How GM AI/ML Slashed Development Time from 15 Hours to One Minute

GM AI/ML is fundamentally compressing the timeline of vehicle development at America's largest automaker. Engineering simulations that once demanded 15 hours of computation now complete in roughly one minute. The shift is not incremental. It represents what Sterling Anderson, GM's chief product officer, calls the third epoch of engineering and design, and it is already reshaping how the company builds cars, manages crash safety, tunes HVAC systems, and even designs the factories that assemble its vehicles.

Anderson arrived at GM a little over a year ago from the self-driving startup Aurora. He co-founded it in 2016. He'd worked at Tesla. But since taking the chief product officer role, he's watched GM accelerate into methods that collapse previously siloed disciplines into a single, computationally driven workflow. Anderson described the first age of engineering: humans looked at birds, thought wings worked well, so they'd design something like them, but it was guess-and-check where they'd build and tweak prototypes until something marginally worked.

Three Epochs of Building Cars

The second epoch introduced computers into the equation. Computational fluid dynamics began informing aerodynamicists. Finite element analysis gave structural engineers new precision. But the underlying process remained stubbornly linear. "The relay race that was development remained the same," Anderson explained. Design passed the baton to aerodynamics, which passed it to structures, and whenever one team found a problem, the baton got tossed back for another round of fixes. The tools were faster, but the rhythm of iteration was not fundamentally different.

So Anderson says GM's third epoch tears down those functional walls, replacing the sequential relay with a probabilistic approach powered by AI and machine learning. They're virtualizing analyses now. Rather than running one simulation at a time and waiting for results, they can run multiple explorations in parallel.

"A collapse of those functions into a single broadly informed, largely probabilistic method for design, development and manufacturing of these assets."

That is how Anderson described it. The language is technical, but the implication is straightforward. The old constraints on how many tests you could run, how many design variations you could explore, and how long you had to wait for answers have evaporated.

15 Hours Down to One Minute

The numbers tell the story cleanly. "Our FEA runs that historically were 15 hours per run? They're now one minute," Anderson said. The compression is staggering. But the strategic logic here is clear. It is not simply about letting engineers go home earlier. When a simulation returns results in sixty seconds, the entire cadence of development changes. Teams can pump through iterations at a dramatically faster clip and run a much broader set of tests than was previously conceivable given the time available.

He's an executive director. Jason Fischer, executive director of virtual integration engineering at GM, underscored this point when discussing crash performance simulations that take about 15 to 18 hours depending on complexity to run. But they've used probabilistic methods and artificial intelligence to cut that time to about less than one minute.

But that framing misses something. The real transformation is not about saving time so someone can rest. "It's the fact that one minute later, we know what the answer is, and we can start optimizing that structural performance, and that gives us the ability to look at other things," Fischer said. Speed unlocks breadth. When answers arrive almost instantly, engineers can afford to test edge cases, explore unconventional designs, and harden vehicles against real-world variability rather than just validating against a single maneuver.

Testing Before You Build

The Virtual Proving Ground

Anderson and Fischer walked through one example, the Consumer Reports avoidance test where a vehicle at speed must swerve to miss an obstacle, and instead of connecting physical electronic control units on a test bench to verify they communicate without errors, GM now models the entire chain digitally. It's all digital. Sensors, domain controllers, and software all run together in a virtual environment that also simulates vehicle physics, so they don't need physical hardware at all.

"We actually have IP protection on how we've set this system up at General Motors where we can put together the vehicle behavior from a physics perspective," Fischer said. The result is a design space that can be explored thousands of times over, with physical parameters changed on the fly to see how control logic responds. Road conditions shift digitally. The vehicle encounters conditions that would be impractical or impossibly expensive to recreate physically. "Then you start getting a result that performs well not in this particular maneuver, but it's actually hardened against the real world," Fischer added.

Crash Tests and HVAC

It's a direct benefit. So engineers can identify weak points and reinforce them long before a physical prototype ever meets an immovable barrier at 40 miles per hour, and the same logic extends to a new vehicle's HVAC system. GM doesn't design and fine-tune individual components in isolation and then struggle to calibrate them together; instead, they can simultaneously balance airflow, refrigerant behavior, and cabin comfort. Work that once consumed months or weeks now finishes in days or hours.

So it really gives our engineers time back to dig deeper and think more creatively in their designs, as opposed to doing repetitive tasks or that iterative grind, Fischer said. One detail's worth pausing on. And the virtual environment isn't limited to the vehicle itself, because digital twins of new assembly lines are now created well in advance of any physical hardware being installed.

From NASCAR to the Moon

The reach of GM AI/ML extends well beyond passenger vehicles. Fischer noted that the company's virtual tools now touch several of its other business units:

• Motorsport programs with NASCAR and Formula One
• Energy and battery development
• Defense projects
• The lunar program

It's particularly symbiotic. Fischer said, "The beauty of these virtual tools is our collaboration with our motorsports team with NASCAR and Formula One." But they co-develop and independently advance tools based on available bandwidth and expertise, and when one pulls ahead, a monthly technology transfer session ensures both racing side and production side benefit from latest techniques.

So look beyond the numbers. The 15-hours-to-one-minute leap signals automotive engineering is being reconfigured at its foundations, and last month IBM and Dallara published research showing that AI-driven virtualizations produce data well-correlated enough with physical testing to be reliable. GM's deployment of GM AI/ML takes that principle and scales it across an enterprise that spans everything from showroom sedans to lunar rovers. It's not simply moving faster. It's being able to ask questions no one previously had the time to ask.