PSI Proved an 18-Year-Old Quantum Conjecture in Three Days. Its Paycheck Comes From Server Rooms

• The result caps input reads, not total compute time, so it does not cap the commercial value of quantum hardware.
• Exponential speedups for structured problems, including Shor's factoring algorithm, sit outside the theorem entirely.
• Quantum simulation of physics and chemistry, the workload vendors sell into most, is untouched.
• A special case was settled in August 2026 by Guy Blanc, Jordan Docter, Carmen Strassle and Li-Yang Tan of Stanford; PSI's work covers the general case.

• A paying customer that renews without a price concession.
• A discovery converted into a device, a process or a patent rather than a paper.
• A result that survives physical measurement, not just simulation.
• Revenue growth that outruns headcount, which would prove the platform scales rather than the consultancy.

• Capital was never the binding constraint. PSI raised $58 million; NVIDIA booked $96.2 billion in one quarter. The scarce resource was early access to an unreleased Anthropic model — a supplier relationship, not a balance sheet.
• Supplier concentration is the structural risk. If Anthropic grants similar access to other operators, PSI's edge narrows to its verification stack, its customer data and the physical laboratory it is still building.
• The incumbent has conceded the risk in its own filing: Synopsys states that adoption of AI technologies has brought new demands and challenges, including potential disruption to business models and existing technology offerings, in the context of competition for Ansys simulation software.

• What changed: exponential speedups on unstructured, random-input problems are now formally closed off, which removes a talking point rather than a revenue line.
• What did not change: Shor's factoring, quantum simulation of physics and chemistry, and every structured problem, all of which sit outside the theorem.
• What to watch: whether vendors pivot their messaging toward chemistry and optimisation workloads, where the load-bearing commercial claims live.
• What actually moves the shares: bookings and revenue per quarter. A theory result changes the narrative for a week, not the model.

• The margin pool in the template's path is real: roughly $690 million of annualized operating profit sits in Charles River's Discovery Services segment at a 28.9% operating margin, and total company revenue already fell 2.2% year over year in Q3 2026.
• The platform premium is thinner and lumpier. Recursion spent $360.0 million on research over the trailing four quarters against $68.2 million of trailing revenue, and its quarterly revenue is milestone-driven: the $18.6 million reported for Q3 2026 included a $10 million Sanofi milestone.
• Schrödinger shows the same split: total revenue rose 19% while the software segment, the part that sells physics as seats, grew only 6%, with hosted software up 49% to $17.2 million.
• The part automation will not touch quickly is regulated quality work and biologics testing, not the coordination of early discovery campaigns.
• The winners of the template are operators with capital and physical assets: they can buy the research system and they keep the experimental data.

• The quantum narrative reprices first: the theorem closes off one class of advantage and gets argued about for weeks, while no model, no contract and no cash flow changes.
• Hyperscaler and neocloud capital expenditure guidance is the demand pool PSI sells into; Vertiv's third-quarter commentary on robust AI and high-performance compute demand is the cleanest public read on it.
• Watch for a named PSI customer or a disclosed contract value. Until one appears, treat traction as unquantified.
• Anthropic's next decision on frontier-model access determines whether PSI's early access is an asset or a rented advantage.
• The next quarterly prints from Recursion and Schrödinger are the first hard test of whether platform discovery demand is accelerating or thinning.

• By 2029 the Fermi Explorer Mission, where PSI is the founding technical partner and says it established feasibility and found a more efficient trajectory, either becomes the first AI-planned interstellar mission or joins the pile of credible-sounding proposals.
• A discovery-to-device conversion is the milestone that matters: until physical measurement blesses a result, the business stays a services firm with an option attached.
• Model-access economics decide the sector. If unreleased frontier models become a supplier input rather than a scarce gift, research-system operators consolidate into a few well-capitalized players.
• In biology, the same template would run straight through the discovery-services layer, and Charles River's Discovery Services margin is the first place a shift would show up in reported numbers.
• The main risk to this thesis: services prove stickier than expected, or frontier models commoditize faster than any operator can build the verification stack that makes them useful.

Nature gets a vote in your models of reality.