Who it's for
Solo engineers, hardware startups, skunkworks proving the loop on one part.
01The mispricing trap
Why you cannot price this like CAD
Seat pricing meters the tool — who is allowed to open the IDE. Prometheus doesn't sell a tool; it sells the output of a senior engineer's iteration loop. The unit of value is completely decoupled from headcount, and that breaks the seat model in four ways.
The value-per-seat is enormous
The whole market tops out near $7.5K/seat·yr for a pure design tool, and ~$65K–$100K for the highest-end multi-physics CAE seat. A 10-person team at $100K a seat is ~$1M/yr — three orders of magnitude below the value one team creates, and nowhere near what underwrites a ~$40B valuation.
Compute trends to free
Surrogate models score a design in milliseconds, not hours. So any cost-plus or per-GPU-hour logic collapses toward zero exactly as the product gets better. Meter the raw compute and you give the value away.
The outcome is decoupled from headcount
The product's whole pitch is exploring 5 → 10,000 candidates. A handful of users produce a fielded program worth tens of millions. Pricing tied to seat count can never track value that lives in the part, not the person.
Seats invert the incentive
A per-seat buyer rations users and under-explores — defeating the exact wide search that is the product. The meter should encourage more exploration, not punish it.
The structural rule
Meter the loop you compress. Capture the real money only at the physical and regulatory gates you cannot — validation, certification, the part actually flying. Seats get you in the door; they are never the profit engine.
02The shape of the model
One price that slides along a value spectrum
Don't pick a single pricing axis — stack three. As the customer's trust and stakes rise, monetization moves from a fixed access floor, through metered consumption, to a share of the value created. Each layer is anchored to a market that already exists.
← Tool · pays for access
Engine · pays for work done
Partner · pays for value created →
1
Platform / Access — the anchor floor
A per-engineer or per-program annual subscription to the design environment: generation, surrogate evaluation, and a bundled monthly credit pool. Tiered by capability — single-physics vs. multi-physics, and by domain pack (aero / auto / materials / drug discovery) — exactly as ANSYS tiers Premium vs. Enterprise. The PLM / data-of-record layer is sold separately, per the Teamcenter precedent.
Single-physics · $18K–$30K / seat·yr
Multi-physics · $40K–$100K / seat·yr
Enterprise program · $250K–$2M+ / yr
Priced above commodity CAD ($1,600 Fusion) and the $7,500 nTop ceiling, because one seat replaces a multi-tool CAE suite plus a senior engineer's iteration judgment — not a drafting tool.
2
Consumption — the scalable engine
Metered Design Credits drawn from a shared, fungible org-level pool (the Altair Units / ANSYS Elastic model) for every generative pass and high-fidelity confirmation beyond the seat's bundled allotment. The pool floats across users and modules and bills on draw, not on named seats.
List · ~$300 / credit
Volume · → ~$120 at 100K+ / yr
Full campaign · 10–40 credits = $3K–$12K
Heavy program · $30K–$150K / yr
Meter the outcome atom (a generative pass / a campaign), never per-candidate-geometry or raw GPU-hours. Autodesk's killed per-export fee proves a per-result tax punishes the exact exploration that is the value. Generous bundles + rollover so credits never ration the search.
3
Value capture — co-development
For flagship, certifiable programs — jet-engine hot sections, clean-sheet structures, new alloys, molecules — a deliberately small upfront, a gated milestone stack, and a royalty on the deployed part's lifetime mass / fuel / cost savings. The structure is borrowed wholesale from AI-pharma biobucks (Insilico–Lilly: $115M upfront on a ~$2.75B headline + royalties).
Upfront · $40M–$120M (~3–4% of headline)
Milestone stack · $0.5B–$2B
Royalty · mid-single → ~21% w/ co-invest
Back-load capture to the physical/regulatory gates the AI cannot compress. The customer pays almost nothing if the design fails — essential, because a confidently-wrong simulation is catastrophic — and Prometheus banks durable upside only on programs that actually field.
03The metering unit
The Design Credit
Engineering buyers won't pay per token like an LLM — they anchor on work done: a study, a part, a validated design. So the meter is an atom of autonomous design work, the way Devin bills an “ACU” of agent labor or Fusion bills a ~33-token generative study.
1 Design Credit (DC) =
~$300One generative pass that spawns and ranks a candidate population, including up to ~1,000 AI surrogate evaluations scoring those candidates in milliseconds. Volume pricing falls to ~$120/DC at 100K+ credits a year.
- Generative pass + ≤1,000 surrogate evals1 DC
- High-fidelity FEA / CFD confirmation (the expensive, non-surrogate solve)3–5 DC
- Full exploration campaign (thousands of candidates, ~5,000 evals, 1–2 confirmations)10–40 DC
- Seat-bundled monthly pool (overage metered from a shared org pool)~20 DC / seat
- Heavy multi-objective program100–500 DC / yr
Anchors: Fusion generative study ≈ 33 tokens ≈ $99 · Devin ACU $2.00–2.25 (~15 min of agent labor) · Abaqus job 15–85 SimUnits, sublinear in cores · Siemens STAR-CCM+ a flat $22/solver-hour regardless of core count — proof that engineering buyers already accept price decoupled from raw compute. A full campaign at ~$3K–$12K is deliberately a sliver of one avoided physical iteration ($300K–$2M+), so the customer stays hugely net-positive and keeps exploring.
04Packaging
Four tiers, from solo builder to flagship partner
The same three layers, packaged for four very different buyers. Each tier shifts weight from access toward outcome as the stakes — and the trust — rise.
Individual / Builder$18K–$30K / seat·yr
(startup rate ~$5K–$8K)
(startup rate ~$5K–$8K)
Structure
Per-seat subscription + ~15–20 bundled credits/mo + metered overage (~$300/DC). Self-serve.
Anchored to
nTop ($7.5K/seat) and Cursor Ultra ($200/mo), repriced 2–4× up — the user is a senior engineer, not a consumer.
Team / Department$40K–$100K / seat·yr + pool
typical ACV $500K–$2M
typical ACV $500K–$2M
Who it's for
10–50-seat teams at automotive, industrial, aerospace-supplier or materials firms running part portfolios.
Structure
Multi-seat + shared floating credit pool ($50K–$250K/yr) + an 18–25% validation-assurance line that funds confirmation accuracy.
Anchored to
ANSYS 100-user deployments >$500K/yr · Schrödinger $500K+ ACV · Palantir $4–7M land expanding via 130%+ NDR.
Enterprise / Program (incl. gov / ITAR)$2M–$15M / yr platform
$10M–$50M+ TCV envelopes
$10M–$50M+ TCV envelopes
Who it's for
OEM prime programs — an aircraft part family, vehicle platform, engine component — plus defense & sovereign buyers.
Structure
Use-case program license (Palantir model — priced per program + compute, not per seat) + a pre-committed credit envelope. On-prem / ITAR carries a 30–60% premium.
Anchored to
Palantir use-case + expansion ($80M–$96M deals) · PhysicsX / Monolith embedded enterprise contracts.
Co-Development Partner$40M–$120M upfront + $0.5B–$2B milestones
+ royalty (mid-single → ~21%)
+ royalty (mid-single → ~21%)
Who it's for
Strategic primes co-developing flagship, certifiable, deployed assets — where the part's lifetime value is the prize.
Structure
Small upfront + gated milestone stack (design-freeze → prototype → first-article → certification → production/EIS) + royalty on lifetime savings. Co-investment dial lifts the royalty band.
Anchored to
Insilico–Lilly ($115M + ~$2.75B) · Exscientia–Sanofi ($100M + $5.2B, royalty to ~21%) · Recursion–Roche ($300M+/program × 40).
05Run the numbers
Three deals, end to end
What a customer actually pays — and why it pencils out for both sides — across three points on the spectrum.
Aerospace bracket — Team tier + consumption
A tier-1 supplier needs a validated, manufacturable structural bracket. Historically 4–9 months and 3–5 physical iterations; now on a 25-seat department license.
Before
4–9 months · 3–5 physical prototype/test loops at ~$300K–$800K each · ~5 variants explored · baseline mass.
With Prometheus
~2 weeks · one campaign spawns thousands of candidates, ~5,000 surrogate evals, 1–2 hi-fi confirmations, one physical confirmation · part ~38% lighter.
What they pay
Multi-physics seats covered by the department ACV; the campaign draws ~25–30 Design Credits = ~$7,500–$9,000 in consumption.
Why it pencils: collapsing 3–5 iterations to ~1 avoids ~$0.6M–$3.2M of iteration cost. Paying ~$8K of credits to capture that is an ~80×–350× return on the consumption layer alone — before counting lifetime mass savings.
Clean-sheet engine hot-section part — Co-Development tier
An engine OEM co-develops a certifiable hot-section component for a flagship program that flies on a platform with a multi-decade service life.
Before
Multi-year design + iteration loop · dozens of physical test articles · certification as the long pole · baseline mass & fuel burn.
With Prometheus
Design loop compressed dramatically (certification unchanged) · manufacturability in the objective from run 1 · a significantly lighter part, improving fleet-wide fuel burn.
What they pay
$60M upfront (~3% of a ~$2B headline) + a gated milestone stack summing to ~$1.94B — $25M design-freeze · $75M prototype-validation · $200M first-article · $540M certification · $1.1B production/EIS — + a 4% royalty on documented lifetime fuel/mass savings, rising toward ~21% if Prometheus co-invests in cert testing.
Why it pencils: the customer risks only ~3% upfront and pays the big sums only after the non-compressible gates prove the design is real and safe. A ~30–40% lighter hot-section part saves millions in fuel per shipset; a mid-single-digit royalty on a fleet-wide saving is durable eight-to-nine-figure capture — indexed to the asset, not the tool.
Automotive battery-pack portfolio — Enterprise + gain-share
An OEM applies Prometheus across ~40 structural/battery brackets on a new ~4-year vehicle platform, against an independently-audited baseline of historical time and mass.
Before
~40 parts each at typical iteration cost and timeline · baseline program engineering spend and vehicle mass established and audited.
With Prometheus
One program license drives the whole portfolio via a shared credit pool · aggregate iteration time cut ~50%, aggregate mass down ~30% — both verified before any gain-share is paid.
What they pay
$3M/yr platform + a $400K pre-committed credit envelope + a 12% gain-share on independently-audited time and mass/cost savings vs. baseline — paid only on parts that pass physical validation. Total ≈ $7.8M–$12.6M/yr.
Why it pencils: ~50% time and ~30% mass reduction across 40 parts documents a $40M–$80M value delta; a 12% gain-share (mid-range of the 10–20% ESCO/McKinsey band) is $4.8M–$9.6M of value-indexed capture on top of the platform floor — and the land-and-expand motion grows it from one bracket to the full portfolio.
06The receipts
Every number borrows from a market that exists
Prometheus is a new category, but nothing in this model is invented from scratch. Each layer copies a pricing mechanic that engineering or pharma buyers already pay.
| Comparable market | Unit of charge | Price anchor | What Prometheus borrows |
|---|---|---|---|
| High-end mechanical CAD Siemens NX · Dassault CATIA |
per seat · yr (+ ~20% maint.) | NX ~$9K/seat + ~$1.8K maint · CATIA $7.1K–$7.6K/yr | Anchor the seat above high-end CAD, never entry CAD; adopt floating token/value-based pools (NX-X) for the shared credit pool. |
| Engineering simulation / CAE ANSYS · Abaqus · STAR-CCM+ |
seat + metered solver throughput | ANSYS CFD Enterprise ~$65K · STAR-CCM+ $22/solver-hr · Elastic heavy users $50K+/yr | Two decoupled meters (seat + evaluation); flat-rate-regardless-of-cores proves buyers accept price decoupled from compute. Price hi-fi confirmations as the expensive solve. |
| Generative / topology tools Fusion · nTop · Altair Units |
seat + cloud-credit per run | Fusion study ~33 tokens ≈ $99 · nTop ~$7.5K/seat · Altair shared pool | The Design Credit atom + the fungible shared pool. Negative lesson: Autodesk killed per-outcome export fees — never charge per candidate geometry. |
| AI agent / developer tools Devin · Cursor · Copilot · Claude |
seat + per-task effort unit | Copilot $19–$39/seat · Cursor Ultra $200/mo · Devin ACU $2.00–2.25 (~15 min) | The autonomy ladder (seat → per-task as the agent moves from assist to autopilot); Devin's ACU as the template for the Design Credit. |
| AI drug-discovery co-development Isomorphic · Insilico · Exscientia |
upfront + milestones + royalty | Insilico–Lilly $115M on ~$2.75B (4.2%) · Exscientia–Sanofi $100M + $5.2B, royalty →21% | The entire Layer-3 structure: ~3–4% upfront, value back-loaded to non-compressible gates, royalty on lifetime asset value, co-investment dial. |
| Value-based / gain-share Palantir · ESCO · Rolls-Royce |
use-case license · % of audited savings | Palantir NDR 134–139%, $4–7M→$20–31M · ESCO 10–20% of audited savings | Palantir use-case (not seat) pricing + land-and-expand; ESCO/McKinsey gain-share on audited savings vs. an agreed baseline; power-by-the-hour as the long-term option once trusted. |
↔ scroll table
07Sequencing
Earn each layer before you charge for it
In safety-critical hardware, trust is the gating constraint — no one bets a 10-year program on AI output they haven't watched succeed. So the model unlocks in order, each stage de-risking the next.
Free / academic — seed the data flywheel
Give the design environment free to students, academics, and research labs (the nTop / Schrödinger / Carbon playbook). Prometheus's moat is training on real experimental + physics data; a free tier seeds adoption and the proprietary data flywheel that compounds surrogate accuracy — the one asset that defends a $40B valuation.
Land with seats — paid pilots on real data
Run low-risk, time-boxed pilots on the customer's own experimental data (the Palantir bootcamp model), proving the loop on a single bracket. This builds the trust required before anyone bets a program on AI output.
Expand to consumption
Once the loop is proven on one part, grow the shared credit pool across the customer's portfolio — converting a few seats into department/program ACV and targeting 130%+ net revenue retention, without renegotiating seats.
Co-development last
Only after the surrogate has demonstrated validated, physically-confirmed accuracy on real parts do you earn the right to milestone + royalty deals. You cannot credibly ask for a royalty on a deployed part until you've proven the design holds at the physical gates.
A standing risk throughout: incumbents (Siemens, Dassault, ANSYS, Autodesk) can bundle “good-enough” generative features into existing seats at near-zero marginal price. Prometheus must compete on validated accuracy and data moat, not features — and, where possible, distribute alongside incumbents (the PhysicsX-via-Siemens motion) rather than purely against them. Distribution beats list price.
08Where the model is fragile
The unresolved questions
A pricing model for a pre-product company is a hypothesis. These are the open questions that would move the numbers most — and the places a real aerospace or pharma buyer would push back hardest.
Liability
If a confidently-wrong simulation contributes to a part that fails in service — catastrophic in safety-critical hardware — who bears liability, and how is it priced into the validation-assurance line and milestone clawbacks?
Ownership
Does the OEM get an exclusive license to the deployed design while Prometheus keeps the underlying method and data flywheel? What stops a customer redesigning around the output to escape a multi-decade royalty?
Manufacturability
A lattice you can't machine affordably is worthless. Is manufacturability-in-the-objective a paid premium or table stakes — and how is it justified when those constraints reduce the very mass savings the royalty is indexed to?
Cert cost-sharing
Certification doesn't compress. Should Prometheus co-fund FAA/regulatory testing to earn the higher (~21%) royalty band, and how is that risk capital priced relative to gates the AI can't influence?
Sovereign / ITAR
Air-gapped, ITAR-compliant on-prem breaks the SaaS economics and severs the central data flywheel. Is a 30–60% premium even enough — or does it permanently weaken the moat that justifies the valuation?
Baseline disputes
Gain-share rests on an agreed historical baseline and an independent auditor. How are baselines negotiated and gaming prevented when the customer's “historical” iteration count is self-reported?
Credit rationing
If credits expire or run out, customers ration the exact 5→10,000 search that is the value. How large must bundled pools and rollover be before rationing shows up as a net-revenue-retention cliff?
Data-as-payment
Should sophisticated buyers pay partly in proprietary experimental data — and how is that valued against the risk that the largest OEMs demand their data never train shared models, starving the flywheel?