Kambi's AI trading share reaches 85 percent as Q3 revenue rises

Kambi Group's third-quarter numbers landed on a morning when most of the industry's attention was pointed at Utah courtrooms and Kalshi's $40 billion valuation. That made it easy to miss what the Swedish sportsbook technology provider was quietly reporting: 85 percent of bets across its network were priced by AI trading systems, up from a figure that would have seemed implausible three years ago.

Revenue reached €46 million for the quarter, a 23 percent increase. Adjusted EBITA more than doubled to €7.9 million. Kambi lifted its full-year guidance on the strength of those figures. The headlines are good. But the 85 percent number is the one worth sitting with.

Kambi's business is infrastructure. It prices markets for sportsbook operators who do not want to build their own trading desk. What that 85 percent share tells you is not just that Kambi's AI works — it is that the operators relying on Kambi have made a structural decision to get out of the manual pricing business almost entirely. They are renting the model along with the lines.

I have seen this pattern in DeFi liquidity provisioning, where protocol-level automation displaced human market-making in certain pools faster than the broader market expected, then created concentration risk that nobody had priced. The speed of adoption looked like efficiency. It was also a hidden single point of failure. When the model misbehaved, every pool using it misbehaved at once.

The parallel with Kambi is not exact, but it is close enough to watch. If 85 percent of bets on a network are priced by one AI system, then the idiosyncratic risk of that system — a bad input, a mispriced correlation, an edge case in a live event — is no longer Kambi's problem alone. It propagates through every operator running on the platform simultaneously. Counterparty diversity is the standard hedge against this kind of systemic exposure. Here, the trading intelligence is centralized by design.

The reporting does not say which sports, which markets, or which operators account for the remaining 15 percent that still involves human judgment. That gap matters. If the manual override is concentrated in high-volume in-play markets on major events — exactly where model errors are most expensive — then the 85 percent figure understates the effective exposure. If it is concentrated in long-tail markets where human expertise genuinely adds value, the architecture is sounder than it looks.

Kambi's guidance upgrade suggests management is not worried. Investors, apparently, were reassured. The consensus read is that the AI transition is complete, the margin improvement is real, and the company has executed well. I do not think the consensus is wrong about execution. I think it is early on systemic risk.

A network where the pricing layer is this concentrated will, at some point, produce a correlated error visible across multiple operators in the same moment. When that happens — a major tournament, a live event with an unusual result pattern, a data feed disruption — the question will not be whether Kambi's model was good on average. It will be whether operators had any independent capacity to respond. At 85 percent, most of them will not.