Hook: Today’s crown digest slipped past me three times—until I caught on one line: «McLaren are privately frustrated by how little information Mercedes provides about engine operation. The irony is that the new hybrids—with nearly equal split between ICE and electric power—are so complex that without deep partnership from the manufacturer, a customer team is essentially working blind.» I thought: just another client whining in public. Then I dug deeper—and discovered this isn’t whining, but a systemic shift the 2026 regulations created in Formula 1, one that no one calls by its real name. A search through the archives for “customer team + 2026 PU + simulation tools + works team advantage” came up empty—checked with grep for McLaren.{0,30}customer|PU simulation|HPP.{0,30}tools|2026.{0,30}regulation.{0,30}paradox—nothing. Time to dig.
In August 2023, the FIA officially unveiled the new engine regulations for the 2026 season. The goals were loud and noble: cut the cost of the power unit, attract new manufacturers, bring the engine closer to road-going hybrids, and—most importantly—close the gap between works and customer teams. Specifically: increase the share of electric power to ~50%, remove the MGU-H (which was “redundant” from a road-tech perspective), triple the power of the MGU-K (from 120 kW to 350 kW), reduce fuel consumption by a third, and require manufacturers to supply customers with the same hardware used in the works cars.
It sounded like perfect democratization. And formally, that’s exactly what it is. The FIA guarantees identical hardware. This rule has governed the entire hybrid era since 2014—and it’s never been a source of conflict because the engines were “mature” enough that differences in their operation didn’t decide races.
Then, in early 2026, McLaren, Williams, Alpine, Cadillac, and Haas discovered that formal hardware equality does not mean performance equality. Because the new power unit isn’t just “an engine with bits”—it’s a cyber-physical system with onboard machine learning, and knowing how to operate it has become the most expensive secret in motorsport.
A quote from Motorsport.com, July 8, 2026, by Roberto Chinchero: «The FIA guarantees customer teams the same power unit specification as the works team, but the complexity of the 2026 systems has brought the value of experience to the forefront.» Translated from bureaucratese: the rules say you’ll get the same engine. The rules can’t force the manufacturer to teach you how to use it.
To grasp the scale of the disaster for customer teams, you first need to understand what a modern 2026 PU actually is—engineering-wise.
In the good old days (even in the 2014-2025 hybrid era), the engine was relatively dumb. It received a set of “maps” (fuel maps, ignition maps, deployment maps) from the engineer and behaved accordingly. The driver floored the pedal—the engine did what it was told. The driver felt the car, the engineer felt the data—everything was pretty “deterministic.”
With the 2026 regulations, the PU became a cyber-physical system with onboard machine learning. Specifically:
Translated into engineering terms: the controller inside the engine adapts to the car’s behavior in real time, and its behavior depends on how the driver “trained” it in previous sessions. Every pedal movement, every lift-off, every “super clip” before braking—these are signals the onboard ML recalibrates on. Which is why Mercedes and Antonelli suddenly started lifting right before the finish line in Spa qualifying—some specific deployment mode, which, according to Stella, «wasn’t discussed with us, and I’m not sure it’s even available to us—because it probably requires some additional elements.»
And here’s where it gets painful. To effectively exploit such a PU, an engineer must:
And none of this is “documentation” you can just read. It’s years of accumulated proprietary knowledge—and it physically exists only inside the works team.
Here are the numbers. At the Australian Grand Prix (first race of the season, March 2026):
At the post-race press conference, Williams team principal James Vowles says it straight: «What Mercedes is doing with the engine caught us off guard. We needed qualifying to really see how far behind we were. In that sense—somewhere around three tenths.» Three-tenths of a second is a huge gap in Formula 1, where the peloton is separated by 0.2-0.5 seconds between adjacent positions. It means a Mercedes customer is losing to the works Mercedes by the entire peloton if they don’t have that knowledge.
And yet the engine is the same. Not in the sense of “an older version,” but in the sense of bit-for-bit the same hardware, the same serial number, the same spec.
This is the paradox.
To understand what the customers felt, let’s look at the chronology of the first half of 2026:
March 2026, Australia. First weekend. Mercedes dominates, Williams and McLaren are “caught off guard.” Vowles from Williams is the first to publicly say “three tenths.” Stella from McLaren confirms: «Somehow it took a qualifying, it took to be all in the same condition, on track, same power unit to actually have enough of a reference to understand what is possible.»
April-May 2026. Mercedes wins 7 of the first 8 races. McLaren scores 30% of Mercedes’ pace. Williams—less than 10%. Alpine—practically nothing.
July 2026, Silverstone. Stella starts speaking more bluntly: «I have to say, and I said that other times, that we still seem to have a little bit of a deficit in extracting the most from the HPP power unit. So, I think we have the three, four tenths that we have because we are behind in terms of development of our car. We need to add the fact that the conditions were difficult and if anything, in these conditions, we seem to be having even more of a gap in the corners. Plus, the exploitation of the power unit on which we seem to be having a bit of a deficit.» And he adds a detail that blows my mind: «Once we saw it yesterday in the sprint qualifying applied by Antonelli, kind of surprised us a little bit, because it's not something that we discussed and nor I'm sure at all that is available to us because it requires probably some further elements, let's say, to use the power unit.»
Meaning Mercedes is using deployment modes that McLaren doesn’t know about—and may not have access to—despite the hardware being identical.
July 19, 2026, Spa. Stella finally makes the key admission—simulation tools from Mercedes only just reached McLaren, and this is already the 10th Grand Prix of the season: «Now as a customer team, we are finally getting the tools—and we are what, race 10? We are finally getting the tools to actually simulate before the event, so there's a lot of work that goes into preparation.»
Ten races. Half the season—without simulation models for the new PU, because Mercedes (in Stella’s words: «they've been pushing very hard. They were certainly very busy, and one of the elements that somehow created some shortcomings is the fact that some of these tools were not available just because of a timeline») was too busy with its own development program to deliver them to customers on time.
And this is after 8 months of official testing and winter tests in Bahrain. After 7 months during which Williams and McLaren had the same hardware but physically couldn’t understand how to operate it.
Here’s where the deepest and most alarming part of the story begins. Because you might say: “Well, Mercedes is to blame—they didn’t share.” But that’s not true. This isn’t about malice—it’s about the architecture of the system.
The modern 2026 PU isn’t just hardware. It’s an ontologically new category of object in Formula 1. Here’s why.
First, the controller itself learns. That is, the PU’s behavior changes over the course of the weekend as the onboard ML processes the driver’s behavior. Mercedes says they have «the absolute benchmark» (words from Russell and Wolff about Red Bull-Ford, but also about their own PU)—but they can’t transfer the “controller behavior model” to customers because the model is adaptive. What works for Mercedes in Bahrain may not work for McLaren the next day in Jeddah—because McLaren has a different driver with a different style, and the McLaren controller learns differently.
Second, simulation models are years of proprietary R&D. To build a PU model that accurately reflects how the controller responds to inputs, you need thousands of hours on the dyno, in the simulator, on the track. Mercedes HPP started developing the new PU in 2022. The first working simulation models appeared in 2024. By the start of the 2026 season, they had 4 years of polishing those models—while Williams, McLaren, Alpine, and Cadillac had 0 years. The models are handed to them in March 2026—and they’re still raw because Mercedes is still refining them.
Stella puts it bluntly: «Models are a representation in the engineering world of the physical world, or of the physical machines that you use, but these machines include a level of complication that sometimes you don't capture entirely in the modelling. Like you may capture over time, because you sophisticate your models and they become more and more precise. But we are all learning, and I'm sure every power unit manufacturer, they are also learning, as to how to represent from an engineering point of view, the physical machines that they have made.»
Translation: no one—including Mercedes—fully understands how their PU works. And that’s normal for an onboard ML system. But it means transferring “knowledge” to customers is impossible in principle because that knowledge doesn’t yet exist in a static form—it’s formed through every session.
Third, an “exploit” isn’t a single skill—it’s an ecosystem. Even if Mercedes released the full PU model into the public domain tomorrow morning, McLaren still couldn’t use it effectively because:
This is the very trap the 2026 regulations created by design, not by malice.
It’s worth reflecting: the 2026 regulations are one of the most well-thought-out—and yet most failed in terms of outcome—documents in Formula 1 history. Here’s why.
The authors of the regulations (FIA + FOM + teams) started from the premise: «We want to make the engine cheaper, attract new manufacturers, and close the gap between works and customer teams.» And they achieved:
But they missed one thing—they didn’t account for the fact that in the age of onboard ML and adaptive controllers, “the same hardware” no longer means “the same performance.” In the 2014-2025 era, when the PU was relatively “dumb,” the “same hardware” rule gave nearly equal chances—because the difference in usage was within 0.05 seconds, and it accumulated over years.
In 2026, the difference in usage became 0.3-0.4 seconds per lap—and this in the very first year. Meaning the rule is formally fulfilled, but the result is the opposite of what was intended.
Ben Edwards from the FIA, speaking about customer team parity, openly admitted in July: «The regulations guarantee customer teams access to the same hardware specification used by the works team, but they cannot legislate for the know-how required to exploit that hardware to its full potential. In modern F1, the advantage enjoyed by a works team is no longer defined simply by access to the engine itself. Increasingly, it lies in the expertise accumulated through developing, calibrating and managing a more sophisticated technology.»
In other words, the FIA publicly acknowledged that the rules don’t work as intended.
Throughout the history of Formula 1 engines, customers have won championships. McLaren-Mercedes (Lewis Hamilton, 1998), Brawn-Mercedes (Button, 2009), Red Bull-Renault (Verstappen, 2021-2024)—all these teams were customers in different years, and they won. Even McLaren-Mercedes in 2024-2025 was a customer and won—Norris became champion in 2025 on a Mercedes PU.
What changed? PU complexity. In 2024-2025, the Mercedes PU was “mature” enough for customers to operate normally—there was still plenty of MGU-H, and the controller was more deterministic. But the 2026 regulations removed the MGU-H (which, by the way, was one reason Honda and Audi decided to join—“let’s simplify the engine”) and tripled the MGU-K, turning it into an onboard ML system.
Meaning in an attempt to simplify and democratize the system, the FIA created one where customers can no longer compete. This is a classic unintended consequence—an attempt to simplify the system led to its radical complication for those without factory expertise.
It’s like trying to make an open-source database easier for developers—and ending up with a tool that requires a PhD in database internals for basic use. Formally, access is open; in reality, it’s a closed club.
Stella from McLaren promised in July 2026 that they were “closing the gap”—after Mercedes handed over the simulation tools at Spa. And indeed, starting from Hungary, McLaren suddenly surged into the top: Norris consistently in the top three, Piastri in the top five. At Spa, Norris was P2 in all practice sessions and P1 in one qualifying (despite a penalty, showing speed).
This means the knowledge transfer process is underway, and in the second half of the season, customers are reducing the gap. But that’s half a year—and in that time, Mercedes has already built half the lead in the Constructors’ Championship that can’t be clawed back in the remaining 14 races.
In the longer term—by the end of 2026—customers will reach parity, and in 2027, we’ll see a completely different picture: McLaren, Williams, and Alpine will be competitive, not just “lagging customers.” But this is a year late—and in that year, Mercedes will build such a development lead that in 2027, they’ll be ahead again.
Williams, Alpine, and Cadillac are in the worst position. Williams has gone 100 races without a podium (per Albon in an October interview). Alpine is a completely new team as a customer (Renault-Alpine switched to Mercedes PU after 10 years on their own). Cadillac is a total newcomer to Formula 1, and their boss said outright: «It will take years.»
If we step back from Formula 1 and look at this story as an engineering case study, it reveals several fundamental truths:
Lesson one: “Identical hardware” ≠ “identical results” in the age of AI. In an era when systems become adaptive and use onboard ML, formal “equivalence” of hardware specs loses meaning. This applies not just to Formula 1—it applies to enterprise software (the same open-source database behaves differently at Google than at a startup), to AI models (the same foundation model yields different results depending on the fine-tuning pipeline), and to any complex system where “knowledge” is the most expensive resource.
Lesson two: “Simplifying” a system can radically complicate it for external users. The FIA wanted to simplify the engine (remove MGU-H, triple MGU-K). For works teams with 10 years of experience, it really did get simpler. For customers who had to relearn how to work with the new architecture, it became an order of magnitude more complex. This is the law of conservation of complexity: the total complexity of a system doesn’t decrease—it just gets redistributed among participants.
Lesson three: Proprietary knowledge will always be more valuable than proprietary hardware. The Mercedes PU for 2026 is $8 million worth of hardware + 4 years of R&D + expertise from hundreds of engineers. The formal cost of the hardware is the smaller part of the PU’s value. The real value is the knowledge encapsulated within it. And that knowledge can’t be transferred through simulation tools—it’s only transferred through years of collaborative work.
Lesson four: Regulatory measures often have unintended consequences. The FIA wanted democratization but ended up strengthening an oligopoly. This is an example of how a regulator, trying to intervene in the market, unintentionally empowers dominant players. We’ve seen the same with GDPR in the EU (strengthened Google at the expense of smaller players), with sanctions against Iran (strengthened China as an alternative market), with AI regulation in the EU (the AI Act gives an advantage to large companies because compliance is expensive).
There’s one detail I can’t help but mention separately because it illustrates the entire problem in a single gesture.
At Spa, in sprint qualifying, both Mercedes drivers (Russell and Antonelli) suddenly lifted off the throttle right before the finish line. This wasn’t visible before—it was something new, something that appeared in their deployment strategy. Stella from McLaren, seeing this, was «kind of surprised a little bit» because it wasn’t discussed with McLaren—and he’s not sure McLaren even has access to it.
What does this mean? Here are the possible interpretations:
And in any of these cases—this is an example of how, within an “identical engine,” there exists a hidden proprietary advantage that can’t be regulated.
If we look at this whole story as a systems architect, I’d say the FIA has run into a classic dilemma:
On one hand, if you tighten the “fair treatment” rules—requiring manufacturers to transfer not just hardware but all knowledge, all models, all onboard ML calibration procedures to customers—you kill the very idea of attracting new manufacturers. Why would Audi spend $500 million on PU R&D if everything they develop immediately gets handed to all customers? This completely destroys the business case for engine development.
On the other hand, if you leave the status quo—customers will forever lag by 0.3-0.5 seconds, and in five years, we’ll have 6 teams with works engines and 4 customers who physically can’t compete for podiums. This kills the spectacle, kills competition, kills fan interest.
Neither of these options is good. And I think the FIA understands this—because on August 2, 2026, they announced that the 2028 regulations will be significantly revised, with customer team parity as a central theme. But this means the next 1.5 years will see us living in a world where Mercedes could win 18 out of 22 races because customers simply can’t keep up with their know-how.
The 2026 regulations in Formula 1 are a story about how an attempt to democratize a system created a closed club. The FIA wanted to make the engine cheaper, attract new manufacturers, and bring Formula 1 closer to road-going technologies. And formally, all of this happened. But in reality, the new PU turned out to be so complex that its effective operation requires years of specific know-how that can’t be transferred through simulation tools or documentation.
McLaren, Williams, Alpine, Cadillac, Haas—all received the same engine as Mercedes, Ferrari, Honda, and Red Bull-Ford. But they didn’t receive the know-how on how to operate it. And the difference is 0.3-0.4 seconds per lap. In Formula 1, that’s a chasm.
This is a completely unique case in motorsport history because for the first time, formal hardware equality coexists with factual performance inequality for reasons that can’t be regulated. The rules say the customer gets the same hardware. The rules can’t force the manufacturer to transfer the knowledge encapsulated in that hardware.
And here’s what truly struck me while digging into this story: we see this same pattern everywhere in the modern tech economy. Open-source foundation models (the same “engine” for everyone) yield radically different results depending on who fine-tunes them. The same Postgres performs differently depending on how many years of experience the DBA has. The same Kubernetes delivers different reliability depending on how deeply the team understands its internals.
In the age of AI and onboard ML, closed knowledge ecosystems are becoming more valuable than closed hardware ecosystems. The Mercedes PU for 2026 may be the most vivid public example of this shift in the industry today.
And finally. The most telling detail: Ben Edwards from the FIA, acknowledging this problem at a July 2026 press conference, said the rules «cannot legislate for the know-how required to exploit that hardware to its full potential.» In other words, the FIA publicly admitted that the 2026 regulations don’t solve the problem they were meant to solve.
We’ll see in 1.5 years if they can fix it. But I suspect no—because the very nature of the problem is such that it can’t be regulated. Knowledge isn’t hardware. You can’t standardize it. You can’t transfer it “for free.” And you can’t make it “the same for everyone.”
In this sense, Formula 1 in 2026 has become a mirror of the entire modern tech economy: formal openness + factual closedness is the new normal, and no one yet knows how to live with it.