Formula One drivers become stars by driving fast. What decides a good many races, though, is a data operation running beside the track — and it now leans heavily on cloud computing and artificial intelligence.
Race strategists at the Mercedes-AMG Petronas F1 Team run thousands of simulations in real time using Microsoft’s Azure cloud computing and Azure Batch, alongside AI.
"Azure Batch has revolutionised our race strategy," said Joseph McMillan, the team’s lead race strategist. "It lets us simulate every lap, pit stop and overtake scenario of the whole race in moments. To be a good strategist you have to be skilled at logical thinking, data analysis and modelling. In the present era, using artificial intelligence and AI coding tools makes it possible to get complex simulations running very quickly."
McMillan completed a PhD in mathematics at Imperial College London before joining the Mercedes F1 team, and now leads its strategy department.
Miami, 3 May
Before the start of the Miami Grand Prix, McMillan sat at a console beside the track working through multiple streams of data. Britain’s George Russell was starting fifth; Kimi Antonelli, 19, was starting from pole.
In car racing a thousandth of a second decides who wins. The race strategist’s core job is tyre changes and the timing of the decisions around them.
The undercut
Antonelli lost some control of the race early on and dropped to second. To recover the situation, McMillan immediately called an undercut — pitting before the car ahead so that fresh tyres give you the pace to take the position back.
Despite cloudy weather, he took the risk of putting Antonelli on light dry tyres for the extra speed. A specific message went over the radio intercom and the instruction was given at the precise moment.
The strategy worked completely. Antonelli returned to the lead of the race and won the trophy.
Afterwards Antonelli said the strategists had done an excellent job and that an outstanding strategy had secured a big win. McMillan said the result showed how important data, and precise strategy from the strategy team, now are in Formula One.
What is actually new here
Teams have always modelled races. What changed is the clock. An undercut is only worth calling inside a window of a few seconds, and historically a strategist had to make that judgement partly on instinct because the full scenario could not be computed in time.
Running the whole race tree in moments does not remove the judgement — someone still decided to gamble on dry tyres under cloud — but it changes what the judgement is made against. McMillan was not guessing whether the undercut worked on paper. He was deciding whether to trust the paper in the weather he could see.
Why the undercut is the hardest call in the sport
An undercut only works inside a narrow window. Pit too early and the fresh tyres have gone off before the race ends; too late and the car ahead has already covered the move. The window also moves while you are looking at it, because it depends on traffic, on how the tyres are actually degrading that afternoon, and on what the other team is about to do.
That is why simulation matters more here than almost anywhere else in the car. The decision cannot be rehearsed in advance, because the inputs do not exist until the race is running.
The transferable bit
Most readers will never call a pit stop. The pattern is worth borrowing anyway: the value of running thousands of scenarios was not that the machine chose, but that it cleared away the options that did not work, leaving a human with a smaller question and enough time to answer it.
That is a more accurate description of where these tools help than most of what is claimed for them.
