The two tenths Piastri could not drive out of at Spa.
Oscar Piastri qualified two tenths of a second behind Lando Norris in Q3 at Spa-Francorchamps, and Andrea Stella told motorsport.com that almost the entire gap opened in a single place: the long run from Stavelot to the Bus Stop, where both cars were draining the last of their electrical energy. Stella's account is unusual because it does not fault the driver. He said Piastri "could have been one to two tenths faster, just if the power unit had behaved as we anticipated," and that the same explanation applied whether or not Piastri had put a wheel wrong.
The energy the car runs out of
Each 2026 power unit draws up to 350kW from its MGU-K, nearly triple the 120kW of the previous formula, while the internal combustion engine gives up a matching share of the total, as Formula 1's own regulations explainer sets out. The trade is a battery that cannot feed that motor for a whole lap. The rules cap recovered energy at 9 MJ per lap and let the battery's state of charge swing by only 4 MJ, which Raceteq's breakdown of the 2026 energy system lays out in full.
Derating is the term for what happens when that budget empties. The software cuts electrical deployment before the end of a straight to protect the battery, and the driver feels it as a car that simply stops pulling. Spa carries one of the longest sustained deployment runs on the calendar, the climb from Stavelot back toward the Bus Stop, and that is precisely the stretch where Stella placed Piastri's loss rather than in any corner.
A model that learns the lap
Stella described a power unit whose self-learning elements calibrate energy deployment from the driver's inputs and the laps just completed, a detail Speedcafe drew out of his Spa media session. Both McLarens ran identical engine settings. Piastri still arrived in the derating zone with less battery in hand than Norris, because the model had planned his energy against a different picture of the lap.
On Friday Piastri lost much of a practice session to a hydraulic leak that forced a gearbox change, and his first flying run in Q3 ran wide across gravel. Those two events fed the forward-predicting model a version of the lap that did not match his second, cleaner attempt, so the deployment map it built was optimised for the wrong run. Stella called the cause "just a minor deviation in how the power unit was operated," language that puts the fault in the software's expectations rather than in the driver's hands.
A forward-predicting controller is only as repeatable as the laps it is fed. Run the same corner twice with a slightly different approach and the model can return two different deployment maps, because it is solving against a moving estimate of the lap ahead rather than a fixed table. That is the charge sitting underneath Stella's polite phrasing: a 2026 qualifying order can be set, in part, by software that does not behave the same way twice.
Not only McLaren
Mercedes saw the same signature in its works cars. Stella said the traces from Kimi Antonelli and George Russell overlaid Piastri's in the same way, so the effect appeared in the factory Mercedes and the customer McLaren on one weekend, which points the finger at the shared power unit rather than at either chassis. Mercedes AMG High Performance Powertrains supplies both, and both teams lost time in the same phase of the lap.
Spa sits close to the worst case for the problem. The circuit offers few heavy braking events to refill the battery and several long stretches of full deployment to empty it, so a small error in the energy plan compounds instead of washing out. The Hungaroring, one week later, is nearly the opposite, a stop-start lap of short straights and near-constant braking, which is why some drivers said openly they were looking forward to it over Spa.
A grid the drivers cannot feel
Piastri put it more bluntly than his team principal did. He said qualifying grids decided by "computers behaving or misbehaving" are "a pretty crap way of going racing," a line motorsport.com carried from his Spa remarks. His frustration is not that he was slow; it is that the deficit was invisible to him while it happened and only legible in the data afterward.
Norris out-qualified a team-mate who, on Stella's telling, did nothing wrong at the wheel. That is a different kind of result from one settled by pace and chassis, and it is harder to accept for exactly the reason Piastri named: a driver cannot correct in real time for a model that has already decided how his lap will spend its energy.
The same software is arriving on the road
Production electric cars are moving toward the same class of energy manager. Researchers are training machine-learning controllers such as LSTM networks to manage battery deployment in real time, and surveys of the field describe model-predictive control and reinforcement learning as the direction of travel for EV energy management. The appeal is efficiency, because a learned controller can shave peak demand a fixed lookup table would miss. The catch is repeatability, because that same controller is optimising against a prediction rather than a rule.
The Nissan Sakura, used as a test bed for one such system, returned a 21.3 percent cut in peak battery current on a standard drive cycle in the Scientific Reports study that trained it. That is the gain the road industry is chasing. Formula 1 runs a version of the same idea under a stopwatch, and the stopwatch is what makes its non-repeatability impossible to hide.
Hungary answers the narrow question on July 26. The Hungaroring hands back the braking zones that Spa withholds, so if the deployment swings settle there, Stella's diagnosis holds and the effect is a circuit-specific one. If Piastri loses another two tenths he cannot explain, the software problem is larger than one wide lap through the Ardennes gravel.