You have pulled fiber to the building, because mining pools care about latency. The building is already cooled for sustained high-density compute, because that is what bitcoin mining is. The hardware currently inside the building mines bitcoin. The hardware you could put inside instead would rent to a frontier-AI company for several times more revenue per megawatt-hour. The only hard parts of switching are the GPUs, which are bought, and the operations team, which is hired.

This is roughly what Core Scientific, the Texas-based miner, did. It converted 590 megawatts of owned capacity into GPU hosting for CoreWeave across a series of 12-year contracts with a projected cumulative value of $10.2 billion; CoreWeave eventually bought the company outright for around $9 billion in stock. Applied Digital, another crypto-miner-turned-AI-operator, signed 15-year leases with CoreWeave worth roughly $11 billion for 400 megawatts of AI data-center capacity in North Dakota.

Telecoms ran the same logic on a different input. If you are a long-haul fiber carrier with tens of thousands of miles of existing conduit across the United States, and a generation of carrier-grade networking expertise, and the hyperscalers need more private fiber between their AI data centers than they can build themselves in any reasonable timeframe, you sell them fiber. Lumen Technologies announced $5 billion in new AI-driven business in August 2024, anchored by a landmark partnership with Microsoft the previous month. Follow-on deals with AWS, Meta, and Google pushed Lumen’s cumulative AI-driven bookings past $8 billion.

European regional operators are a version of the same story with a sovereignty twist. Nebius, the Amsterdam-based spin-out from the former Yandex B2B cloud, signed a $27 billion five-year agreement with Meta last month and now sits on a roughly $46 billion contracted backlog with Meta and Microsoft combined. Nscale, the UK-based operator, raised $1.1 billion in the largest European Series B on record in September 2025 and signed a $24 billion contract with Microsoft for approximately 200,000 NVIDIA GB300 GPUs across European and US sites. Both are central to OpenAI’s Stargate UK and Stargate Norway programs.

These are large movements. They are largely a story about power, sited capacity, and physical networking, which are the scarce resources hyperscalers cannot replicate at speed.

The specialist tier underneath

The more interesting second-order effect is that a distinct specialist layer is forming underneath.

Hyperscalers and large neoclouds are optimized for frontier training and very large inference. They are not especially well-suited to regulated, on-premises, or domain-specialized deployments. A US law firm that wants a custom LLM tuned on its own case files does not want that model running on shared multi-tenant infrastructure. A European hospital network doing automated patient scheduling cannot move that workload to a data center on a different continent. A telecommunications operator doing automated voice-call analysis at regulatory scale wants the compute inside its own facilities.

Those are four of the verticals SkyBiometry names in its launch: legal, healthcare, telecom, and publishing. They are also the kinds of deployments that require the operator to understand the domain, not only the hardware. Neurotechnology’s Aadhaar and elections work is 35 years of on-prem, high-stakes, national-scale deployment experience, which is a different asset base from what a crypto-miner-turned-neocloud brings to the same table.

Similar positioning shows up elsewhere in the European operator tier. Scaleway and OVHcloud both pitch EU sovereignty as their primary differentiator. Ori Industries, in London, sells managed AI cloud services rather than raw capacity. The common theme is fit to a narrower customer profile that hyperscalers are unlikely to serve well.

Why this tier probably persists

The usual critique of specialist operators is that hyperscalers eventually eat the market through scale. That argument has force at the commodity end, where crypto miners and telcos are increasingly likely to be absorbed into the hyperscaler supply chain over time. CoreWeave’s acquisition of Core Scientific already demonstrated what that absorption looks like.

The specialist end is harder to consolidate. Regulated industries, sovereign deployments, and vertically tailored workflows require operators who speak the domain and whose business model is compatible with customer-specific integration work. That is a less scalable model; it is also a less commoditized one. Firms with deep adjacent expertise, which in Neurotechnology’s case is more than three decades of biometric deep learning across nearly two billion people, are reasonably well positioned to occupy it.

The AI buildout story is usually told from the top of the stack downward, in the form of hyperscaler capex numbers, frontier-lab deals, and neocloud backlogs. The specialist tier is where most enterprises will actually meet AI infrastructure in practice, and its shape will be determined less by which firms have the most gigawatts and more by which firms have the most relevant domain experience. SkyBiometry is one example. It is unlikely to be the last.

Sources: SkyBiometry, Neurotechnology press releases, Core Scientific investor releases, Applied Digital investor releases, Lumen newsroom, Nebius newsroom, Nscale press releases, CNBC, TechCrunch, Data Center Knowledge

–
By the Control Plane Editorial Team