A San Francisco company turned high-performance computing into something you can swipe a credit card for. Boeing engineers, Samsung chip designers, and a Pentagon or two have noticed.
On any given Tuesday, somewhere in Detroit, an engineer at General Motors Motorsports is meshing a tire deformation problem. The mesh is enormous. The deadline is closer than the answer. She clicks submit. Within minutes a few hundred GPU cores spin up in a cloud region she will never visit, chew through the simulation, return the result, and disappear. The bill arrives. The race team moves on.
That dispatcher is Rescale. The company does not build chips. It does not run the cloud. It sits in the middle and translates the language of CFD, FEA, and crash sims into the language of EC2, A100s, and TPU pods. Quietly, it has become a critical piece of plumbing for the industrial economy.
Rescale calls the result a "digital engineering platform." Most of the customer base just calls it the thing that lets a 90-person aerospace startup run the same calibre of simulations as Airbus. As of 2026 the company has roughly 240 employees, headquarters at 981 Mission Street in San Francisco, more than 300 enterprise customers, and a backer list that reads like a Silicon Valley group chat: Sam Altman, Jeff Bezos, Peter Thiel, Richard Branson, Chris Dixon, Paul Graham. Nvidia put in money in 2025. Microsoft did earlier. So did Samsung, Hitachi, and Foxconn.
Pre-Rescale, the real bottleneck in product design was not creativity. It was the queue. Want to simulate the airflow over a new wing? Submit a job to the in-house cluster, wait three days for someone in IT to schedule it, then learn the cluster was at capacity and the answer would be ready next Monday. The wing might ship before the simulation finished.
This was not a small problem. According to Rescale, its current customers collectively spend more than a billion dollars a year on HPC infrastructure. Most of that money used to buy hardware that sat idle most of the week and was undersized the rest of it. Capital-intensive, capacity-mismatched, and unloved by the procurement team.
"The simulation tools were built for the 1990s. The engineers were trying to build products for the 2030s. Something had to bend."
The obvious answer, in the era of AWS, was the cloud. The non-obvious problem was that engineering software did not particularly want to live there. Licenses were node-locked. Datasets were enormous. Latency mattered. Compliance frameworks were thick. The big cloud providers were happy to sell you the raw cores; nobody was happy to make the math actually run.
Joris Poort spent the late 2000s in a cubicle at Boeing Commercial Airlines, optimizing structural elements of the 787 Dreamliner. The optimization runs took weeks. He had a mechanical engineering degree from Michigan, a master's in aerospace from Washington, and the dawning suspicion that the slowest part of building an aircraft was waiting for the computer to finish thinking.
Poort went to Harvard Business School, did a stint at McKinsey, and in 2011 talked his Boeing collaborator Adam McKenzie into co-founding Rescale. They got into Y Combinator. Paul Graham, Sam Altman, Peter Thiel, Jeff Bezos, and Richard Branson all put in early money. The pitch was not "AI." The pitch, in 2011, was that a tire-shaped CAD model should not require a tire-shaped procurement cycle.
For most of the 2010s this was a hard sell. Enterprise IT did not want a startup standing between them and AWS. So Rescale did the boring, defining work: licensing deals with the big simulation vendors, integration with every flavor of HPC scheduler, security certifications for the defense customers, multi-cloud orchestration so a job could span Azure today and GCP next Tuesday. By 2021 the company had crossed $100M in total funding. By 2022 it was the first unicorn in cloud HPC. The boring work paid.
The Rescale Platform is the front door. An engineer logs in, picks a simulation application from a catalog of more than a thousand pre-configured packages, uploads inputs, and chooses how much horsepower to throw at the problem. Underneath, Rescale finds capacity across AWS, Azure, Google, Oracle, or an on-premise cluster, provisions it, handles licensing handshakes, runs the job, ships back the result, and tears the infrastructure down. It is, depending on how you squint, Stripe for compute or Heroku for physics.
The orchestration layer. HPC, simulation software, and data, across any cloud.
An open environment for training and deploying surrogate models that accelerate physics calculations.
Launched October 2025. Turns simulation output into structured knowledge that feeds AI.
Simulation-native AI agents that handle input checks, troubleshooting, reporting, and hardware selection.
The newer layers are where the company is making its second bet. AI Physics OS lets a customer train a surrogate model on prior simulations, so the next ten thousand runs can be approximated in seconds rather than days. Data Intelligence, announced in October 2025, treats simulation results as a corpus and makes them queryable. Agentic Digital Engineering, announced in May 2026, lets agents do the dull work an engineer used to do by hand: validate inputs, pick the right cluster, write the report.
Rescale's customers include Samsung, Arm, General Motors Motorsports, SLB, and the U.S. Department of Defense. That last one is not a casual reference. Doing classified compute in a commercial cloud is a paperwork triathlon, and Rescale has spent the years getting through the paperwork. The result is a federal practice that is hard to clone from a Sand Hill garage.
The cap table reads like a who-stayed-in-touch list from early YC. Bezos, Altman, Thiel, Branson, Graham, Dixon all appear. Microsoft's M12 fund joined. Nvidia followed. So did Foxconn, Hitachi, Samsung, and Saudi Aramco's Prosperity7. Strategic money tends to be smarter money than financial money - it shows up because it intends to buy the product. Most of these investors do.
Rescale's stated mission is to empower engineers and scientists to make new discoveries by giving them on-demand access to the world's most powerful computing. That is the kind of sentence a press release writes. The interesting test is whether it survives contact with reality.
It mostly does. A grad student working on a battery chemistry problem can, in theory, log into the same platform a Fortune 50 aerospace customer uses, run an identical CFD code on identical hardware, and pay only for the minutes consumed. The expensive part of being a small lab - the cluster you cannot afford - becomes an operating expense rather than a capital one. The expensive part of being a giant company - the cluster that is always either too big or too small - becomes someone else's problem.
This is the part of the Rescale story that is easy to underrate. The company's pitch is not about democratizing in a soft sense. It is about flattening the cost curve so that two engineers with the same idea, in San Jose and in Sao Paulo, can run the same simulation in the same week. Whether that actually accelerates breakthroughs is an empirical question. The early evidence, in chip design, electric propulsion, and biologics, is that it does.
The current wave of AI is, for the most part, a wave of language. The next wave is going to need to know how a wing flexes, how a transistor heats, how a drug folds, how a battery degrades. That is physics. Physics requires simulation. Simulation requires HPC. HPC, in 2026, increasingly lives in the cloud.
Rescale spent fourteen years sitting in the exact spot where those four sentences intersect. The investor mix - Nvidia, Microsoft, Samsung, Foxconn, Hitachi - is not a coincidence. It is a forward-positioning statement: when AI agents start designing actual products, they will need a substrate that speaks both physics and cloud. Rescale would like to be the substrate.
Back in Detroit, the GM Motorsports engineer is on her fourth tire deformation run of the day. The cluster she is using does not exist when she is not using it. The bill at the end of the quarter looks like a phone bill rather than a capital expenditure request. Rescale built the dial tone. Whatever the engineer dreams up next, the compute will be there before the deadline is.