Somewhere between Amazon’s warehouse shelves and the pit lane at the Monaco Grand Prix lies one of the strangest technological chains of our time. This is the story of how problems delivering books gave rise to the infrastructure now used to calculate the odds of an overtake at 300 km/h.
In the early 2000s, Amazon was a company growing faster than its own systems could digest that growth. Every new product category—electronics, clothing, home appliances—added another layer of complexity to its internal architecture, and those layers stuck together like wet newspapers. The monolithic application, where everything depended on everything else, was becoming an engineer’s nightmare: a change to the recommendations module could bring down payments. Jeff Bezos, according to people who were there, at some point sent out an internal mandate—all teams had to interact with one another exclusively through software interfaces (APIs), no exceptions, no workarounds, no calling a colleague through the wall to ask them to “tweak this bit here.” Violators faced dismissal. It sounds almost comical—to fire someone for speaking directly to the department next door—but it was precisely that severity that laid the foundation for a service architecture.
The point of the transition was decomposition: instead of one enormous application—dozens, then hundreds of autonomous services, each responsible for a single function and communicating with the rest through standardized interfaces. Data storage—a separate service. Payment processing—a separate service. Search—a separate service. Each could be scaled, fixed, and updated independently. It was a painful process that took years, and Amazon went through it not out of love for elegant architecture, but because otherwise the site simply crashed under Black Friday traffic. The problem of scaling retail, bluntly put, forced it to build a universal construction kit—and at some point, someone realized that construction kit could be sold.
2006. Amazon launches two services that look modest but turn out to be a tectonic shift. Amazon S3—data storage, as simple as a mailbox: you put in a file, get a link, pay for the volume. $0.15 per gigabyte per month—a price that seemed aggressively low at the time. At its core were principles that Werner Vogels, Amazon’s CTO, championed with a preacher’s persistence: decentralization, resilience to hardware failures (hard drives break—the system should not notice), and a reliance on cheap commodity hardware, with redundant copies of data spread across hundreds of disks. A little later came Amazon EC2—virtual servers you could rent by the hour, by the day, by the minute. No need to buy hardware and build a server room, no need to hire a team to replace burned-out power supplies at three in the morning.
One detail fascinates me personally: by the time S3 and EC2 became public products, Amazon had already been using them internally for its own retail operation for several years. These were not laboratory prototypes but production systems, tested by millions of transactions. Andy Jassy, who headed AWS and later became CEO of all Amazon, started the project as an internal initiative—a proposal to give the company’s developers a standardized set of infrastructure tools. The logic was almost domestic: if we have already built plumbing for ourselves, why not run water through it for the neighbors? The price of S3 has since fallen to around $0.02 per gigabyte—more than six times lower, reflecting both the growth in scale and Moore’s Law, which may have slowed but has not stopped.
Formula 1 is a sport in which data was currency long before it became fashionable. Telemetry, aerodynamic calculations, tire-wear simulations—all of it had existed for decades, but it lived inside the teams, in closed systems, on their own servers. The viewer at home saw the cars, heard the roar of the engines, and wondered why Hamilton had come in for a pit stop right then. The answer was somewhere in the data, but the data remained invisible.
Every modern car carries more than 300 sensors, generating over 1.1 million data points per second. Brake temperature, tire pressure, fuel consumption, throttle position, steering angles, suspension loads, hybrid power-unit parameters—a stream of information that is useless to the viewer in raw form and taxing even for an engineer. Then there is the historical database amassed over more than 65 years of the championship’s existence: race results, qualifying results, weather conditions, track layouts, the evolution of the regulations. This database is not a neat Excel file but more like an archaeological layer: data formats and collection systems changed, while the rules of racing themselves were radically rewritten.
In 2018, Formula 1 and AWS announced a partnership. Pete Samara, F1’s Director of Innovation and Digital Technology, and Mike Clayville, AWS Vice President, became the faces of this alliance. The task was framed in two parts: move the championship’s infrastructure to the cloud and—more importantly for the viewer—create analytical graphics that appear during broadcasts and explain what is happening on track in the language of numbers and probabilities.
The graphics viewers see during broadcasts under the F1 Insights powered by AWS brand are the result of a pipeline in which every element solves its own task. Telemetry from the cars arrives in real time through Amazon Kinesis—a data-stream processing service capable of receiving and routing millions of events per second. Processing takes place in AWS Lambda—serverless functions that launch on an event, perform a calculation, and disappear, requiring no constantly running server. Machine-learning models trained on historical data live in Amazon SageMaker. Storage is split between Amazon DynamoDB for live data and Amazon Glacier for archives. Video integration—the overlay of graphics onto the broadcast—runs through AWS Elemental Media Services.
The individual insights work differently. Battle Forecast calculates how many laps it will take for a chasing driver to get within attacking range, and as they close in, assigns an overtaking-difficulty rating: the model takes into account the speed differential, track characteristics, historical statistics of wheel-to-wheel battles, and the tire condition of both drivers. Predicted Pit Stop Strategy models strategy options: what happens if you pit now for hard tires, and what happens if you stay out another five laps and fit mediums. Braking Performance shows how drivers brake before a specific corner—at what distance, with what intensity, and how this changes over the course of the race as fuel burns off and the car gets lighter. Car Exploitation visualizes how fully a driver is using the car’s potential. Energy Usage and Hybrid Energy System reveal the workings of the hybrid power unit—how much energy is recovered under braking, how much is deployed under acceleration. Close to the Wall—my favorite—measures how close a car passes to the wall on street circuits like Monaco, where the gap between carbon-fiber bodywork and a concrete barrier can be a matter of centimeters. Alternative Strategy shows what would have happened if a driver had chosen another strategy—a kind of parallel race universe, calculated in real time.
All of it works like a layer cake: raw data → stream processing → model → visualization → the television viewer’s screen. The delay between an on-track event and the appearance of the graphic is seconds—and that is with twenty cars on track at any given moment, each with its own stream of more than a million data points.
Precision is needed here, because a fog of misunderstanding regularly gathers around the F1-AWS partnership. AWS works with data that Formula 1, as the championship organizer, collects and has the right to use. This includes broadcast data, timing data, and standardized telemetry available equally to all teams. AWS does not gain access to the confidential engineering data of individual teams—to Ferrari’s suspension setups, Red Bull’s aerodynamic maps, Mercedes’ strategic models. Every team is its own technological fortress, and the data inside it is guarded as jealously as fighter-jet blueprints. Some teams, incidentally, use AWS themselves for internal computing, but those are separate contracts, separate infrastructure, and one team’s data does not intersect with another’s.
AWS does not run team strategies. When the engineer on Ferrari’s pit wall decides on which lap to call Leclerc in for a pit stop, he uses his own models, his own data, his own experience. The Predicted Pit Stop Strategy graphic on the television screen is a parallel calculation based on public data, and it may or may not match the team’s decision. In essence, it is a new kind of journalistic tool: not a paddock insider tip, but a mathematical reconstruction. The difference between “AWS calculates strategy for F1” and “AWS calculates strategy about F1” is one preposition—but an abyss of meaning.
What if Amazon had not made this move toward service architecture in the early 2000s? What if Bezos’s API mandate had remained an internal memo, and S3 and EC2 had never appeared?
Cloud computing would probably have emerged anyway—Microsoft, Google, and IBM were moving in the same direction. But AWS set the pace and the standard. Without it, by the most cautious estimates, the cloud market would have lagged by three to five years. And three to five years in technology is a geological epoch. In 2018, when F1 was looking for a partner to modernize its analytics, cloud infrastructure might have been at roughly the level of 2013–2015: more expensive, less flexible, with a smaller range of specialized services such as SageMaker for machine learning.
What would that have meant for the Formula 1 viewer? Most likely—the absence of real-time analytical graphics. Or a heavily simplified version of them: static tables instead of dynamic visualizations, historical reference notes instead of probabilistic models. Battle Forecast, which assesses overtaking chances in real time, requires data-stream processing, rapid access to a trained model, and instant visualization all at once—three technological layers, each of which became a mature product precisely thanks to the scale and competition unleashed by AWS. Without that scale, F1 might have been building its own infrastructure on dedicated servers—expensive, slow, with limited ability to scale from race to race.
There is a less obvious consequence, too. A historical database spanning more than 65 years is not the sort of thing easily digitized, standardized, and made fit for machine learning. Without cloud tools for handling heterogeneous data, this archive would probably have remained a collection of scattered records—valuable to historians, useless to algorithms. Glacier stores archival data for pennies; without an equivalent service, the cost of storing more than six decades of telemetry, timing data, and video would have been prohibitive.
The strangest thing about this story is its genealogy. The problem Amazon was solving in 2002—“our online store crashes under load”—has nothing in common with the problem F1 is solving in the 2020s—“how do we show the viewer that Verstappen has a 78% chance of winning this duel?” There is no direct engineering line between them. There is an intermediate product—the cloud platform—that turned out to be universal enough to serve both a warehouse full of books and a racetrack.
Pete Samara described the task of F1 Insights as translating data into a language the viewer can understand. In a sense, it is the same task Amazon solved for shoppers: turning complexity into simplicity. Amazon’s recommendation algorithm hides tens of millions of calculations behind the button that says “You might like.” A Battle Forecast graphic hides, behind a single number on screen, a stream of a million data points per second, historical overtaking patterns, and a model trained on thousands of races.
Formula 1 continues to expand its range of insights. Every season brings new graphics, models become more accurate as data accumulates, and the cost of computing keeps falling. The partnership announced in 2018 had, by the mid-2020s, become one of AWS’s most prominent showcases—and one of the few where cloud computing is visible not to a developer but to an ordinary person sitting in front of the television on a Sunday afternoon. What comes next depends on how deeply F1 is prepared to let algorithms into its ecosystem—and how ready viewers are to trust the numbers more than their own eyes.