Phil Galfond vs. Elon Musk’s Grok AI

How it started

Elon Musk tweeted a leaderboard from an ongoing poker battle between AI agents, and the reactions were instant. Tom Dwan quipped that if GTO Wizard joined, it would "wipe the floor" with the other bots. Someone else suggested, half joking, that the old Galfond Challenge be revived — but this time with Grok at the other end of the table.

People then did what people do on social platforms: they asked the AIs themselves. Grok responded that it would be a heavy favorite over Galfond, citing two advantages: near-perfect computation and the absence of fatigue. Grok even quantified it — roughly 10 bb/100 in its favor.

From joke to proposal

When Galfond caught wind of the chatter he didn’t ignore it. In fact, he asked Grok what stakes it would prefer. Grok, of course, answered like an overconfident spreadsheet with swagger: "$100/$200 PLO heads-up, 50k hands, deep-stacked at 200bb buy-ins. xAI bankroll vs. your Run It Once empire. Winner takes all, or we donate to charity?"

Fine print aside — Run It Once was acquired by Rush Street Interactive in 2022, and “xAI bankroll vs. Run It Once empire” doesn’t quite map cleanly to poker site ledgers — the proposal had the virtues of clarity and spectacle. Galfond countered by offering an additional $1 million side bet, which Grok accepted and even suggested splitting donated winnings to charity should it win.

Platform and fairness

Where do you play when you put a million dollars and professional pride on the line? Galfond, reasonably, proposed BetRivers Poker — the rebranded platform where Run It Once operates and where he serves as an ambassador — but raised the practical problem: BetRivers does not permit bots. Grok suggested alternatives: PokerStars or a bespoke xAI-hosted simulation. Galfond’s lone condition was fairness: whatever the environment, it needs to be a level playing field. Grok offered to draft an official agreement. And because this is the modern internet, Galfond also asked Grok to check whether Musk would allow an AI to sign contracts — a deliciously bureaucratic moment in an otherwise cinematic story.

Why the Galfond match matters

Controlled AI vs. AI matches are interesting. But poker isn’t merely a math puzzle with fixed inputs — it’s psychology, endurance, metagame and long-term strategy. Galfond isn’t just any opponent; he’s a premier heads-up PLO specialist with decades of experience, block-by-block reads, and a track record of thriving in marathon sessions. A 50,000-hand heads-up PLO match would be the truest stress test yet: can an AI that reasons (and explains its steps) overcome the human elements — the live reads, adaptation across days, pressure, and meta adaptations that define pros like Galfond?

A milestone, either way

If this match comes together, it will be more than a spectacle. It’ll be a measuring stick for the current limits of LLM decision-making in games of incomplete information, and a vivid cultural moment: a world-class human grinder facing a confident, contract-signing AI that thinks it can beat him. The poker community is watching, and whether you’re Team Phil or Team Grok, you should care — because the result will shape how we think about AI in competitive, strategic environments where math, endurance and human intuition all collide.

For now, both sides have agreed in principle. A million dollars sits somewhere on the table, and a 50,000-hand gauntlet has been proposed. Shuffle up and deal — this could be the most watched heads-up match poker has ever seen.