On Monday, May 11, OpenAI made an announcement that, on the surface, looked like a corporate housekeeping note. The Deployment Company — the $4 billion enterprise services venture OpenAI quietly assembled in early May with 19 institutional partners — would acquire Tomoro, an applied AI consulting firm OpenAI itself helped create back in 2023. The deal pulls about 150 forward-deployed engineers (FDEs) into a single OpenAI-controlled entity, valued at $10 billion at close.
Read it again. OpenAI just spent the better part of a year quietly assembling a private-equity-backed services arm. It then bought the consulting firm it had previously seeded. The closing move was to staple 150 embedded engineers onto the front line. The structure is not subtle. The frontier labs have decided that the next revenue dollar in enterprise AI does not come from selling tokens, prompts, or seats. It comes from putting engineers in your finance department.
That is the actual story of May 11–13, 2026. The fact that OpenAI committed $4B to it. The fact that Anthropic committed $1.5B to a parallel venture nine days earlier with Blackstone, Hellman & Friedman, and Goldman Sachs. The fact that the combined $5.5 billion was raised explicitly to fund a Palantir-style FDE model. None of that is the news. The news is where those FDEs are going to land first — and why the CFO is the new beachhead for enterprise AI.
This article unpacks the structural shift, the data that explains why finance is the target, two decision frameworks every CFO and CIO should run before the next AI services pitch lands on the desk, and the 90-day action items that follow.
The $5.5 Billion Confession
Spending $5.5B on services structure is not a routine market move. It is a confession.
For the first two years of the GenAI cycle, the dominant narrative from both OpenAI and Anthropic was that model capability would drive enterprise adoption. Bigger models, longer context, better reasoning, lower cost per token — and customers would integrate. That narrative is now over. Between late 2023 and mid-2025, OpenAI's enterprise market share dropped from roughly 50% to 25%, while Anthropic gained share aggressively. The reason was not model quality. Claude and GPT-class models are now functionally substitutable for most enterprise workloads. The reason was deployment velocity. The customers who chose Anthropic were not picking a better model. They were picking a vendor that showed up with engineers.
That insight is what funded May's announcements. The $5.5 billion is the labs' admission that models are no longer the moat. Implementation is.
Three data points reinforce the point:
- 88% of organizations running AI agents in production reported a security incident in the past year — not because the models were broken, but because the deployments were under-engineered.
- 42% of C-suite executives say AI is creating organizational conflict — finance fights IT, ops fights legal, governance fights product. The model is irrelevant to any of those fights.
- 60% of finance AI initiatives fail due to organizational readiness gaps, not technology limitations, per Gartner's most recent CFO AI tracker. Eleven percent of executives cite the technology itself as the chief barrier. The other 89% cite people, process, and integration.
OpenAI did not buy Tomoro because Tomoro has better technology than OpenAI. OpenAI bought Tomoro because Tomoro has 150 engineers who can walk into a Fortune 500 CFO's office and ship working code. That is what enterprise customers will pay for in 2026, and that is what the frontier labs have decided to sell.
Why the CFO is the Beachhead
Of all the seats where an embedded engineer can land — IT, ops, security, sales, R&D — the labs are pointing first at finance. The reasons are structural.
Reason 1: Finance has the cleanest data. Compared with marketing, sales, or operations, finance data is structured, audited, and standardized. ERP systems (SAP, Oracle, NetSuite, Workday) enforce schema. GL accounts and chart-of-accounts hierarchies are stable. Month-end close cycles produce predictable inputs and outputs. An FDE walking into a CFO's office does not have to spend six weeks cleaning a CRM. The data is already organized.
Reason 2: Finance has the highest pain density per workflow. Close cycles, FP&A, treasury, AR/AP, expense reconciliation, audit prep — every one of these is repeatable, time-consuming, and currently absorbing hundreds of hours per quarter in a typical mid-cap finance team. The "before" state is well-instrumented; the "after" state is measurable.
Reason 3: Finance has budget authority. The CFO does not have to convince anyone else to approve the spend. If an FDE engagement reduces close time by four days or cuts FP&A headcount needs by 30%, the CFO writes the check and books the savings against the same P&L line.
Reason 4: Finance has the audit appetite. Unlike marketing or product, finance teams are accustomed to working under audit, internal control, and regulatory frameworks (SOX, GDPR, IFRS, GAAP). They will engage with governance overhead that other functions resist. That makes them a better customer for AI systems that need policy, logging, and human-in-the-loop controls.
The market is following the logic:
- 56% of finance leaders now use AI — roughly double the 2023 adoption rate.
- 17% have AI in production, 34% are actively piloting, 28% are planning to pilot. Only 21% are still just considering.
- 83% of CFOs at $1B+ revenue companies plan AI budget increases above 15% over the next two years. 42% expect increases above 30%.
The same data shows the gap that funds the FDE model. Only 17% of finance teams have AI in core workflows. 45% remain in limited pilot mode. Among CFOs who have piloted AI, only 4% report a pilot success rate above 50%. 36% cannot justify a return on investment. 71% cite model inaccuracy as their top concern. 86% say legacy tools are a significant or moderate barrier.
That is the addressable market. Eighty percent of CFOs want AI in production. Less than one in five have it. The labs have decided to close that gap with embedded engineers, on the customer's premises, until the system works.
The Palantir Playbook OpenAI and Anthropic Are Copying
The forward-deployed engineer is not a new role. It is Palantir's signature go-to-market motion, refined over fifteen years across U.S. defense, intelligence, financial services, healthcare, and energy customers. The mechanics are simple to describe and brutal to execute:
- An FDE is a software engineer employed by the vendor, embedded full-time inside the customer's organization, typically for three to twelve months.
- The FDE writes production code in the customer's environment, against the customer's data, integrating the customer's systems.
- The FDE stays until the deployment works — until the dashboard renders the right numbers, the workflow runs end-to-end, the audit log passes review, the user count crosses a threshold.
- The vendor charges premium pricing — typically seven-figure annual contract values per FDE — and the customer accepts that price because the alternative is a Big Three system integrator engagement that takes 2–3x as long and ends with a Powerpoint instead of code.
Palantir compensation data shows the cost basis. Forward-Deployed Software Engineer total compensation at Palantir runs from $171K at the low end to $415K at the median high, with staff-level FDEs clearing $630K+. New listings at the FDSE level start at $135K–$200K base before equity. Anthropic, OpenAI, and Cohere are paying in the $200K–$400K+ range with material equity grants on top.
The financial math is unforgiving. An FDE costs the vendor roughly $300K–$500K all-in per engineer per year. Tomoro's 150 FDEs therefore represent a fully-loaded annual cost basis of $45M–$75M for OpenAI before the engineers generate any margin. That is why FDE-led services require seven-figure ACVs to clear. The vendors have decided the math works because every successful FDE-led deployment locks in a customer relationship across the model, the application layer, and the integration substrate — economics traditional model-tokens revenue cannot match.
Tulika Sheel of Kadence International framed the trade-off precisely: vendor-provided services reduce short-term deployment risk but create "deeper dependency across the stack, from models to data pipelines and workflows." Faisal Kawoosa of Techarc called it the natural evolution: model providers "moving beyond being platform vendors to actively shaping the entire AI value chain." Neil Shah of Counterpoint described the endgame as becoming a "one-stop shop" — model, application, deployment, governance — and pricing accordingly.
The structural risk is the same risk that defined the Oracle/SAP era of the 1990s: when the vendor owns the implementation, the vendor owns the upgrade path, the integration roadmap, and the exit cost. The difference now is that the lock-in extends into the agent layer — the autonomous systems that touch your GL, your treasury, your forecast. Switching vendors a year into a deployment means rewriting the agents, retraining the workflows, and re-auditing the controls. That is a one-time exit cost most CFOs will not pay.
Framework #1: The FDE Investment ROI Calculator
Most CFOs will receive at least one FDE-style services pitch in the next six months — from OpenAI's Deployment Company, from Anthropic's services venture, from EPAM's 10,000-Claude-architect program, from Cognizant's secure AI services unit, or from Accenture/Deloitte/McKinsey responding to the threat with their own embedded-engineer offerings. Before any of them lands on the desk, the math has to be run.
Here is a decision rule grounded in the public cost data:
Step 1 — Set the engagement parameters.
| Variable | Typical Range |
|---|---|
| FDE annual cost (vendor's price to you) | $750K–$2.5M per engineer (loaded with margin) |
| Engagement length | 6–12 months |
| FDE team size | 2–6 engineers |
| Total engagement cost | $1.5M–$15M |
Step 2 — Quantify the in-scope value pool. Identify the specific finance workflows targeted: close acceleration, FP&A automation, AR/AP, treasury, audit prep, expense reconciliation, forecasting. For each, compute the dollar value of the time saved, errors avoided, and capital efficiency gained, on a three-year horizon (not one-year — FDE economics fail on one-year math).
Step 3 — Apply the 3x rule. A defensible FDE engagement should target a 3x return on engagement cost over three years. A $5M engagement should produce $15M of P&L impact over three years to be justifiable. If the value pool is below 3x, walk.
Step 4 — Compare to alternatives at parity. For each candidate workflow, build a parallel estimate for:
- Internal team build (2–3 senior engineers + PM + 12–18 months)
- Platform-based deployment (a no-code/low-code AI agent platform; industry data shows $77K Year 1 average vs. $228K for consulting-led equivalent — a 66% delta)
- Big Three SI engagement (Accenture/Deloitte/McKinsey traditional)
Step 5 — Score on five dimensions. Each candidate gets 1–5 across:
| Dimension | What it measures |
|---|---|
| Speed to value | Months to first measurable P&L impact |
| Total cost | All-in over 36 months including model, tools, services |
| Vendor lock-in | Cost and time to exit / switch |
| IP retention | Who owns the code, prompts, and agent definitions |
| Governance fit | How well controls, audit, and policy survive |
A score below 18/25 on the FDE path means a different delivery model is probably correct. A score above 22/25 means the engagement is worth running — but with a contractual right to retain code, prompts, and agent definitions on exit.
The most common mistake CFOs are making in early 2026: comparing an FDE engagement only against the "do nothing" baseline. That is a false comparison. The right baseline is the alternative delivery model that gets to the same outcome — internal team or platform — at lower cost and lower lock-in. The FDE wins on speed. It does not always win on total cost of ownership.
Framework #2: The CFO AI Readiness Triangle
The second framework is for the prerequisite question every CFO should answer before signing any AI services contract: are we ready to absorb an embedded engineering team? The data is unambiguous. Sixty percent of finance AI initiatives fail on organizational readiness gaps. The frontier labs cannot fix that for you, regardless of how senior the FDE is. Run this assessment first.
The Readiness Triangle scores three pillars, 5 points each, for a total of 15.
Pillar 1: Data Readiness (5 points)
| Question | 1 point each |
|---|---|
| Is your GL chart of accounts standardized and stable across business units? | Y/N |
| Are your top three finance data sources (ERP, CRM, HRIS) connected via documented APIs or a data warehouse? | Y/N |
| Do you have a defined data steward role with sign-off authority on schema changes? | Y/N |
| Is more than 80% of your finance reporting drawn from a single source of truth (not Excel pulls)? | Y/N |
| Have you completed a data quality audit in the last 12 months? | Y/N |
Pillar 2: Process Readiness (5 points)
| Question | 1 point each |
|---|---|
| Is your month-end close documented step-by-step with owners and time estimates per step? | Y/N |
| Have you identified your three highest-value automation targets with explicit ROI hypotheses? | Y/N |
| Do you have a defined change management process for finance system updates? | Y/N |
| Do you run quarterly process retrospectives with measurable cycle-time improvement targets? | Y/N |
| Have you piloted at least one finance automation (RPA, AI, or scripted) in the last 18 months? | Y/N |
Pillar 3: Talent Readiness (5 points)
| Question | 1 point each |
|---|---|
| Do you have at least one finance analyst with SQL or Python skills (not just Excel)? | Y/N |
| Does your finance team include a designated "AI champion" with executive sponsorship? | Y/N |
| Have you trained at least 25% of your finance team on prompt engineering or AI tool use in the last 12 months? | Y/N |
| Do you have a formal partnership with IT or data engineering with defined SLAs? | Y/N |
| Does your CFO sponsor a quarterly innovation budget independent of cost-saving targets? | Y/N |
Scoring:
- 0–6 (Red): Do not engage an FDE team. The deployment will fail on the customer side and you will be paying $5M+ to discover that. Spend the first six months on data hygiene, process documentation, and talent upskilling. Pilot a low-stakes use case (expense classification, invoice matching) with a SaaS tool first.
- 7–10 (Yellow): Engage a limited FDE engagement (2 engineers, 6 months, single workflow). Require explicit knowledge-transfer milestones in the contract.
- 11–15 (Green): Ready to engage a multi-FDE team across multiple workflows. Negotiate hard on IP retention, exit terms, and platform portability.
A Green score does not mean the engagement is justified — it means the organization is ready. The Framework #1 ROI test still applies. A Green organization with a sub-3x value pool should still pass on the FDE pitch.
The Competitive Landscape Right Now
The market for "embedded AI engineering" is consolidating fast. Here is the May 13 picture every CFO and CIO should have on the desk:
| Vendor | Structure | Capital | Key Differentiator |
|---|---|---|---|
| OpenAI Deployment Company | $10B JV, OpenAI-controlled | $4B raised (May 11) | 150 Tomoro FDEs, 19 PE/SI partners, GPT-class models |
| Anthropic Services Venture | Standalone, Blackstone/H&F/Goldman | $1.5B (May 4) | Mid-market focus, embedded model access, Claude-native |
| EPAM-Anthropic | 10,000-architect program | (terms undisclosed) | Largest Claude-trained engineering bench |
| Cognizant Secure AI Services | Wholly owned services unit | Internal | Governance-led, "agentic trust gap" positioning |
| Palantir Foundry FDE | Native FDE-led delivery | (legacy operation) | 15-year track record, defense/regulated focus |
| Accenture / Deloitte / McKinsey | Big Three response | Various partnerships | Brand trust, regulatory depth, breadth |
| Platform-based (Sierra, etc.) | No-FDE deployment | Sierra: $950M raised | 66% lower Year-1 cost, less lock-in |
The four questions Deepika Giri at IDC says every CIO should pressure-test in 2026:
- How modular can our architecture remain after this engagement?
- What is the cost and time to switch providers in 18 months?
- Which implementation partner controls the governance framework — and do we own the policies, or do they?
- Who owns the agent definitions, prompts, and integration code on exit?
Those questions are unwelcome in every FDE pitch deck. That is precisely why they need to be the first questions on the table.
The 90-Day CFO Action Plan
For finance leaders fielding services pitches in the next quarter, here is the sequence that protects optionality without slowing the AI program.
Days 1–30: Assess and prioritize.
- Run the Readiness Triangle internally. Document the score with evidence.
- Identify your top three workflow value pools with 36-month ROI estimates. Pressure-test against the 3x rule.
- Inventory the AI tools and platforms already in your finance team. Calculate what is in production vs. pilot vs. shelfware.
Days 31–60: Open a bake-off.
- Solicit proposals from at least three different delivery models: an FDE-led vendor (OpenAI, Anthropic, EPAM), a Big Three SI, and a platform-led alternative (Sierra, no-code agent platform, internal-build estimate).
- Require each to bid against the same scoped workflow with the same success metrics.
- Demand fixed-price or capped time-and-materials, not open-ended retainer.
Days 61–90: Contract for exit, not just entry.
- IP retention clauses: you own the code, prompts, and agent definitions.
- Knowledge transfer milestones at 30/60/90 days post-launch.
- Portability requirement: the deployment must run on at least two model providers (OpenAI + Anthropic, or one frontier + one open-weight).
- Governance: your audit framework wins. Their governance can supplement, not replace, your controls.
The CFOs who do this in 2026 will get the speed advantage of an FDE-led deployment without the long-term lock-in cost. The CFOs who skip these steps will discover in 2027 that their finance stack is functionally married to one frontier model, with switching costs measured in years and millions.
The Strategic Bottom Line
The $5.5B that OpenAI and Anthropic committed to enterprise services in May 2026 is the clearest signal yet that the frontier model business is no longer a model business. It is an implementation business, wrapped in a model wrapper, sold at enterprise services margins.
For CFOs, the implication is direct. The next AI vendor pitch on your calendar will not be a model demo. It will be a forward-deployed engineering team, ready to embed in your finance function for the next six to twelve months. They will ship code. They will accelerate close. They will demonstrate ROI. And they will, in the process, become structurally embedded in your enterprise stack in a way no SaaS vendor has been since the Oracle era.
The question is not whether to engage. The market has decided that question — finance is the beachhead, and the engagement is inevitable. The question is how to engage on terms that preserve your optionality while capturing the speed advantage the FDE model genuinely delivers.
Run the ROI calculator. Score the Readiness Triangle. Open a bake-off. Contract for the exit, not just the entry.
OpenAI just put 150 engineers inside your CFO office. Decide what they're allowed to take with them when they leave.
Continue Reading
- AI Vendors Just Bought the Consulting Industry. Same Day. — The May 4 launch of both Anthropic's and OpenAI's enterprise services ventures, with $11.5B in committed capital
- EPAM + Anthropic: 10,000 Claude Architects — How EPAM is responding to the FDE threat with its own Claude-native engineering bench
- Cognizant's Secure AI Services Bet — Governance-led positioning against the FDE model
- 78% Can't Pass an AI Audit. Alation Just Made It a Score. — The governance layer that should ride alongside any FDE engagement
- Anthropic Passes OpenAI in Enterprise AI — The deployment-velocity argument that funded the May services launches
Sources
- OpenAI Launches $4 Billion Company to Accelerate Enterprise AI Adoption (PYMNTS)
- The Second Coming of Secondments? FDEs Hit the CFO Office (PYMNTS)
- OpenAI Launches the Deployment Company (OpenAI)
- OpenAI Pushes Beyond Models With New AI Deployment Business (CIOL)
- OpenAI, Anthropic Expand Services Push, Signaling New Phase in Enterprise AI Race (CIO.com)
- Why Anthropic and OpenAI Are Copying Palantir's FDE Playbook (MindStudio)
- Palantir's FDE Model Drove 640% Returns — Now Anthropic and OpenAI Are Copying It (MindStudio)
- OpenAI and Anthropic Spent $5.5B on AI Consultants in 2026 (Beam)
- Palantir Forward Deployed Software Engineer Salary (Levels.fyi)
- Tech's Secret Weapon: 2026 Guide to the Forward Deployed Engineer (Hashnode)
- State of AI in Finance 2026 (CFO Connect)
- Top 5 AI Adoption Challenges Facing CFOs in 2026 (CFO Dive)
- So Far, Few CFOs See Substantial ROI from AI Spending (CFO.com)
- What Microsoft's New Research Tells CFOs About the ROI of AI (Fortune)
- Anthropic and OpenAI Both Launching Joint Ventures for Enterprise AI Services (TechCrunch)
