The Real Reason OpenAI Is Building a $4B "Deployment Company"
TL;DR
OpenAI’s new “deployment company” is really a massive services play — the unit launches with $4 billion in backing from TPG, Advent, Bain Capital, Brookfield, and Goldman Sachs, plus 150 forward deployed engineers from its acquisition of Tomorrow.
FDEs are basically technical consultants embedded inside customers — Paul Roetzer says the simplest translation is “yes, it’s a consultant,” but one who can redesign workflows, build agents, and price against business outcomes instead of software seats or tokens.
The real prize isn’t just software revenue — it’s the services and labor market around AI adoption — Roetzer frames this as vendors chasing the old “$1 in software drives $6 to $10 in services” dynamic, with an even bigger target of roughly $4 trillion to $6 trillion in U.S. knowledge-work wages.
Everyone is racing because enterprises can’t absorb agentic AI fast enough on their own — Google Cloud is hiring more FDEs, Salesforce has committed to 1,000 of them, and Indeed/Financial Times data showed postings for the role jumped more than 800% from January to September 2025.
The hype around FDEs risks missing the harder half of AI transformation — Ali K. Miller’s pushback is that deployment alone is not the plan, because change management, communication, education, and training are often the actual blockers inside companies.
This creates a real opening for both new grads and specialized service firms — Aaron Levie argues college counselors should point CS students toward FDE paths, while Roetzer says agencies can win by going all-in on one platform, like OpenAI, Anthropic, or Google Cloud, before those vendors hire the talent away.
Summary
Why “forward deployed engineers” suddenly took over the AI conversation
The episode opens with the role du jour: forward deployed engineers, or FDEs, engineers who sit inside customer organizations and help design and ship AI systems with frontline teams. The catalyst was OpenAI’s new “deployment company,” a business unit built to identify valuable workflows, redesign processes, and turn AI gains into durable production systems.
OpenAI and Google both made their move in the same week
OpenAI’s effort starts with more than $4 billion in investment and heavyweight backers including TPG, Advent, Bain Capital, Brookfield, and Goldman Sachs, plus consulting partners like Bain, Capgemini, and McKinsey. It also acquired Tomorrow, an applied AI consulting firm serving clients like Tesco, Virgin Atlantic, and Supercell, instantly adding 150 experienced FDEs; meanwhile, Google Cloud CEO Thomas Kurian announced a parallel AI-focused org and tied it to Google’s previously announced $750 million ecosystem commitment.
Paul’s blunt translation: “Isn’t this just a consultant?”
Roetzer says yes, basically — it’s a consultant with technical depth who can customize AI models and agents to solve business problems. The twist is pricing: instead of billing for software access, these teams can walk into a company, identify a problem worth, say, $100 million, and charge $25 million on an outcome basis, which he says is a massive business-model shift.
The Trojan horse behind all this hiring
He argues the vendors won’t say it directly, but they are absolutely coming for traditional consulting work. Even if OpenAI, Salesforce, and Google frame this as partner-friendly, Roetzer calls FDEs a “Trojan horse” driven by two realities: enterprises genuinely need help adopting AI, and model companies need huge new revenue streams to justify valuations and future IPO stories.
Salesforce offers the clearest blueprint for how this works
Roetzer pulls from a March 2026 Salesforce post describing FDEs as a hybrid of “personal tech guru, business consultant, and handholder.” He highlights the pod structure: one deployment strategist plus two FDEs, focused full-time on one client for about three months, sometimes on-site, with strategists finding the use cases and engineers building the agents.
Agentic AI made this feel urgent, but the real bottleneck is people
The reason this is accelerating now, he says, is that agentic AI turns software into something closer to work output, so vendors must understand the customer’s business process end to end. That’s why Aaron Levie says agent deployment is more complex than normal software rollouts — but Roetzer sides with Ali K. Miller too, emphasizing that nontechnical work like education, training, communication, and change management may be just as important, if not bigger.
The HubSpot lesson: software companies always drift toward services
Roetzer uses his own history running a HubSpot agency to explain the pattern: if a vendor can’t trust the ecosystem to deliver consistent value, it starts doing onboarding and services itself. He says AI vendors don’t have time to wait for a mature partner ecosystem, so they’re hiring directly now — not only competing with consultancies for revenue, but also for the exact talent those firms would otherwise recruit.
What this means for careers, agencies, and AI Academy
By the end, the conversation turns practical: yes, this could be a great path for computer science grads, but also for AI-forward marketers, operators, and advisors who can bridge business and technical work. Roetzer says his own team is building something similar around AI Academy — not just courses, but advisory services for real transformation — and closes with a warning for service firms: there’s huge opportunity if you go deep on one platform, but the risk is the platform company hires your best people.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.