Theses›Software & applications›Physics AI

Physics AI

Physics-AI value accrues to vertical specialists holding cross-customer data rights, not to horizontal platforms, by 2030

  • Bull
  • Not dated
  • $39B by 2035

The callLearned physics models will replace slow solvers in engineering design; the money goes to the vertical specialist that wins rights to train across customers' data.

D1D2R1R2R3C1foragainst D1D2R1R2R3C1foragainst
Fig. 1The balance. Drivers and catalysts against restraints, each block sized by its weight; catalysts hang dashed because they have not happened yet, sized by their odds. It tips toward the drivers.
  1. D1Design moves from a handful of day-long simulations to searching millions of options in seconds, which first-principles solvers cannot do and do not learn from.
  2. D2A model that learns across customers gets better with each one, so the vendor that wins cross-customer training rights compounds.
  3. R1The incumbents are consolidating, Synopsys buying Ansys is the clearest event of 2024-26, and Ansys and COMSOL are 40-year franchises.
  4. R2Fabs usually keep all their data, and nobody has built a bottom-up figure for the slice a startup can reach.
  5. R3The horizontal platforms are already priced at $2.4B.
  6. C1A leading-edge fab standardises on a third-party surrogate vendor, opening the fab process-development slice to merchant tools.end-2028
Rests onthe beliefs underneath
Supported by · confidence
What would prove itdated tests, open and resolved
202720282029today1234567predictionswatch 202720282029today1234567predictionswatch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1A physics-AI simulation company announces a round of $100M or more between 6 Nov 2026 and 31 Jan 2027.Jan 2027 · open
  2. 2Zenithon's first model and its imec and Ruhr-Bochum collaborations produce disclosed customer results, or not.Mar 2027 · open
  3. 3A horizontal platform raises again at a higher mark with no semiconductor-process deployment: the market disagreeing with the view.Jun 2027 · open
  4. 4No horizontal physics-AI platform reaches disclosed production deployment in semiconductor process development — as distinct from CFD, structural or automotive work — by end-2028.Dec 2028 · open
  5. 5At least one physics-AI vendor publicly discloses broad cross-customer training rights inside a fab or equipment-OEM contract by end-2028.Dec 2028 · open
  6. 6Synopsys ships an ML-surrogate product that displaces a named first-principles solver workload in customer production by end-2028.Dec 2028 · open
  7. 7A vendor discloses cross-customer training rights in a fab or equipment-maker contract, and a bottom-up figure for surrogate spend appears.Dec 2028 · open
Expressionslisted, private, technologies
  • NVDANVIDIA
    • cheap
     Dominant share of AI training compute.
  • SNPSSynopsys
    • full
     EDA duopoly with Cadence.
  • SIE.DESiemens Digital Industries Software (Simcenter, incl. Altair)
    • expensive
     Sells the Simcenter simulation portfolio, including the Altair tools Siemens bought in 2025, and the Simcenter PhysicsAI surrogate add-on to manufacturers; parent Siemens AG is the listed entity.
  • CDNSCadence Design Systems (incl. MSC Software, BETA CAE)
    • not screened
     Sells electromagnetic, thermal and CFD solvers and, since Feb 2026, the MSC Software structural tools bought from Hexagon, to electronics, automotive and aerospace engineers.
  • DSY.PADassault Systèmes (SIMULIA)
    • not screened
     Sells the SIMULIA structural, CFD and electromagnetic simulation tools inside its 3DEXPERIENCE platform to automotive, aerospace and industrial engineers.
  • SVCOSilvaco
    • not screened
     Sells TCAD process and device simulation, with AI surrogate modelling for fab process development (FTCO), to semiconductor fabs and chip makers.
Private · 12 private companies, 6 of them in the US.

PhysicsX · Luminary Cloud · Neural Concept · Pasteur Labs · Navier AI · Zenithon · SandBox Semiconductor · Cognichip · Vinci · SixSense · Monolith AI · Emmi AI

1 more private company tracked, not shown

Technologies
Who has already bet on itprivate rounds, from public sources

$969M raised by 10 companies, $936M of it in the last three years. Wave under way: Large rounds in the last three years; the category is funded and getting crowded.

Backed by a top-tier venture firm: PhysicsX, Luminary Cloud, Emmi AI, Monolith AI. Backed by a strategic investor: PhysicsX, Luminary Cloud.

2 more funded companies in the market.

What changedthe record, newest first
  • 3 Sep 2026A senior plasma physicist, asked about something else, volunteered that no commercial software can predict the behaviour of the plasma he works with. The equations that would describe it are approximations that stop working in his regime, so his lab writes its own code and feeds it with measurements from real experiments. That is the clearest statement of why machine-learning models are moving into physics simulation at all: not because they are more elegant, but because in some regimes there is no equation to solve, only data to learn from. It also quietly sets up the harder problem, which is that the data has to come from somewhere.
  • 1 Sep 2026There is a fast-growing category of companies replacing slow physics simulations with machine-learning models that approximate them, and a real argument inside it about whether one general model beats several specialised ones. The interesting part is that the deciding factor may not be technical at all. If the advantage comes from data collected while working on a customer's process, then the question is simply who is contractually allowed to keep that data and reuse it on the next customer — and in most industries the answer is nobody, because it is the first clause the customer's lawyers strike. That is a legal constraint dressed up as a technology race, and almost nobody prices it.
Reading
Show the reading