Theses›Industrial›Uncertainty-propagation compute

Uncertainty-propagation compute

Deterministic uncertainty-propagation compute: carrying probability distributions through existing scalar code, in a compiler and a co-processor, replaces Monte Carlo for low-to-mid dimensional problems

  • Bear
  • Early
  • $746M by 2035

The callRobots and vehicles will have to certify what they don't know, but no safety standard names quantified uncertainty yet, and the market today is spreadsheet add-ins.

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 restraints.
  1. D1Physical AI has to prove what it does not know, and safety regimes (UL 4600, ISO 21448, the EU AI Act) are where certified uncertainty could become required.
  2. D2Monte Carlo cannot run on a robot: thousands of runs, a different answer each time, too slow for a control loop, and nobody rewrites sensor fusion for a GPU.
  3. R1A desk-tool market with no disclosed exit above $50M: Palisade went into a PE roll-up, Dynardo to Ansys undisclosed, and even after the catalyst the winner does about $80M.
  4. R2The compiler is the moat and the easiest thing for Arm, NVIDIA, MathWorks or Ansys to absorb by shipping distribution arithmetic natively.
  5. R3Past a few tens of jointly uncertain variables the representation costs more than Monte Carlo, handing those workloads back to GPU sampling.
  6. C1A safety standard or a top-5 robotics maker requires reproducible uncertainty on every unit, making it a mandated line item.2030
Rests onthe beliefs underneath
Supported by · confidence
What would prove itdated tests, open and resolved
2027202820292030today12345predictionswatch 2027202820292030today12345predictionswatch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1The leading co-processor company's first chip samples show a measured energy or latency gain on a named robotics workload, and its £5M seed closes by Jun 2027.Dec 2027 · open
  2. 2By end-2028 no safety standard for autonomous systems names quantified uncertainty of perception or state estimates as a required safety-case artefact.Dec 2028 · open
  3. 3An autonomy safety standard names quantified uncertainty, or a merchant design-in on a 10k-unit robotics or automotive platform.Dec 2028 · open
  4. 4By end-2029 Arm, NVIDIA, Intel, MathWorks or Ansys ships distribution arithmetic as a native product feature.Dec 2029 · open
  5. 5Arm, NVIDIA, Intel, MathWorks or Ansys ships distribution arithmetic natively, which closes the window.Dec 2029 · open
Expressionslisted, private, technologies
  • ORCLOracle (Crystal Ball)
    • not screened
     Crystal Ball, a spreadsheet Monte Carlo and forecasting add-in, sold to corporate planning and engineering users.
  • SNPSSynopsys (Ansys optiSLang)
    • full
     optiSLang robustness and uncertainty-quantification module (from the 2019 Dynardo buy) sold inside Ansys simulation suites to engineering firms.
Private · 2 private companies, 2 of them in the UK.

digiLab · Intellegens

1 more private company tracked, not shown

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

$3M raised by 2 companies, $3M of it in the last three years. Wave building: Money is arriving: recent rounds and tier-1 funds are in, but no runaway leader yet.

  • $3MdigiLab
    • Founded 2021
     Seed, Dec 2024 $3M (Oxford Technology Management, East X Ventures)
  • ·Intellegens
    • Founded 2017
     Grant, Aug 2023 (Innovate UK)
What changedthe record, newest first
  • 7 Sep 2026Carrying probability distributions through ordinary code deterministically, instead of re-running it thousands of times, is now a shipping product with a first chip taped out. It replaces Monte Carlo where the number of uncertain inputs is small and the code already exists, which is a real desk-tool market and, so far, not a venture-scale one. The bet that would change that is a safety rule requiring robots and vehicles to know how sure they are.
  • 7 Sep 2026The one company shipping deterministic distribution arithmetic says its robotics case is speed and ease of writing filters and planners, not a safety requirement to quantify uncertainty. That would be a bigger market if it holds, but no published benchmark on a robotics workload exists yet; every number so far is finance and engineering Monte Carlo.