This is more than just a financial transaction. When the British Business Bank — the UK government’s vehicle for industrial development — allocates €25 million to EQT Life Sciences’ Health Economics 3 Fund, it signals something beyond mere support for the life sciences sector. It lays a brick in a trajectory that inevitably intersects with on-premise artificial intelligence: the kind running on local hardware under full data ownership.

EQT HE3 is a multi-stage fund targeting therapeutics, medical devices, diagnostics, and healthtech. It has already co-invested alongside the Bank in companies like Phagenesis ($42 million Series D in 2023) and Cyted Health ($44 million Series B in 2025), both developing technologies where processing sensitive data – diagnostic images, clinical records, digital biomarkers – is central. And it is precisely the nature of such data that makes local AI not a whim, but a necessity.

Consider how healthtechs use Large Language Models or deep learning more broadly: assisted reporting, predictive analysis from electronic health records, drug discovery. In each scenario, the volume of personal data involved forces compliance with GDPR and digital sovereignty rules. Uploading everything to the public cloud becomes a gamble, not only for regulatory reasons but also for latency and intellectual property control. That is why the capital flow triggered by a fund like EQT HE3 — adding to the €3.7 billion raised by the platform over thirty years — is far from neutral for the infrastructure market: it directly fuels demand for on-premise servers, GPU clusters such as NVIDIA A100 or H100, and self-hosted inference solutions for LLMs.

A flywheel for on-prem AI infrastructure

The UK’s industrial strategy has singled out life sciences as one of the eight growth sectors. But scaling these companies, as British Business Bank’s Christine Hockley pointed out, requires specialist investors who can inject both capital and expertise. No less critical is the enabling technology layer: bringing a medical device or a digital diagnostics platform from clinical phase to commercialization means processing ever-growing information loads in environments that guarantee security and auditability. Here the cloud gives way to hybrid and on-premise setups.

Those following this path will face the classic self-hosting trade-offs: total cost of ownership (TCO) versus scalability, operational complexity versus data control. Inference serving frameworks for LLMs – like vLLM, TGI, or custom solutions – are making local inference more manageable even on consumer GPUs or non-specialized servers, but without an upstream capital ecosystem that funds healthtech startups, the demand for such stacks would remain hypothetical. The British public bank’s commitment proves it no longer will: every company backed by the fund becomes a potential buyer of on-prem AI hardware, driven by the need to train and serve models without exposing data outside the corporate perimeter.

Ultimately, this liquidity injection does nothing but accelerate the convergence of two previously parallel worlds: thematic venture capital and deployable AI engineering. A circle that closes to the advantage of those producing or integrating compute systems for local inference, and that serves as a warning to those still betting everything on an indistinct cloud.