Enterprise AI has a stalling problem, and it is not the models. Companies have poured money into chips, private clouds, and pilot projects, then watched most of them stop before they ever reached production. A London company called Deliverance AI thinks it knows why, and on Tuesday it came out of stealth to sell the answer.
Founded less than a year ago, the company said it has reached £6m in annual recurring revenue, hired more than 30 people, and signed six enterprise customers within three months of incorporation. Those are figures stated by the company, not audited accounts, and Deliverance AI has named none of the customers.
The launch was timed to London Tech Week, where the UK government has spent the week promoting a sovereign compute agenda backed by billions in private investment.
What Deliverance AI sells is what it calls an agentic operating system: a governed layer that runs AI agents inside an organisation’s own environment, routes tasks between different models, logs what those agents do, and attributes the cost back to a budget line.
The target buyers are governments, regulated industries, and large enterprises, the organisations that hold the most sensitive data and answer to the most regulators.
“Enterprise AI will not scale on trust-me promises,” said Mick McNeil, the company’s founder and chief executive. “The organisations with the most valuable data need AI that can operate inside their own environment, under their own controls, with clear governance over which models are used, where data goes, and how decisions are made.”
McNeil’s pitch rests on a distinction between infrastructure and outcome. Buying GPUs and cloud capacity, he argues, does not by itself deliver a working AI system, and the missing piece is somewhere to run agents, govern them, give them context, and hold them accountable.
He would know the spending side of that equation: before Deliverance AI, McNeil held senior roles across cloud, high-performance computing, and AI at Microsoft, Northern Data Group, and Logicalis.
The sovereignty pitch
Deliverance AI’s central argument is jurisdictional. The company is UK-founded and UK and EU-headquartered, and it says its platform can run in customer-controlled, on-premises, or fully air-gapped environments rather than on US-operated infrastructure.
The selling point is exposure: running AI through US-controlled systems, the company argues, leaves European data within reach of extra-territorial measures such as the US CLOUD Act, which can compel American providers to hand over data regardless of where it is stored.
That framing lands in the middle of a live European debate. Brussels and national governments have spent the past year wrestling with the bloc’s dependence on US cloud providers, now widely treated as a political risk rather than a purely technical one, and a wave of startups has rushed to position itself as the home-grown alternative as Europe races to resolve its sovereignty problem.
Deliverance AI joins a field that already includes the likes of Cosine, which has rallied British banks and defence firms behind a sovereign frontier model. “Sovereign” has become one of the most contested words in the sector, and not every product wearing the label keeps data out of foreign hands.
From pilots to production
Underneath the positioning, the product is an orchestration and governance layer. It provides a runtime for agents, a model-routing capability that directs each task to the model judged best on performance, cost, risk, and governance, plus audit trails, cost attribution, and an embedded engineering team.
The routing pitch doubles as a hedge: customers avoid being locked into a single model, cloud, or vendor in a market that reprices itself every few months.
The company says one customer deployment cut costs by nearly 75 per cent while also reducing the time taken to start and finish tasks. That is a single, unnamed deployment and a self-reported number, so it should be read as a claim rather than a benchmark.
The underlying problem it points at is real enough: most enterprise AI pilots stall before production, and spending is shifting fast towards AI-native, agentic tools precisely because per-seat software has stopped delivering. Governance, not raw capability, is increasingly where deployments live or die.
Built on HPE and Nvidia
Deliverance AI is not building the stack alone. It says it runs on HPE Private Cloud AI, the private cloud system HPE co-developed with Nvidia, and that the arrangement can put governed agentic workflows inside a customer’s own environment within four weeks.
On the compute side, the company has deployed on Nvidia DGX systems and the smaller DGX Spark, and is working with Nvidia NemoClaw, the chipmaker’s framework for running autonomous agents under runtime controls, on workflows in private and regulated settings.
Both partners offered supportive statements. Anthony Hills, Nvidia’s regional director for UK and Ireland enterprise and public sector, said the integration “provides a managed infrastructure designed for running governed agentic workflows within private and regulated environments”.
James Brooks, HPE’s UKIMEA hybrid solutions leader, said Deliverance AI’s emergence “shows the growing demand for sovereign AI among enterprises”, with HPE Private Cloud AI keeping customers “firmly in control of their data, models and decisioning”. Both are partners with an interest in the category growing, which is worth keeping in mind when reading the endorsements.
The harder questions sit with the numbers Deliverance AI has chosen to disclose, and the ones it has not. Six customers in three months is fast, but stealth-stage revenue is easy to state and hard to check, and a company that builds its entire proposition on transparency and auditability has so far asked the market to take its own metrics on trust.
Whether sovereignty proves a durable moat or just the season’s most marketable adjective will be settled in production, inside the air-gapped rooms where none of this is yet visible.
Get the TNW newsletter
Get the most important tech news in your inbox each week.