The deployment.
Within seventy-two hours in early May 2026, the two largest AI labs in the world made the same move. Anthropic announced a $1.5 billion enterprise-services venture with Blackstone, Hellman & Friedman, and Goldman Sachs to embed Claude inside mid-market companies. Hours apart, OpenAI announced its $4 billion Deployment Company — “DeployCo” — at a $10 billion pre-money valuation, with 19 investment partners and an immediate acquisition of the consulting firm Tomoro to bring 150 forward-deployed engineers in on day one.
The structure is, by the labs’ own framing, copied from Palantir — “almost line for line.” The forward-deployed engineer flies to the client, sits with the operators, learns the workflow, ships software that wraps a frontier model around the actual problem, and stays until the production deployment works. Palantir refined this over years of defense and intelligence work; the labs are now applying it to the broad enterprise market.
The reason is a number. For every dollar companies spend on software, they spend roughly six on services — the ratio that has made consulting a multitrillion-dollar industry. The labs sell the software. The services layer — the integration, the workflow redesign, the change management — is six times larger, and it is the layer where enterprise AI adoption is actually stalling.
The bottleneck is not the model. OpenAI’s own framing is explicit: model performance is no longer the constraint. Integration, security review, evaluation harnesses, and the slow work of redesigning business processes around AI are the actual bottleneck — and MIT research that 95% of generative-AI pilots fail to move beyond the experimental phase is the data behind the move. The labs have concluded that the next phase of enterprise AI is defined not by who has the best model but by who can get the model into production.
The structural argument I want to make: the AI labs are vertically integrating into the services layer via the Palantir forward-deployed-engineer model — because the model layer is commoditizing, the services layer is six times larger, and the FDE is not a consulting arm but a product-formation mechanism that converts deployment work into embedded, expanding, token-metered revenue. This is the move that connects the prior pieces in this track: it is the labs building the machine that produces the consulting compression (the pyramid), generates the enterprise revenue (the runway), and deepens the lock that justifies the valuation.
The headline integrative finding: The FDE model is genuinely powerful and genuinely risky in the same structure. Powerful, because — done the Palantir way — the embedded engineer is not delivering a recommendation but building the production system, creating operational dependency and switching costs that produce expansion and retention; in the token economy, that embedded customer’s revenue is essentially uncapped, scaling with the work the AI does. Risky, because the FDE model is labor-intensive in a way pure software is not — it “resembles consulting more than pure software licensing,” and the open Palantir question is whether deployment scales (margins expand as the platform standardizes) or remains a permanent drag (margins compress as the customer base grows and each new customer needs proportional FDE hours). The labs are betting it is product formation, not services overhead. Whether that bet holds is the question the whole vertically-integrated structure rests on.
This essay walks the simultaneous move, the six-to-one ratio that explains it, the Palantir model the labs are copying and what it actually is, the two labs’ parallel structures, the token-economy economics that make the embedded customer uncapped, the scalability question that decides whether it works, and the structural reading of the labs becoming the consulting industry they are compressing.
The deployment.
How the AI labs vertically
integrated into the services
layer — the Palantir model
at scale.
the identical structural move
the labs had the smaller half
why the embedded customer is rational
the unresolved scalability question
- Blackstone, H&F, Goldman ($300M / $300M / $150M)
- Apollo, General Atlantic, Leonard Green, GIC, Sequoia
- Embed Claude in PE portfolio companies — hundreds of mid-market firms
- Aligned with ~80% enterprise mix
- $10B pre-money · 19 partners (TPG, Bain, Advent, Brookfield)
- Bought Tomoro — 150 FDEs day one (Tesco, Virgin Atlantic, Red Bull)
- Builds the enterprise depth it lacked
- ~2.7x the capital of Anthropic’s vehicle
(the labs sold this)
(the deployment move claims this)
↓
build &
own
The labs have concluded the model is not the product — the deployment is — and moved, in the same week, to own the layer where the model meets the operation. Whether that makes them something larger than software companies or merely rebuilds a labor-bound consulting business at consulting margins is the Palantir question they have all inherited.Thorsten Meyer · The Deployment · Enterprise Reorg 03
By Thorsten Meyer — May 2026
This is the third dispatch in the AI Enterprise Reorganization track. The first walked the CFO’s new operating system; the second walked the consulting pyramid’s compression; the fourth walked the IPO that prices the enterprise disruption. This one is the mechanism beneath all three: the forward-deployed-engineering operation through which the labs deliver the disruption, capture the revenue, and deepen the lock — the machine that makes the rest of the track happen.
The structural argument I want to make: the labs discovered that selling the model is not enough, because the model is not the bottleneck — and so they are buying and building the deployment capacity that turns model access into operational dependency. The move is an admission and a strategy at once: an admission that the model alone does not change how a company operates, and a strategy to own the layer that does. The FDE is the instrument of both.
The headline integrative finding: The labs are doing to consulting what their agents are doing to software — disintermediating it by absorbing its function — but with a twist: they are not eliminating the forward-deployed engineer, they are becoming the firm that fields them, paired with the model the consultant used to recommend. The traditional consultant recommends; the FDE builds and is accountable for the outcome. By owning both the model and the deployment, the labs collapse the recommend-then-implement split that consulting was built on — and capture the six-to-one services dollar that used to flow to the firms their pyramid piece described compressing. Whether they can do it at software margins, or whether deployment stays a labor-bound drag, is the Palantir question they have all inherited.
This essay walks the simultaneous move (Section I), the six-to-one ratio (Section II), the Palantir model and what it is (Section III), the two parallel structures (Section IV), the token economics (Section V), the scalability question (Section VI), and the structural reading of labs-as-consultants (Section VII).
Autonomous AI-Driven Enterprise Software From Development to Deployment
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
I · The simultaneous move · two labs, one structure, seventy-two hours
The convergence crystallization. When the two fiercest competitors in a market make the identical strategic move within three days, the move is not a bet — it is a recognition. Both labs saw the same constraint and reached the same answer.
Anthropic’s venture
The structure: a $1.5 billion enterprise-services venture, announced May 4, 2026, with Blackstone, Hellman & Friedman, and Goldman Sachs as founding partners — Anthropic, Blackstone, and H&F each committing roughly $300 million, Goldman around $150 million — backed by a wider consortium including Apollo, General Atlantic, Leonard Green, GIC, and Sequoia.
The mechanism: a standalone, AI-native enterprise-services firm with Anthropic engineering and partnership resources embedded directly in its team, deploying Claude inside the participating PE firms’ portfolio companies “and beyond.” The consortium’s networks provide a built-in initial customer base of hundreds of mid-market companies.
The framing: Anthropic’s CFO Krishna Rao — “Enterprise demand for Claude is significantly outpacing any single delivery model.” Goldman’s Marc Nachmann described the goal as “democratizing access to forward-deployed engineers” for companies that cannot afford the talent or the consulting fees to build AI systems themselves.
OpenAI’s Deployment Company
The structure: announced within hours, the OpenAI Deployment Company (“DeployCo”) — over $4 billion in initial investment at a $10 billion pre-money valuation, 19 founding investment and consulting partners (TPG, Bain Capital, Advent, Brookfield, and others), majority-owned by OpenAI.
The mechanism: the immediate acquisition of Tomoro, a London/Edinburgh applied-AI consulting firm founded in 2023 in alliance with OpenAI, bringing roughly 150 experienced forward-deployed engineers and deployment specialists from day one — with a client roster including Tesco, Virgin Atlantic, Red Bull, Mattel, and Supercell.
The framing: OpenAI’s own language — “help organizations build and deploy AI systems they can rely on every day across their most important work,” by embedding “Forward Deployed Engineers, or FDEs, into organizations working on complex problems in demanding environments.”
Why the simultaneity matters
Not coincidence — recognition: two labs that compete on every other axis reached the identical structure (standalone services entity, embedded engineers, investor-network distribution, FDE model) within seventy-two hours. When competitors converge this precisely, the move is not a differentiated bet; it is both companies reading the same constraint and concluding there is only one answer.
The acqui-hire signal: OpenAI did not build the FDE org from scratch — it bought one (Tomoro) to start with 150 FDEs already operating. That willingness to pay for an experienced deployment team is a statement that the deployment work matters enough to the strategy that building it organically was too slow.
The simultaneous-move observation
Within seventy-two hours, the two largest AI labs built the same structure — a standalone, investor-backed, FDE-driven enterprise-services firm — because both read the same constraint and reached the same answer: the model is not enough. The convergence is the strongest possible evidence that the services layer, not the model layer, is where the next phase of the competition is being fought. Both labs are now, in addition to model companies, deployment companies — and they became so in the same week.
The AI-Driven Leader: Harnessing AI to Make Faster, Smarter Decisions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
II · The six-to-one ratio · why the services layer is the prize
The market-size crystallization. The move is explained by a single ratio that reframes what the enterprise-AI opportunity actually is — and reveals that the labs have been competing for the smaller half of it.
The ratio
Software versus services: for every dollar companies spend on software, they spend roughly six on services — implementation, integration, customization, change management, and ongoing operation. This ratio is what has made consulting and systems integration a multitrillion-dollar industry, structurally larger than the software industry it serves.
What it means for the labs: selling the model is selling the software dollar. The services dollar — six times larger — flows to the consultants, integrators, and implementation partners who make the software work. By selling only the model, the labs were competing for one-seventh of the value their own technology was unlocking. The deployment move is the claim on the other six-sevenths.
The bottleneck the ratio reflects
The model is not the constraint: the six-to-one ratio exists because making software work inside a real organization is harder than building the software. For enterprise AI specifically, the labs have concluded that model performance is no longer the bottleneck — integration, security review, evaluation harnesses, workflow redesign, and change management are.
The pilot-failure data: MIT research that 95% of generative-AI pilots fail to move beyond the experimental phase, and Deloitte’s finding that insufficient worker skills are the biggest barrier to integrating AI into workflows, are the empirical backbone. The gap between “we have access to the model” and “the model runs our core operations” is enormous — and that gap is the services dollar.
The talent bottleneck
The scarce input is the engineer who understands both sides: the constraint is the shortage of people who understand both the frontier technology and the business context well enough to redesign workflows around it. Job postings for forward-deployed engineers reportedly rose more than 800% in 2025. The labs are not just claiming the services dollar; they are racing to corner the scarce talent that earns it.
The six-to-one observation
The deployment move is explained by one ratio: the services layer is six times the software layer, and the labs had been competing for the software dollar while the services dollar flowed to consultants. The model is not the bottleneck; deployment is — which means the value is in deployment, which is six times larger than the model business the labs built. The vertically-integrated move is the labs reaching past the software dollar they already own toward the services dollar they did not — by fielding the engineers who earn it.
Redesigning Workflows for AI: How to Transform Processes, Decisions, and Organizations in the Age of Intelligent Systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
III · The Palantir model · what the FDE actually is
The mechanism crystallization. The labs are copying Palantir, and the copy is precise — but the FDE model is widely misunderstood as a high-touch services arm. Its real function is something else, and the difference is the whole bet.
The mechanics
The embedded engineer: an FDE flies to the client, sits with the operators, learns the workflow, ships software that wraps a frontier model around the actual problem, and stays until the production deployment works. Palantir refined this over years of defense and intelligence engagements, where software had to work inside complex institutions rather than sit on top of them.
Accountable for the outcome, not the recommendation: the defining difference from consulting. The same team that identifies the opportunity builds the production system and is accountable for the operating result. The consultant delivers a deck; the FDE delivers a running system and owns whether it works.
What the FDE actually is
Not a services arm — a product-formation mechanism: the most important and most misread point. The FDE is not primarily a revenue-generating services business; it is Palantir’s product-discovery and product-formation engine. Forward-deployed engineers operate upstream of the roadmap — the bespoke systems they build inside clients become the patterns that get generalized into the product (Gotham, Foundry, AIP). Consultants operate downstream of the contract; FDEs operate upstream of the roadmap. The deployment work is how the product learns what to become.
Why this distinction is everything: if you treat the FDE as a services arm, its labor cost looks like a permanent margin drag. If you treat it as product formation, its cost is an investment in building the product — temporary per-customer, generalizable across customers. Palantir’s investors who misread early deployment cost as a permanent drag rather than a product-formation investment systematically misjudged the business. The same misread is available now for the labs.
The compression of the sales cycle
The boot-camp evolution: Palantir’s answer to the scalability problem was the AIP boot camp — compressing the traditional 12-18 month enterprise sales cycle into five days, where a prospect’s team builds a working deployment on their own data. Conversion from boot camp to paid contract exceeds 75%; average initial deal size exceeds $1 million; over 1,300 boot camps were completed by end-2025. The boot camp is the mechanism for reducing FDE hours per new customer — the path from “labor-bound services” toward “scalable product.”
The dependency it creates
Operational dependency, not contractual lock-in: once embedded, the customer expands along multiple dimensions, and switching costs accrue not through contract terms but through operational dependency — the system becomes woven into the institution’s operating fabric. That is a deeper lock than a license: you can cancel a license; you cannot easily extract a system your operations now run on.
The Palantir-model observation
The FDE is not a consulting arm; it is a product-formation mechanism that operates upstream of the roadmap, builds operational dependency rather than contractual lock-in, and — via boot-camp-style compression — can convert from labor-bound services toward scalable product. This is what the labs are buying when they copy Palantir: not a services business, but the machine that turns bespoke deployments into product and embedded customers into expanding, hard-to-dislodge revenue. Whether the labs understand it as product formation or run it as services overhead will determine whether they get Palantir’s economics or merely Palantir’s labor cost.
AI Programming Made Practical: A Step-by-Step Guide to Building AI-Powered Applications, Writing Better Code Faster, and Using Modern AI Tools with Confidence
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
IV · The two parallel structures · same model, different distribution
The comparison crystallization. Both labs copied Palantir, but the structures differ in instructive ways — and the differences reveal each lab’s distinct path into the services layer.
Anthropic’s path
The PE-portfolio distribution: Anthropic’s venture is built around the participating private-equity firms’ portfolios — Blackstone, H&F, Apollo, General Atlantic, Leonard Green span healthcare, financial services, logistics, manufacturing, retail, real estate. The venture gets a built-in pipeline of hundreds of mid-market portfolio companies as its initial customer base. A distribution channel no software vendor has previously had at this scale: direct access to the operating companies the PE firms own.
The enterprise-first fit: the venture aligns with Anthropic’s already enterprise-heavy mix (~80% enterprise) and its financial-services penetration — Claude in production at JPMorgan, Goldman, Citi, AIG, Visa. The deployment venture is the implementation muscle for an enterprise motion that was already the company’s center of gravity.
The standalone structure: a standalone entity with Anthropic engineering embedded — at roughly $300M each from the anchors, a leaner vehicle than OpenAI’s, leaning on the PE network for reach rather than on scale of capital.
OpenAI’s path
The acqui-hire-and-scale: OpenAI’s DeployCo launched larger — $4B initial, $10B pre-money, 19 partners — and bought its FDE capacity (Tomoro, ~150 engineers) rather than assembling it. The TPG/Bain/Advent/Brookfield backing provides capital and reach; the Tomoro acquisition provides immediate operating capability.
The catch-up dimension: OpenAI’s move comes amid reports it missed internal revenue and weekly-active-user targets in early 2026 against intensifying competition from Anthropic and Gemini. DeployCo is, in part, OpenAI building the embedded-enterprise muscle that Anthropic’s enterprise-first mix and Claude Code wedge already gave it. The consumer-scale leader is buying its way into the deployment depth the enterprise-first competitor grew organically.
The larger scale: OpenAI’s vehicle is roughly 2.7x the capital of Anthropic’s and built to operate at larger scale — consistent with OpenAI’s broader strategy of converting consumer dominance into enterprise entry through sheer capacity.
What the difference reveals
Two routes to the same layer: Anthropic leans on PE-portfolio distribution and an enterprise-first fit; OpenAI leans on capital scale and an acqui-hired FDE base. Both are claiming the services dollar, but Anthropic is doing it from enterprise strength while OpenAI is doing it partly to build enterprise strength. The structures encode each lab’s starting position — the same convergence on the FDE model, approached from opposite sides of the enterprise/consumer divide that the runway piece described.
The parallel-structures observation
Both labs copied the Palantir FDE model within seventy-two hours, but Anthropic built a leaner, PE-distribution-led vehicle aligned with its enterprise-first strength, while OpenAI built a larger, acqui-hire-led vehicle partly to construct the enterprise depth it lacked. The convergence is on the model; the divergence is on the path — and the paths mirror exactly the enterprise-versus-consumer split that defines the two labs’ IPO stories. The deployment move is the same recognition reached from two different starting positions.
V · The token economics · why the embedded customer is uncapped
The unit-economics crystallization. The reason the FDE investment is rational — the reason it is product formation and not just cost — lives in how AI revenue is metered. The embedded customer is worth something a seat-based software customer never was.
The pricing shift
From seats to tokens: traditional enterprise software is seat-based — a developer is a $20-or-$x/month subscription, capped by the number of seats. AI is increasingly token-based — that same developer using Claude Code or Codex through the API consumes hundreds or thousands of dollars per month in tokens, and the consumption scales with the value the AI is generating.
The uncapped ceiling: as token-based pricing becomes the norm, the revenue ceiling for a single enterprise customer becomes essentially uncapped — bounded only by how much economically valuable work the AI is doing inside that customer. A seat-based customer’s revenue is capped by headcount; a token-based customer’s revenue is bounded only by the work, and the FDE’s job is to put the AI on more of the work.
Why this makes the FDE rational
Deployment acquires an uncapped annuity: spending on the FDE to embed the model deeply acquires a customer whose token consumption then scales with the work the AI does. The deployment cost is front-loaded; the token revenue is recurring, expanding, and uncapped — which is exactly the profile that makes a high-touch acquisition cost rational. You spend on the engineer to acquire the consumption.
The margin story underneath: Anthropic’s inference margins were reported around 70%, up sharply from 38% a year earlier — efficient enough that the unit economics work at scale. If inference is high-margin and the embedded customer’s consumption is uncapped, then the FDE that embeds the model is buying a high-margin, expanding annuity — which is the bull case for the whole structure. Palantir CTO Shyam Sankar’s line captures it: “Tokens are the new coal. Palantir is the train.” The FDE is infrastructure for the token economy.
The expansion dynamic
Land deep, then expand: like Palantir, the model grows by going deep rather than wide — embed in one workflow, then expand across the institution as operational dependency accrues. The token meter turns that expansion directly into revenue: more workflows, more tokens, more revenue, with switching costs rising as the dependency deepens.
The token-economics observation
The FDE investment is rational because token-based pricing makes the embedded customer’s revenue uncapped — bounded only by the work the AI does — and the FDE’s function is to put the AI on more of the work. Front-loaded deployment cost buys a recurring, expanding, high-margin (if inference margins hold) token annuity. This is why the labs can justify the high-touch acquisition cost: they are not buying a seat-capped subscription; they are buying an uncapped consumption stream and paying an engineer to maximize it. The economics work — if the inference margins and the scalability both hold.
VI · The scalability question · what decides whether it works
The risk crystallization. Everything above is the bull case. The bear case lives in one inherited problem: the FDE model is labor-intensive in a way pure software is not, and whether it scales is the question Palantir itself has not fully answered.
The labor-intensity problem
It resembles consulting: the FDE model embeds technical staff at customer sites to build and maintain deployments — which “resembles consulting more than pure software licensing” and introduces labor-cost sensitivity that traditional SaaS avoids. Palantir’s professional services run at roughly 18-20% of revenue. If commercial scaling requires proportional FDE deployment — more customers needing proportionally more engineer-hours — margins will not expand at the rate the software framing implies.
The scalability test
The standardization bet: Palantir’s answer is the AIP boot camp, explicitly designed to reduce FDE requirements per new customer — the data on this is promising but, as analysts note, not yet definitive at scale. The whole question is whether deployment can be standardized faster than the customer base grows. If standardization wins, margins expand as the platform matures; if it loses, each new customer drags the margin and the model stays labor-bound.
The misread risk, both directions: treating early deployment cost as a permanent drag understates the business (if it is product formation); treating it as fully scalable product overstates the business (if standardization does not arrive). The honest position is that it is unresolved — Palantir, after years, runs services at 18-20% of revenue and a 50% adjusted operating margin, which is neither pure software nor pure services. The labs are inheriting that exact ambiguity, at larger scale and with less operating history.
The labs’ specific exposure
They have even less proof than Palantir: Palantir spent years refining the FDE model before the boot-camp compression; the labs are launching their FDE operations now, at scale, with the scalability question unproven for their specific products. The token-economy tailwind (uncapped consumption, high inference margins) is a genuine advantage Palantir’s per-seat-era model lacked — but it does not eliminate the labor-cost question; it offsets it, by an amount not yet measured.
The where-bull-and-bear-meet
The bull case: token economics + standardization make the FDE a product-formation engine that acquires uncapped, high-margin annuities — and the labs scale to consulting-displacing revenue at software-like margins.
The bear case: deployment stays labor-bound, standardization lags the customer base, and the labs end up running large, capital-intensive services operations at consulting margins — having vertically integrated into exactly the lower-margin business their model layer was supposed to transcend. The labs would have become the consulting industry rather than transcended it.
The scalability observation
The scalability question decides whether the labs get Palantir’s product-formation economics or merely its labor cost — and it is genuinely unresolved, even for Palantir, which after years runs services at 18-20% of revenue. The token-economy tailwind helps; it does not settle the question. The labs are betting the FDE is product formation that scales. If it is, they capture the six-to-one services dollar at software margins. If it is not, they have vertically integrated into a labor-bound services business at consulting margins — the very business their agents are compressing.
VII · The structural reading · the labs become the consultants they compress
The synthesis crystallization. Step back, and the deployment move closes a loop the prior pieces in this track opened. The labs are not just entering the services layer; they are becoming the thing their own technology is compressing — and the irony is the insight.
Observation 1 · The model layer commoditizes, so value migrates to deployment
The empirical signal: the labs themselves frame the model as no longer the bottleneck; deployment is. The model layer is commoditizing (multiple competitive frontier models); the value migrates to the layer where the model meets the operation.
The structural reading: as the model commoditizes, the durable value is not in having the best model but in being woven into the customer’s operations — which is a deployment property, not a model property. The labs are racing to own the deployment layer precisely because the model layer, their original moat, is eroding. The forward-deployed engineer is the new moat, because operational dependency is stickier than model superiority.
Observation 2 · The labs are absorbing consulting’s function, not eliminating it
The empirical signal: the deployment ventures compete directly with McKinsey, BCG, Accenture, Deloitte for corporate-AI-transformation work — Anthropic “takes a shot at the consulting industry,” in Fortune’s framing.
The structural reading: the labs are doing to consulting what their agents are doing to software — disintermediating it by absorbing its function — but they are not eliminating the FDE, they are becoming the firm that fields them, paired with the model the consultant used to merely recommend. The pyramid piece in this track described consulting’s compression; this piece describes who captures the compressed value. The consultant recommended a model and an integration; the lab now sells the model and fields the integration engineer — collapsing the recommend-then-implement split that the consulting pyramid was built on. The labs are not killing consulting; they are vertically integrating it.
Observation 3 · The move closes the track’s loop
The empirical signal: the pyramid (compression of consulting), the runway (the IPO that prices enterprise revenue), and the deployment (the FDE machine) are the same phenomenon at three layers.
The structural reading: the FDE operation is the machine that produces the consulting compression the pyramid piece described, generates the enterprise revenue the runway piece priced, and deepens the operational-dependency lock that justifies the valuation. Deploy the model inside the operation → compress the consulting that used to do the integration → generate the embedded token revenue → which is the enterprise-revenue lock → which is the load-bearing valuation argument → which funds the next deployment. The deployment move is where the abstract enterprise-disruption thesis becomes a concrete operating company with engineers on the ground.
Observation 4 · The scalability question is the whole bet
The empirical signal: the FDE is labor-intensive; Palantir runs services at 18-20% of revenue; standardization is promising but unproven at scale.
The structural reading: the entire vertically-integrated structure rests on whether the FDE is product formation that scales or services overhead that drags — and that question is unresolved. If it scales, the labs capture the six-to-one services dollar at software margins and become something larger than software companies. If it does not, they have integrated into a labor-bound services business at consulting margins — having become the consultants they set out to compress, at the consultants’ economics. The bet is that token economics plus standardization make deployment scale. The risk is that they have rebuilt consulting and called it product.
What this is not
It is not a claim that the move is a mistake. The six-to-one ratio is real, the bottleneck is real, and the token-economy tailwind genuinely differentiates the labs’ FDE model from Palantir’s per-seat-era version. The move is rational; its outcome is uncertain.
It is not a claim the two labs are equivalent. Anthropic enters from enterprise strength via PE distribution; OpenAI enters partly to build enterprise strength via acqui-hire and scale. The structures differ; the model is shared.
It is not a claim that the FDE scales or fails. That is the unresolved question — for the labs and for Palantir itself. The honest position is that it is a bet with a real bull case and a real bear case, and the inference-margin and standardization data will settle it, not assertion.
The synthesis observation
The two largest AI labs vertically integrated into the services layer within seventy-two hours, copying Palantir’s forward-deployed-engineer model almost line for line — because the model commoditizes, the services layer is six times larger, and the FDE is a product-formation mechanism that converts deployment into uncapped, token-metered, operationally-locked revenue. It is the machine beneath the whole Enterprise Reorganization track: it produces the consulting compression, generates the enterprise revenue, and deepens the lock that prices the IPO.
There is no single answer. Anyone offering one is selling something. What is unambiguous is that the labs have concluded the model is not the product — the deployment is — and have moved, in the same week, to own the layer where the model meets the operation. They are becoming the consulting industry their agents are compressing, paired with the model the consultants used to recommend. Whether that makes them something larger than software companies — capturing the six-to-one services dollar at software margins — or merely rebuilds a labor-bound consulting business at consulting margins is the Palantir question they have all inherited. The forward-deployed engineer is the bet. The scalability of the forward-deployed engineer is whether the bet pays.
That is the structural editorial question the deployment move sits on top of. It is the model layer commoditizing and value migrating to deployment. It is the labs absorbing consulting’s function rather than eliminating it. It is the machine that closes the track’s loop — and the unresolved scalability question that decides whether the loop produces software margins or consulting ones. And it is the layer where the next phase of the enterprise-AI competition is being fought — not in who has the best model, but in who can field the engineers that get the model into the operating guts of the world’s largest businesses, and do it at a margin that justifies having become a services company to do it.
About the Author
Thorsten Meyer is a Munich-based futurist, post-labor economist, and recipient of OpenAI’s 10 Billion Token Award. He spent two decades managing €1B+ portfolios in enterprise ICT before deciding that writing about the transition was more useful than managing quarterly slides through it. He runs StrongMocha News Group, a network of more than 450 niche WordPress magazines built on the DojoClaw editorial engine. More at ThorstenMeyerAI.com.
Related Reading · the AI Enterprise Reorganization track
This dispatch
- This piece · The deployment · how the labs vertically integrated into the services layer via the Palantir forward-deployed-engineer model — the machine that produces the consulting compression, generates the enterprise revenue, and deepens the lock · empirical-clay dominant, structural-slate and alternative-sage balance
The track
- The pyramid cracks · Enterprise Reorg 02 · the consulting compression that this deployment move captures the value of — the firms being disintermediated by the labs fielding their own FDEs
- The runway · Enterprise Reorg 04 · the IPO that prices the enterprise revenue this FDE machine generates — the deployment move is how the enterprise-revenue lock gets built
- The CFO’s new operating system · Enterprise Reorg 01 · the finance-function compression that the deployment ventures will implement inside the enterprise
Adjacent tracks
- The referral · Post-Wire 03 · the publisher-side mirror of the same intermediation dynamic — value migrating off the layer the incumbent owned
- The clause · AI Governance 03 · the corporate-structure governance of the labs now becoming services companies
- The mandate · Agentic Commerce 03 · the European regulatory architecture the deployment ventures will meet when they cross the Atlantic
Sources
The simultaneous move
- TechCrunch · Anthropic and OpenAI both launching JVs for enterprise AI services — the seventy-two-hour convergence · Anthropic’s $1.5B venture ($300M each from Anthropic/Blackstone/H&F) · OpenAI’s $4B from 19 investors at $10B valuation · both embedding engineers, both investor-network distribution · techcrunch.com
- OpenAI · launches the Deployment Company — primary source: DeployCo to embed FDEs “into organizations working on complex problems in demanding environments,” Tomoro acquisition for FDEs “from day one” · openai.com
- Blackstone · Anthropic partners with Blackstone, H&F, Goldman — primary source: standalone AI-native services firm with Anthropic engineering embedded · consortium (General Atlantic, Leonard Green, Apollo, GIC, Sequoia) · Jon Gray on breaking “one of the most significant bottlenecks to enterprise AI adoption” · Krishna Rao on demand “outpacing any single delivery model” · blackstone.com
The Deployment Company specifics
- TheNextWeb · OpenAI acquires Tomoro as founding piece of Deployment Company — Tomoro from Edinburgh, “copying Palantir’s forward-deployed engineer model” · the model/application/services-layer framing (“the services layer is where the margins are migrating”) · the $14B framing · thenextweb.com
- Kingy AI · OpenAI launched a consulting empire — DeployCo internal name, $4B initial, $10B pre-money, 19 partners, ~150 FDEs via Tomoro · Tomoro clients (Tesco, Virgin Atlantic, Supercell, Mattel, Red Bull) · the $375B market framing · kingy.ai
- Let’s Data Science · OpenAI launches $4B Deployment Company with TPG — “deploy OpenAI models inside Fortune 500 buildings the way Palantir deploys Foundry” · the FDE adapted “almost line for line from Palantir’s playbook” · Tomoro founded 2023 in partnership with OpenAI · letsdatascience.com
- Constellation Research · OpenAI launches Deployment Company, acquires Tomoro — $4B investment, 150 FDEs · “the immaturity of AI deployments and complexity wrangling AI agents have spurred on a run on the FDE model popularized by Palantir” · constellationr.com
The six-to-one ratio and the bottleneck
- Fortune · Anthropic takes shot at consulting industry — “for every dollar companies spend on software, they spend six on services” · the multitrillion-dollar consulting target · Nachmann on “democratizing access to forward-deployed engineers” · “the future of AI revenue may not look like software licensing, but consulting, rebuilt from the model up” · fortune.com
- The Vanderbilt Report · Anthropic’s $1.5B venture exposes the real bottleneck — “it’s a people problem, not a technology problem” · Deloitte 2026: insufficient worker skills the biggest barrier (3,200 professionals, 24 countries) · MIT: 95% of GenAI pilots fail to leave the experimental phase · FDE job postings +800% in 2025 · 91% of mid-sized companies using GenAI, 53% only “somewhat prepared” · vanderbiltreport.com
- Prime AI Solutions · DeployCo and FDEs — “the same team that identifies the opportunity builds the production system… accountable for the operating outcome, not the recommendation” · OpenAI bought rather than built the FDE capability · primeai.solutions
The Palantir model and economics
- MindStudio · Palantir’s FDE model drove 640% returns — Anthropic inference margins ~70% (up from 38% a year earlier) · seat-to-token pricing shift making the single-customer ceiling “essentially uncapped” · ARR $9B→$44B doubling ~every six weeks (~$96M ARR/day per Ming Li) · Sankar: “Tokens are the new coal. Palantir is the train” · the FDE “too expensive and too weird” for others to copy · mindstudio.ai
- Medium / balaji bal · Understanding Palantir — the FDE as “product discovery and product-formation mechanism,” not a services arm · “FDEs operate upstream of the roadmap; consultants operate downstream of the contract” · operational dependency vs contractual lock-in · “treating early deployment cost as a permanent margin drag rather than a product-formation investment leads to systematic misinterpretation” · medium.com
- Techi · PLTR stock analysis — the boot-camp compression (12-18 month cycle → 5 days), >75% conversion, >$1M average initial deal, 1,300+ boot camps by end-2025 · full-year adjusted operating income $2.254B (50% margin) · the FDE “resembles consulting more than pure software licensing” labor-cost sensitivity · techi.com
- PitchGrade · Palantir business model — professional services at 18-20% of revenue · “if commercial scaling requires proportional FDE deployment, margins won’t expand at the rate the model implies” · the AIP boot camp “explicitly designed to reduce FDE requirements per new customer — promising but not yet definitive at scale” · pitchgrade.com
The two parallel structures
- CNBC · Anthropic teams with Goldman, Blackstone on $1.5B venture — embed engineers inside mid-sized companies to redesign workflows · “having the model alone doesn’t change your workflows… you need people who can combine the technology with what’s actually happening in the business” (Nachmann) · the enterprise IPO battleground framing · cnbc.com
- Fortune · Anthropic deepens push into Wall Street — the JV as “essentially a forward-deployed engineering operation” · Claude in production at JPMorgan, Goldman, Citi, AIG, Visa · the two-track strategy (largest institutions self-serve; mid-market via the PE-backed JV) · Argenti’s three waves at Goldman · fortune.com
- BeInCrypto · OpenAI borrows Palantir’s playbook — DeployCo $4B, 150 FDEs via Tomoro · OpenAI reportedly missed internal revenue/WAU targets in early 2026 amid Anthropic/Gemini pressure · the timing relative to Anthropic’s venture · beincrypto.com
The track backbone
- The pyramid cracks / The runway / The CFO’s new operating system · Thorsten Meyer · Enterprise Reorg 01-02-04 · the consulting compression this move captures, the enterprise revenue this machine generates, the IPO that prices the lock it deepens — the deployment move is the mechanism beneath all three
Key reference figures crystallized
- The simultaneous move: Anthropic $1.5B venture (May 4; $300M each Anthropic/Blackstone/H&F, $150M Goldman; Apollo, General Atlantic, Leonard Green, GIC, Sequoia) · OpenAI DeployCo (May 11; $4B initial, $10B pre-money, 19 partners incl. TPG/Bain/Advent/Brookfield) + Tomoro acqui-hire (~150 FDEs; Tesco, Virgin Atlantic, Red Bull, Mattel, Supercell)
- The ratio: $1 software : $6 services · the multitrillion-dollar consulting layer · model no longer the bottleneck (integration, security, eval harnesses, workflow redesign) · MIT 95% of GenAI pilots fail to leave pilot · FDE job postings +800% in 2025
- The Palantir model: FDE = product-formation mechanism, upstream of the roadmap, operational dependency not contractual lock-in · boot camp 12-18mo → 5 days, >75% conversion, >$1M initial deal, 1,300+ by end-2025 · adjusted operating margin 50%, services 18-20% of revenue
- The token economics: seat → token pricing, single-customer ceiling “essentially uncapped” · Anthropic inference margins ~70% (from 38%) · “tokens are the new coal; Palantir is the train”
- The scalability question: FDE “resembles consulting more than pure software licensing” · services 18-20% of Palantir revenue · standardization “promising but not yet definitive at scale” · bull = product formation at software margins; bear = labor-bound services at consulting margins
- The structural loop: model commoditizes → value migrates to deployment → FDE captures the six-to-one services dollar → produces consulting compression (pyramid) → generates enterprise revenue (runway) → deepens the lock that prices the IPO
