Three quarters. That’s roughly how many of the bets on Kambi’s sportsbook network its AI trading system handled in the second quarter of 2026, the supplier reported in July, up from nearly half across 2025. During the World Cup, it said, markets were “priced, managed, and settled end-to-end without manual intervention.” Kambi’s partners include Rush Street Interactive and Hard Rock Digital, so this isn’t a lab demo.
Can AI set odds, then? For most of one supplier’s bets, it already does. How far that answer stretches depends on which price you mean. Kambi’s 2026 releases count what the system handled, and none of them names a model or explains how it works.
It helps to separate three ways a betting price gets set. A trader works it out with a model, an automated engine reprices it during play, or an order book produces it with no house model involved at all. Each method hands the machine a different job, and each keeps people around for a different reason.
DraftKings Says Its Odds Come From Algorithms And People
DraftKings describes the first method in its annual report for 2025. Its sportsbook, the filing says, offers “a huge spectrum of betting markets across dozens of sports,” and the odds “are set through a combination of algorithmic and manual odds making.” Bet acceptance “is also a combination of automatic and manual acceptance.” The passage sits in a risk disclosure written for investors, and it answers the question directly.
The filing is candid about the target, too. DraftKings says it sets odds “with the objective of providing an average return to the bookmaker over a large number of events,” and that its gross win percentage can still swing “event-by-event and day-by-day.” Anyone who trains models will recognize a calibration target in that sentence. The book doesn’t need to call any single game. It needs its prices to be right on average across a very large number of them.
The split between algorithm and person makes sense once you break a price into parts. A model can estimate a probability. The report says the odds are also meant to carry “a built-in theoretical margin,” which is a commercial decision, and somebody still decides how much the book will take at that number, and from whom.
That last decision is where machines start meeting machines. DraftKings limits bets customer by customer, “based on individual customer profiles and risk level to the enterprise,” and among its reasons it names “a small segment of highly sophisticated syndicates and algorithmic bettors.” So the book’s model faces other people’s models, and a customer-level limit is one of its answers.
People make mistakes too. The same filing warns that its odds compilers and risk managers “are capable of human error,” and in the first method those people still share the odds making with the model while they judge money and customers.
In-Play Markets Already Belong To The Machines
The second method takes the person out of the moment. Once a game starts, every price has to move with the play, and the markets multiply, from the next point in a tennis match to a bet builder that combines several selections from one game into a single ticket. No trading floor could price all of that by hand at that speed.
Kambi’s reports show the handover moving sport by sport. Its automated system priced and traded all of its soccer in the second quarter, the World Cup included, and it managed a significant share of the Roland Garros and Wimbledon tennis majors after early rollouts in tennis. In July, Kambi said it had finished testing the system on basketball and ice hockey and was waiting for top leagues such as the NBA and NHL to resume before taking control of those sports.
The World Cup also showed what the machine is for. Kambi’s live Bet Builder made up 22 percent of all live bets during the tournament, against 3 percent at the 2022 World Cup. That’s a correlation problem, and a hard one. Two selections from the same match aren’t independent, so the combined price has to estimate how strongly they move together, for every ticket anyone builds, while the match is running. That’s estimation at a volume only software can repeat.
DraftKings paid to add more of this method. In late 2024 it closed the purchase of Simplebet, an in-play pricing specialist, for about $36 million in cash and $45 million in stock, plus contingent payments tied to performance targets that it valued at $53.5 million when the deal closed.
Two cautions travel with claims like these. DraftKings’ filing folds generative AI, machine learning, data science and “similar technologies” into one defined term, AI, so the label can cover a logistic regression as easily as a neural network. The report also concedes that trading can get so heavy “that even automated systems would be unable to address and mitigate all risks.” Speed won machines the in-play job, and speed has a ceiling.
On Exchanges, Traders Set The Price Of An Oscar
Awards can be priced either way. Some licensed sportsbooks post house odds on the Oscars where their regulators allow it, which is the first method applied to an awards show. An event-contract exchange has no house model at all. Each contract settles at $1 when its event occurs and at zero otherwise, and Kalshi’s help pages say a user is “always trading against another member of the platform, not the exchange itself.”
An awards race also has no play-by-play for an in-play engine to read, so an exchange price moves when traders react to news. Kalshi’s Best Actor market for the 2026 Oscars left a clear record of that. Timothée Chalamet led it for months. Michael B. Jordan’s odds began climbing on March 1, after his win at the Actor Awards, and Variety reported Chalamet’s contract falling from 68 cents to 51. Kalshi’s recap has the two swapping places on March 6. Academy balloting, Variety noted, had closed on March 5. Jordan went into the ceremony at 58 percent and won.
Traders produced those moves: they read a guild result and a news cycle, then kept repricing a winner already sealed in the Academy’s ballots. Algorithms can take part as traders too, and Kalshi runs a program for “designated market makers who agree to provide consistent, two-sided liquidity.”
Those firms may earn reduced fees and adjusted position limits for meeting quoting and volume requirements. The program’s published list of covered markets runs mostly to financial, economic and sports contracts, with no awards market on it. A market maker’s model can post a quote. It can’t make anyone accept it.
That makes these contracts the cleanest test of a model quote against a crowd’s price, and the category has grown fast: Kalshi says its Oscar contracts traded well over $100 million in 2026, up from $29.6 million in 2025 and $2.3 million in 2024. Awards are one strand: GamingToday breaks entertainment odds betting into awards, box office, Rotten Tomatoes thresholds, chart positions, Survivor eliminations and mention markets, and it treats settlement wording as a risk of its own, ties on chart markets included.
Prices like these are only as informed as the people trading them, though. The favorites on Kalshi and on its rival Polymarket each won 19 of the 24 categories, The Hollywood Reporter counted, and Kalshi’s recap puts three of its misses in markets that traded under $800,000 each, against $24 million on Best Actor.
One Oscar Tie Paid Out Two Different Ways
None of the three methods writes its own rulebook, and the Oscars showed it in March. Two films, The Singers and Two People Exchanging Saliva, shared Best Live Action Short, the seventh tie in Oscars history. Kalshi had listed a tie as its own outcome, priced at about 1 percent, and about 40 buyers put $2,365 on it.
Polymarket’s rules, written before the nominations were even announced, resolved a tie in favor of the film “whose listed name comes first in alphabetical order.” The Singers won there, and Two People, the narrow favorite, went to zero. The same envelope settled one way on Kalshi and another on Polymarket, and no price, however it was set, could have changed that.
People draw the other boundaries too. Kalshi bars Academy members from trading Oscar contracts and runs surveillance for suspicious trading, Variety reported. On the sportsbook side, DraftKings’ filing describes “palpable” errors, such as inverted lines between teams, and says some US jurisdictions require a regulator’s approval before those bets can be voided. In those places, a pricing mistake becomes a regulator’s decision.
Integrity monitoring works the same way. Sportradar says its fraud detection system, “enhanced by machine learning,” flagged 56 percent more suspicious matches through AI analysis in 2025 than a year earlier, within a total of 1,116 suspicious matches across more than a million monitored events. The software raises the flag. What happens next belongs to investigators and sports bodies.
The rule underneath the third method is still being argued in court. Some judges have treated sports event contracts as federally supervised derivatives and others as gambling their states may police, so for now the status of a given contract turns on where its trader sits.
A Pending Bill Would Bar AI From Creating Microbets
The next limit on the second method may be written in Congress. The SAFE Bet Act, which Rep. Paul Tonko introduced in the House on March 11, 2025, would prohibit using AI “to create gambling products, such as microbets,” and using it to track a player’s gambling habits and offer individualized promotions, according to his office’s summary. It has sat in committee since the day it was referred.
If a version like that ever passed, the markets that depend most on machine pricing would be the ones the law stopped machines from building. The industry is heading the other way for now. Kambi’s chief executive describes the company as in transition to an AI-first organization, and DraftKings’ filing says AI “may become more important in our operations over time.”
Those questions will be settled in court calendars and committee rooms, which makes them a story for the gambling press as much as the AI press. That press has used YouTube over the years too, GamingToday’s YouTube channel included.
For anyone judging a claim that AI sets the odds, the more useful question is who can overrule the model, and on what terms.
