It is 3:14 a.m. in a cleanroom somewhere. A wafer the size of a dinner plate moves between machines that cost more than a downtown block. Nobody is awake to choose what runs next. Something else is choosing - and it is, in fact, very good at it.
The room that never sleeps
Every modern semiconductor fab is a small republic of decisions. Thousands of lots. Hundreds of tools. Tens of thousands of recipe steps. Constraints that change every shift. Cycle times measured in weeks. Capital expenditures measured in billions. Every decision is small. The sum of them is enormous.
For most of the industry's history, those decisions sat on the shoulders of human planners armed with spreadsheets and rules of thumb. The planners got good. Then the fabs got harder. The number of state variables blew past the number a human can hold in working memory roughly a generation ago. The industry has been quietly looking for a new kind of dispatcher ever since.
minds.ai is one of the answers. The company - founded in 2014, headquartered in Santa Cruz, with research offices in Amsterdam and Bangalore - builds AI agents that schedule, plan, forecast and control the floor of a wafer fab. Its product, minds.ai Maestro, runs on top of an internal engine called DeepSim. Its agents are queried millions of times a month by silicon. They do not take coffee breaks.
The bet is not subtle: that reinforcement learning, combined with hybrid simulation and a few generations of supervised models, can outperform the schedules a human team produces by hand. The early evidence is that, in narrow but lucrative slices of fab operation, it can.
A startup with three time zones and one obsession
Itzik Gilboa
CEO & CO-FOUNDER
Runs the company from California. Spends his weeks talking to fab operators about KPIs nobody outside semiconductors has heard of.
Sumit Sanyal
CO-FOUNDER & COO
Operations, growth, and the Bangalore office. The connective tissue between research and customer.
Tijmen Tieleman
CO-FOUNDER & CHIEF SCIENTIST
PhD under Geoffrey Hinton. Brought serious deep-learning credentials to a domain that historically didn't have many.
Two products, one engine
minds.ai Maestro
The customer-facing product. A suite of AI agents that handle fab scheduling, planning, forecasting and process control. Maestro improves the three KPIs that matter most to fab managers: on-time delivery, cycle time, and utilization. It does this by treating the fab as a stateful environment and itself as a player that learns the optimal policy.
DeepSim
The engine underneath. A cloud-based platform that stitches open-source AI tooling and proprietary models into a stable industrial-grade stack. It runs the large-scale simulations that train Maestro's agents. It is also sold standalone via the Microsoft Azure Marketplace.
Illustrative emphasis based on public product descriptions. Actual mix varies by customer.
The dossier in four digits
Maximize the potential of semiconductor manufacturing with AI.- minds.ai, on the company mission
From pharma to wafers
minds.ai did not start in semiconductors. It started, in 2014, as a deep-learning consultancy chasing whichever industry would take its calls. Pharmaceuticals. Automotive. General enterprise. The team applied neural networks to problems other firms wouldn't touch yet.
The pivot came when the founders realized that the same techniques worked unreasonably well on a particular class of problem - high-dimensional, slow-to-feedback, expensive-to-test - that turned out to describe the inside of a chip fab almost perfectly. By the late 2010s, minds.ai had narrowed. By 2023, when Monta Vista Capital led a $5.3M seed round and Momenta joined, the company was a focused, semiconductor-only operation with a working product.
In October 2025, GlobalFoundries announced a multi-year partnership with minds.ai. That is the kind of announcement that does not happen unless the smaller party has been quietly proving itself in a customer's plant for some time.
The partner list reads like an industry roll call: Microsoft (Azure distribution), Intel (Partner Alliance Gold), Ansys (simulation), SEMI Fab Owners Alliance (the operator club itself). Most startups gather logos. minds.ai gathered ones that buy chips.
Selected dates from a quiet company
What you can actually do with it
Schedule lots
Let an RL agent choose which lot runs next on which tool. Optimize for whichever KPI mix the fab cares about this quarter.
Forecast bottlenecks
Predict where the queue will pile up next week and re-route before it does. Hybrid simulation makes the prediction trustworthy.
Improve cycle time
Compress the weeks-long path a wafer takes through the fab. Small percentage gains translate to large dollar amounts.
A company built by readers
There is a particular flavor to a team that lists its founder's PhD advisor on the about page. minds.ai has it. The company publishes research, contributes to conferences, and treats the line between academia and engineering as porous. The Amsterdam office leans research-heavy. Santa Cruz leans commercial. Bangalore covers engineering depth. The cadence is closer to a lab than a typical SaaS vendor.
Customers like that. Fab operators do not want a vendor who will iterate a feature flag every two weeks. They want one who will still be there in five years, running the same agent against a tougher plant.
Back to the room
Return to that 3:14 a.m. cleanroom. The wafer moves. The decision that put it on that machine, at that minute, in front of that recipe, was made in milliseconds by an agent trained against a digital twin of the plant itself. Somewhere in Santa Cruz, the agent's authors are asleep. Somewhere in Amsterdam, someone is awake reviewing a chart. Somewhere in Bangalore, the engine logs another query.
The remarkable thing about minds.ai is how quiet all of this is. There is no consumer brand to launch. There is no demo to go viral. The product proves itself in a metric that shows up six weeks later on a fab manager's dashboard, in numbers that a small handful of people in the world can fully appreciate. It is, in the best sense, infrastructure - the kind of company whose work you will never see, and whose work you will absolutely benefit from the next time you buy a phone, a car, or anything else with a chip inside it.
Which is to say: the room never sleeps. Neither does the agent. Somewhere, a wafer is moving. That is the company in one sentence.