Anthropic and Hg reportedly plan a European AI consultancy

Anthropic and private-equity group Hg plan to back an independent consultancy that would embed AI engineers and advisers in European companies, the Financial Times reported on 23 October. London and Berlin were named as expected hubs, with a launch targeted before the end of 2026. The report cited people familiar with the plans; both companies declined to comment.[1]

The reported venture would move the model developer further into the work that follows a software purchase: connecting data, redesigning tasks and keeping an application operating inside a customer’s business. It would also give Hg’s software portfolio access to implementation expertise. The commercial question is how much capability customers retain once the embedded team leaves.

Implementation is the expensive middle

For a business adopting AI, a model subscription is often the easy part. The hard work is connecting authorised data, designing a workflow that survives imperfect outputs, integrating systems of record, controlling permissions and proving that results justify costs. The organisation must also train employees, handle errors and negotiate accountability across vendors. Engineers physically embedded in a company can see those operational obstacles far more clearly than a distant account team.

A model developer can bring detailed knowledge of its own systems: their tool interfaces, common failure modes and performance constraints. Embedded engineers can translate that knowledge into a workflow built around a client’s records and staff. The value would come from making the application reliable and useful in that setting, rather than from the model’s benchmark position alone.

Hg brings a second asset: a set of software companies that may become early users. In September, before the latest report, Hg said its existing relationship with Anthropic dated back to early 2024 and cited more than 1,600 live AI projects across its portfolio. It associated those programmes with over $260 million of budgeted EBITDA impact. That wording is essential: budgeted impact is not necessarily audited or realised profit, and a portfolio total does not tell us whether any one deployment succeeded.[2]

The buyer’s comparison still matters

Suppose a European insurer wants claims staff to retrieve policy clauses faster. An implementation partner backed by one model maker may reasonably propose that model first. But a properly designed procurement should also test whether retrieval with conventional search, a smaller model, or a competitor works as well at lower cost. The buyer must decide how that comparison is conducted, by whom, with which data, and what happens if the partner's preferred technology loses.

That hypothetical insurer would need a way to compare complete workflows. Retrieval accuracy, handling time, the cost of checking answers and the ability to escalate uncertain claims are more informative than a generic model score. A conventional search system or a different model should remain a plausible outcome of the comparison.

Ownership of connectors, testing datasets and operational records also affects what the client has bought. If those assets can be exported and understood by its own staff, the business can change a component without rebuilding the entire process. If the implementation depends on proprietary tools and continuing specialist access, the recurring service cost becomes part of the investment.

Consultancy earnings and model revenue might reinforce each other. A deployment team that makes a model indispensable increases recurring usage, while access to customers helps the lab discover valuable product features. That can create a productive feedback loop. It can also create lock-in if models, orchestration, audit trails and professional services are bundled so tightly that customers cannot change one without replacing the rest.

Hg’s September announcement and the October consultancy report should be read separately. The former documents an existing commercial collaboration and portfolio ambitions; the latter describes a proposed new organisation. Its actual staffing, customer agreements and delivery record will become clearer only if the reported launch proceeds.[1][2]

For customers, the strongest outcome would be a workflow that continues to meet agreed performance and cost measures after handover. That would show that the consultancy had transferred operating capability as well as installed a model. The proposed venture’s success can be assessed at that level, without assuming that backing from a model maker either guarantees expertise or disqualifies its advice.

Report an error or suggest a correction