Palantir’s stock returned 640% over five years on the back of one idea: send engineers into the customer’s building, map the data, build the software, and don’t leave until it works. On May 11, 2026, OpenAI launched a $4 billion standalone company built on the exact same model.
The OpenAI Deployment Company (internally called “DeployCo”) is a majority-owned business unit backed by 19 investment firms, consultancies, and system integrators. TPG leads the investment round. Advent, Bain Capital, and Brookfield serve as co-lead founding partners. Goldman Sachs, SoftBank Corp., Warburg Pincus, B Capital, BBVA, Emergence Capital, and WCAS round out the financial side. McKinsey & Company, Bain & Company, and Capgemini join as consulting and integration partners.
The pitch: OpenAI will embed specialized Forward Deployed Engineers directly inside your organization to build production-grade AI systems connected to your data, your workflows, and your infrastructure.
This is not an API company anymore. It is a services company with a $4 billion balance sheet and the world’s most powerful consulting firms as founding partners.
19 Partners, and the Investor Mix Tells the Story
The financial partners (TPG, Goldman Sachs, SoftBank, Warburg Pincus) provide more than capital. They supply deal flow from their portfolio companies, which collectively span thousands of mid-market and enterprise organizations that need AI implementation but lack in-house expertise. Goldman Sachs alone manages over $2.8 trillion in assets; its portfolio companies are ready-made Deployment Company customers before the first cold call.
The consulting partners signal something more consequential. McKinsey, Bain & Company, and Capgemini are three of the world’s largest management consulting and systems integration firms. By joining as founding partners rather than standing aside as competitors, they gain first-mover access to OpenAI’s frontier models while OpenAI gains their enterprise client relationships and implementation methodology.
The arrangement creates an unusual dynamic. These firms are simultaneously OpenAI’s distribution channel and its direct competitor in the AI implementation market. The companies that built $375 billion in combined annual consulting revenue selling expertise by the hour now have a financial interest in a venture that aims to replace much of that work with AI-assisted engineering teams working on faster timelines.
The Tomoro Acquisition Gives OpenAI a Running Start
OpenAI didn’t build this team from scratch. The company agreed to acquire Tomoro, an applied AI consulting firm founded in 2023 that already runs Forward Deployed Engineering operations for major enterprises.
Tomoro brings approximately 150 experienced Forward Deployed Engineers and deployment specialists, along with a client roster that includes Tesco, Virgin Atlantic, Supercell, Fidelity International, Red Bull, Mattel, DPD, and the NBA. The firm built AI concierges for Virgin Atlantic, in-game support agents for Supercell, and AI deployment systems for Fidelity International’s investment operations.
The acquisition is subject to regulatory approval and expected to close within months. Once complete, those 150 engineers become the Deployment Company’s operational core from day one.
For context, Palantir employed roughly 200 forward deployed engineers when it went public in 2020. OpenAI is launching at comparable scale, but with an existing book of enterprise clients already running production AI workloads across retail, gaming, financial services, and logistics.
What Forward Deployed Engineers Actually Do (and Why an API Key Cannot)
The “Forward Deployed Engineer” term comes from Palantir, where it describes engineers who physically sit with customers, map internal data into a structured ontology, and build production software that the customer keeps running long after the engagement ends.
OpenAI’s version follows the same logic. FDEs work alongside a company’s leadership, technology teams, and frontline staff to identify high-impact AI integration opportunities, redesign workflows around AI capabilities, and connect OpenAI’s models to the company’s existing data, tools, controls, and business processes.
Denise Dresser, OpenAI’s Chief Revenue Officer, described the challenge: “AI is becoming capable of doing increasingly meaningful work inside organizations. The challenge now is helping companies integrate these systems into the infrastructure and workflows that power their businesses.”
McKinsey’s 2025 Global AI Survey found that while 88% of organizations report regular AI use, only one third have begun scaling AI programs enterprise-wide. That 55-point gap between experimentation and production deployment is the entire business case for the Deployment Company. Selling an API key takes five minutes. Getting a 50,000-person organization to change how it actually operates requires engineers in the building who understand both the models and the institutional context that no documentation captures.
The structural difference between DeployCo and a standard consulting engagement matters for buyers. DeployCo books its delivery engineers directly against OpenAI’s balance sheet, rather than routing through reseller partners the way Google’s Gemini Enterprise and IBM’s Anthropic distribution deals operate. OpenAI owns the customer relationship end to end, from the engineer’s desk to the model API call.
Where the Palantir Playbook Applies, and Where It Diverges
Palantir proved this model generates exceptional returns at scale. Its U.S. commercial revenue grew 137% year over year by Q4 2025, driven almost entirely by FDE-led deployments that turned pilot projects into enterprise-wide platforms. The company went from a $19 IPO price in 2020 to delivering 640% total returns over five years.
The lock-in mechanism is well understood. Once Palantir engineers have mapped an organization’s operational logic into the company’s Ontology layer and built production systems on top of that structure, removing the platform becomes a multi-year cost rather than a procurement decision. Switching costs compound with every workflow integrated.
OpenAI’s version carries the same structural advantage, plus one that Palantir never had: the underlying models improve continuously. Palantir’s software is powerful but largely static between major releases. OpenAI’s FDEs build on top of models that get meaningfully better every few months. A system designed around GPT-5.5 today will perform better on the next generation without the FDE team rebuilding it from scratch. This creates a compounding return: the embedded systems get smarter even when no one is actively working on them.
The risk is also familiar from Palantir’s history. FDE-driven deployments are capital-intensive and difficult to scale linearly. Each engagement requires senior engineering talent that is expensive and scarce. This is a high-margin business at maturity, but it demands significant cash burn during the growth phase.
Both Frontier Labs Are Running the Same Play
OpenAI is not alone. Anthropic, now valued at roughly $900 billion, launched a parallel enterprise deployment venture in May 2026, backed by Blackstone, Goldman Sachs, and Hellman & Friedman. That venture targets mid-market organizations that need Claude integrations but lack the internal AI engineering teams to build them.
Between the two frontier labs, billions of dollars are flowing into what amounts to AI-native consulting. The traditional consulting firms face a strategic fork: partner with the AI labs (as McKinsey and Bain chose to do with OpenAI) or risk losing their most profitable implementation contracts to teams that pair frontier models with embedded engineers who can ship production code in weeks rather than months.
The global management consulting market generates roughly $375 billion in annual revenue. The segment most exposed to disruption is technology implementation, which accounts for the majority of large consulting engagements. AI-native firms that deliver faster, more measurable outcomes with embedded engineering teams can undercut traditional billable-hour pricing on exactly this type of work.
What This Means If You’re an Enterprise AI Buyer
Running IT operations at a Saskatchewan telecom for two decades taught me that enterprise technology adoption follows a predictable pattern. The vendor demo looks transformative. The pilot project runs smoothly on a clean dataset. Then the production rollout hits 15 years of accumulated edge cases, undocumented integrations, and quiet organizational resistance, and the timeline doubles.
The Deployment Company model addresses the hardest part of that pattern by putting engineers in the building who live through those edge cases alongside internal teams. But three practical considerations matter for any buyer evaluating this option.
Pricing is still opaque. OpenAI has not disclosed the Deployment Company’s cost structure. Given Palantir’s historical pricing, expect six-figure and seven-figure annual contracts for embedded engineering teams, likely bundled with API consumption. Smaller organizations may need to wait for a mid-market tier that does not exist yet.
Lock-in is the strategy, not a side effect. The FDE model works precisely because it creates switching costs. Any enterprise that embeds an OpenAI engineering team deeply into its workflows will find migration increasingly expensive over time. Buyers should treat this as a platform commitment, not a vendor trial. The comparison is choosing AWS over Azure, not choosing a SaaS tool you can swap out next quarter.
150 engineers cannot serve the Fortune 500. The initial headcount from Tomoro is a starting point. Watch for aggressive hiring over the next 12 months. The constraint on this business is not demand; it is supply of engineers who understand both frontier AI agents and enterprise operations.
The AI industry’s center of gravity is shifting from model development to model deployment. Building a frontier model still costs billions. But the revenue, the margins, and the competitive moats now belong to whoever gets those models running inside the world’s largest organizations first.
OpenAI just bet $4 billion that the best way to do it is the oldest way: put people in the room.