Distyl AI raised $175M at a $1.8B post-money valuation in September 2025 on an estimated ~$31M of ARR — about 58x, the highest multiple in the atlas register (PR Newswire; Sacra; ARR estimate from GetLatka) [WEAK — Latka estimate; the atlas treats all Latka figures as unreliable].
Distyl describes itself as an "AI-native transformation consultant combining software subscriptions with high-touch professional services", with multi-year enterprise contracts "reaching tens of millions of dollars over several years" (Sacra). Strip the framing and it is 111 people doing implementation work, marked at a software price.
That mark is the purest available test of the services-as-software thesis, and the thesis is explicit about what has to be true. Sequoia's "Services: The New Software" argues that as AI crosses a competency threshold, companies that sell the work itself beat companies that sell the tool (Sequoia). The load-bearing assumption: that AI compresses delivery cost fast enough for gross margin to migrate from services levels (30–40%) to software levels (70–85%) before the multiple has to be defended. If it does, 58x is early. If it does not, it is 58x on a consultancy.
Whose budget
Large, new, and unusually — the buyer is frequently an AI company rather than a startup buying from one.
Reported: ~$9B committed by AWS, Microsoft, OpenAI and Anthropic to FDE implementation services between January and September 2025, an 800% rise in FDE job listings across the same window, and 729% growth from April 2025 to April 2026 (VC Cafe) [WEAK — secondary aggregation; no primary source for the $9B]. AWS alone is reported to have allocated ~$1B to FDE programmes; Microsoft runs its Frontier Company initiative; OpenAI's John Deere engagement is the canonical example.
Two sub-markets. Labs and AI-infra companies renting FDEs to land enterprise deals: new budget, effectively uncapped, sized against ACV rather than headcount. Enterprises hiring implementers to deploy agents: partly a displacement of Accenture and Deloitte spend — a good displacement, because the incumbent price is high and the capability weak.
budget: 5. Nobody had this line in 2023. Deployment is the acknowledged bottleneck between a capable model and revenue, and the money to unblock it comes out of the same board approval that funds the compute.
Can you get the supply
Meaningfully hard, and the 729% listing growth is the price signal.
An FDE is a rare combination: a strong engineer who is client-facing, tolerant of ambiguity and willing to travel. That is not the same pool as "AI engineer", and it is not what the contractor marketplaces supply. Toptal, Turing and Andela sell senior engineers; the temperament is the scarce input, and temperament does not appear on a CV.
supply: 4. Harder than design and content, where the moat is operational; easier than Adversarial evals and red-team crowds. The The law is about to arrive question is live too: an FDE embedded full-time in a client's office for a year looks a great deal like an employee.
What the spread looks like
Two routes, with very different economics.
The marketplace route. Turing is the one firm with a published number: 50–55% margin on top of what the engineer actually earns, i.e. a 33–36% take on the billed amount (Futureproofing). Buyer-side rates: Toptal senior AI engineers $100–$200/hr ($16k–$32k/mo at 160 hours); LLM specialists median $240/hr; Turing $100–$200/hr or $17k–$35k/mo; Andela $12k–$14k/mo for a senior engineer on a 12-month minimum plus a $50k conversion fee. Toptal and Andela do not disclose their spread; Andela's published band against Eastern European and African engineer costs implies a comparable or larger one [UNVERIFIED].
The agency route. AI development projects $50k–$500k+; AI consulting $5k–$25k/mo or $100–$450/hr; automation setup $2,500–$15,000 with $500–$5,000/mo retainers (Digital Agency Network).
spread: 3. 33–36% is mid-table on What a rake can actually be — better than compute brokerage's low single digits, far below Expert networks' 70–80%. Distyl did something else: it hired the people directly and wrapped them in software, trading near-term margin for reusable product. That is the only version of this business with a route out of the labour multiple.
Palantir is the proven case: ~82% gross margin on a business built on forward-deployed engineers [UNVERIFIED — widely cited but not sourced in the atlas research notes; confirm against a 10-K before using]. The mechanism matters more than the number. Palantir's FDEs are a cost of customer acquisition and installation for a renewing licence — they go in, make the platform fit, and leave behind software that bills every year whether or not an engineer is present. In an FDE staffing or agency business the engineers are the product: revenue stops when they stop. Same job title, opposite unit economics. Any FDE company claiming a software multiple is claiming it is doing the first thing. Ask what renews.
Can you hold it
Middling, with one strong feature.
Multi-year enterprise contracts "reaching tens of millions of dollars over several years" are genuine hold — an enterprise mid-transformation does not switch implementation partners casually, and every month of embedded work raises the switching cost. That is better than Voice, speech and low-resource language data, where the transaction ends on delivery.
Against it, two leaks. The engineer can be hired directly — Andela prices this explicitly with a $50k conversion fee, which tells you how routine it is; see Getting cut out. And the buyer at the lab end is a handful of companies, so One customer is a binary event applies with full force: if AWS, Microsoft, OpenAI and Anthropic are the source of the ~$9B, then four buyers hold the vertical. Nobody has published a concentration figure for Distyl or anyone else. hold: 3.
What AI does to it
ai: 4. Better models create more deployment work, not less — every capability increase produces a new class of enterprise workflow that somebody has to go and install. The bottleneck moved from model quality to integration, and that is where this budget came from.
The counter-pressure is on the delivery side. If AI compresses what an FDE can do per hour, the billable hour shrinks even as the number of projects grows, and a staffing spread priced per hour is on the wrong side of that. A licence is not. This is the same fork as What better models do to each layer describes elsewhere, sharpened: the technology that creates the demand also deflates the unit you are selling.
What would kill it
Services revenue depresses multiples, and this vertical has already priced as if it will not. Distyl at 58x only works because it is classified as software-plus-services rather than as a consultancy. Compare the public anchors: Appen trades at ~0.9x revenue and Innodata at ~6.2x on ~40% gross margins. If Distyl's revenue turns out to be project-based services revenue rather than recurring licence revenue, it is the single most aggressive mark in the register — and the correction in this sector historically arrives as layoffs, customer loss and silence rather than as a printed down round. See What the public market pays for labour.
Two others. The labs build their own FDE armies — they already are, which is simultaneously the source of the budget and the mechanism by which it gets taken in-house; the same pattern that runs through Expert data for frontier labs. And Accenture fights back on price: the incumbent consultancies are slow but enormous, and the displacement thesis assumes they stay bad at this.
Who is already there
| Player | Model | Position |
|---|---|---|
| Distyl AI | Direct-hire FDEs plus software subscriptions | $1.8B, ~$202M raised, 111 people, ~$31M ARR [WEAK] |
| Turing | Engineer marketplace, disclosed 33–36% take | $2.2B (Mar 2025), $300M+ 2024 revenue |
| Toptal | Freelance marketplace, $100–200/hr | Bootstrapped, independent, spread undisclosed |
| Andela | $12k–14k/mo, 12-month minimum, $50k conversion | $1.5B SoftBank mark (2021), quiet since 2022 |
| AWS / Microsoft / OpenAI / Anthropic | In-house FDE programmes | The source of the ~$9B, and the competition |
The lab programmes are the important row. They are the buyer and the substitute, which is not a position any of the vendors above controls.
Where the record is thin
The headline figure for this vertical — ~$9B committed by AWS, Microsoft, OpenAI and Anthropic between January and September 2025 — comes from a single secondary aggregation with no primary source cited (VC Cafe). The 800% and 729% listing-growth figures come from the same page. They are the only market-sizing numbers this vertical has, and none of them has been checked against a company disclosure or a jobs-data provider.
Distyl's revenue is the other hole, and it is load-bearing. The ~$31M ARR is a GetLatka estimate, and the atlas's own spot-checks found Latka unreliable to the point of listing Surge AI — a company doing over $1B — at $330K of revenue and three employees. The 58x in this page's opening paragraph is therefore a real valuation divided by an unreliable estimate. Distyl has never disclosed revenue, its split between subscription and services, or its gross margin — and that split is the whole argument.
Nor is Palantir's 82% sourced here. It is the anchor for the only proven escape from the labour multiple, and the atlas is carrying it on reputation rather than on a filing. Two hours with a 10-K would fix it, and it is worth doing before anyone builds a pitch on it.
speed: 3. Faster than Expert networks or Robotics teleoperation and physical-world data — a first FDE engagement can be sold and staffed in a quarter — but slower than Adversarial evals and red-team crowds, because the enterprise buyer runs a procurement process and the lab buyer wants references. See Marketplace, staffing firm, BPO or agency for why almost everyone in this vertical ends up choosing the agency shape whatever they call themselves.