Theses›Datacentre & AI compute›Probabilistic computers

Probabilistic computers

Probabilistic computers (p-bits / Ising / thermodynamic samplers) own server-side sampling + optimisation before fault-tolerant QC

  • Bear
  • Early
  • $347M by 2035

The callSampling could become a native chip function, but nobody has beaten a tuned GPU or a classical solver on a real customer problem, and the displaced incumbent is tiny.

D1D2R1R2C1foragainst D1D2R1R2C1foragainst
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. D1Generative AI and agents make sampling a datacentre workload, and GPUs simulate randomness with deterministic arithmetic, which wastes most of the energy.
  2. D2Transistor noise as the random source works at room temperature on a standard fab, and the US government is putting up to $75M behind Extropic.
  3. R1Nobody has beaten a tuned GPU or a classical solver on a real customer problem; the 10-year story is always five years from advantage.
  4. R2The incumbent being displaced, Gurobi, is a sub-$100M business, D-Wave grew 0% year on year in Q2 2026, and Fujitsu's annealer ambition was never reported as met.
  5. C1Extropic's Z1 reaches early access in 2027 and converts to an eight-figure enterprise order.end-2028
Rests onthe beliefs underneath
Supported by · confidence
What would prove itdated tests, open and resolved
20272028202920302031today1234567predictionswatch 20272028202920302031today1234567predictionswatch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1A probabilistic substrate demonstrates practical advantage vs top classical solvers (Gurobi 12+/CPLEX 22+) on a published commercial benchmark (N>1000) by end-2027.Dec 2027 · open
  2. 2A p-bit / thermodynamic / photonic sampler reaches paying commercial deployment on a sampling or probabilistic-inference workload (not pure optimisation) by end-2027.Dec 2027 · open
  3. 3A published win over Gurobi 12+ or CPLEX 22+ on a commercial benchmark above 1,000 variables.Dec 2027 · open
  4. 4A sampler vendor (Normal, Extropic or another) discloses a paying deployment.Dec 2027 · open
  5. 5At least one probabilistic-computing vendor reaches $50M+ ARR by 2028.Dec 2028 · open
  6. 6Any vendor at $50M+ of recurring revenue; D-Wave's revenue and cash read each quarter.Dec 2028 · open
  7. 7Probabilistic computers stay workload-specific and server-side, NOT a general-purpose or edge compute platform, through 2030.Dec 2030 · open
Expressionslisted, private, technologies
  • 6501.THitachi
    • full
  • 6702.TFujitsu
    • fair
  • QBTSD-Wave Quantum
    • speculative
     Annealing quantum — different paradigm.
  • 9432.TNTT
    • not screened
     LASOLV coherent Ising machine (photonic) for optimisation, offered through research and enterprise pilots.
Private · 6 private companies, 3 of them in the US; none in continental Europe.

Normal Computing · Extropic · Toshiba · 1QBit · Ludwig Computing · LightSolver

3 more private companies tracked, not shown

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

$96M raised by 4 companies, $141M 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 strategic investor: Normal Computing.

What changedthe record, newest first
  • 7 Sep 2026A Cambridge company now ships deterministic hardware and a compiler that carry probability distributions through ordinary floating-point code, replacing Monte Carlo at 1/N convergence on standard 22nm silicon. That is the classical solver closing the gap the probabilistic-hardware bet depends on, and it strengthens the case that uncertainty-aware computing is captured in software and conventional chips before any thermodynamic or p-bit substrate is needed.
  • 21 Jul 2026A useful map of where the alternatives to conventional computing sit: digital holds until circuits get small enough that noise takes over, then you are in probabilistic territory, then quantum. The claim worth testing is that a reversible approach extends the digital runway, pushing back the date when the alternatives become necessary.
  • 23 Jun 2026D-Wave more than doubled revenue to $24.6M and is sitting on near-$900M cash, so the path to a $50M optimisation business now looks ~2x out rather than a leap, and the first thermodynamic production chips are shipping, which turns a simulation argument into a measurable one.
2 earlier
  • 22 Jun 2026Probabilistic chips could own a real optimisation and sampling niche before fault-tolerant quantum arrives, if the moat is the fabric and compiler rather than the device.
  • 11 Jun 2026The thermodynamic-sampling cohort's quiet problem reads less like bad timing and more like building for a model class (energy-based / heavy-sampling) the field routed around to autoregressive transformers. The near-term pull is in the narrow optimisation/sampling vertical, not the grand "replace the GPU" story; what would flip it is test-time / inference-time search making cheap hardware sampling central.