Behavox’s $8m Civo deal turns compute into a product line

On 7 October Behavox, the London-based compliance-software company whose customers include global banks, hedge funds, asset managers, commodity traders and insurers, announced an $8m (£6.1m) commitment with Civo, a sovereign cloud and AI provider. The agreement secures dedicated GPU capacity to run the AI models behind Behavox’s products, so that client data such as employee communications and trading records can be processed in Civo’s domestic data centers.

A day later, on 8 October, Gigatokens, a Behavox company also headquartered in London, named Civo its infrastructure partner. It sells inference on open-weights models at published per-token prices, with zero data retention by default. It said it already serves customers from existing GPU clusters, and that the capacity secured under the $8m commitment “adds scale as it comes into service”.

Behavox said the Civo commitment is “part of a wider build-out”, in which it is securing compute from other neoclouds “in response to very strong demand from regulated industries for affordable tokens and for agentic AI built for regulated entities”. That capacity, the company said, is sold through Gigatokens.

A company built on selling surveillance and controls software to banks is now committing to dedicated GPU capacity and reselling it by the token. In Behavox’s model, compute has moved from the cost of running its products to a product in its own right, and the place where that compute sits has become part of what it sells.

From cost line to price list

The Gigatokens website says Gigatokens powers the Behavox product suite, including Quantum communication surveillance and Polaris trade surveillance, and that “the same Behavox models are available per token on the direct API”. Its price list sets Behavox Quantum and Behavox Polaris at $5 per million input tokens and $25 per million output tokens, alongside open-weights models such as Kimi K3 and GLM 5.3.

The site lists four ways to buy: the direct API, aggregators such as OpenRouter, Behavox’s compliance products and Gigatokens Conductor, a platform for running AI agents. It says Behavox runs its own AI workforce on Conductor, with “over half a million agent machines to date”, and claims more than 50 billion tokens served a day.

Behavox says it serves more than 120 institutions, including 70 global banks. Per-token inference reaches a different buyer: developers and agents paying through an API, some of them via aggregators. The GPUs serving both are the same, which gives Behavox two revenue lines from one block of capacity.

The balance sheet behind the commitment changed in June 2026, when Behavox took a $175m (£132.5m) preferred equity investment from HPS Investment Partners, part of BlackRock. Behavox said at the time that as part of the transaction it had fully repaid and retired its $70m (£53m) venture-debt facility with Hercules Capital, and that it had been profitable since 2023. That position gives it room to commit to capacity before the token revenue from it comes in.

Where the data sits

Kiryl Trembovolski, co-founder and chief operating officer of Behavox, put location first: “The world’s largest regulated institutions rely on Behavox for compliance software and trust us with their most sensitive data. UK data protection law is the gold standard, and clients around the world trust it.” Civo, he said, lets Behavox process that data in-country.

The Gigatokens release names its buyers as financial institutions and enterprises that hold data “they cannot send to an AI service abroad”, listing client communications, deal information, trading records and personal data. Aleksei Shishkin, vice-president of technology operations at Gigatokens, put the supplier choice in terms of ownership and speed: “We needed dedicated GPUs in the United Kingdom from a company that owns and runs them, and we needed them quickly. Civo had the capacity available and committed to deliver it within months, not years.”

For these customers the location of the model is part of the compliance case, and Behavox now prices that location into a token.

What Civo gets, and what Behavox carries

For Civo, the deal arrived 15 days after it announced on 22 September the first of 40 planned edge data centers, in Hertfordshire, opening at 8MW and rising to 38MW on the same plot, with five further locations totaling 150MW secured. It is taking orders for Nvidia Vera Rubin capacity from $10 per hour at early-access pricing. A named customer serving financial institutions gives that sovereign pitch a reference deal. Mark Boost, Civo’s chief executive, said bringing Behavox’s AI processing onto domestic infrastructure was “proving the UK can offer a true, high-performance alternative to the dominance of overseas tech giants”.

Neither release gives the term of the $8m commitment, the number of GPUs or the names of the other neoclouds. A commitment is a fixed obligation, while token revenue moves with usage. Behavox pays Civo under its commitment and Gigatokens sells by the token, and Behavox’s margin, or its loss, sits in the gap between the two. If demand from regulated buyers runs below the capacity contracted, Behavox rather than its supplier pays for idle GPUs.

The Civo deal shows a compliance vendor treating compute as inventory: bought ahead, priced per token and sold under its own name to buyers for whom the location of processing decides the purchase. Whether that inventory earns more than it costs depends on token demand that neither release puts a number on.