Why Anthropic and OpenAI Are Copying Palantir's Forward-Deployed Engineer Playbook
Palantir dropped to $6 in 2022 then returned 640% in 5 years. Now both major AI labs are cloning its FDE deployment model for enterprise.
Palantir Figured Out Enterprise AI Deployment Before Anyone Called It AI
Palantir IPO’d at roughly $19 in September 2020, slid to around $6 by late 2022, and then quietly returned 640% over the following five years. That trajectory isn’t just a stock story. It’s a proof of concept that both Anthropic and OpenAI are now explicitly copying.
The model that drove Palantir’s recovery was the Forward Deployed Engineer — the FDE. And if you want to understand why two of the most valuable AI companies in the world just announced major enterprise deployment ventures within weeks of each other, the FDE is where you start.
Last week, Anthropic announced a joint venture focused on deploying enterprise AI services, backed by Blackstone, Hellman & Freeman, and Goldman Sachs as founding partners. The venture is valued at $1.5 billion, with a $300 million founding commitment split between Anthropic, Blackstone, and Hellman & Freeman. Additional backing comes from Apollo Global Management, General Atlantic, GIC, Leonard Green, and Suko Capital. Almost simultaneously, OpenAI is raising $4 billion from 19 investors for something it’s calling the “development company,” at a $10 billion valuation. There is reportedly zero investor overlap between the two ventures.
Both of them are, in essence, building FDE machines at scale.
What the Forward Deployed Engineer Actually Is
Day one: idea. Day one: app.
Not a sprint plan. Not a quarterly OKR. A finished product by end of day.
The standard software sales motion goes like this: build product, hand it to sales, sales sells it to the customer, customer tries to install it with maybe some help from a customer success team. Then the customer mostly figures it out alone.
Palantir broke that model. Instead of handing off the product and walking away, they took their best engineers and embedded them directly inside the customer’s organization. These weren’t account managers writing up documentation. They were shipping real code, building custom integrations, configuring the actual system to work inside the customer’s specific environment. Forward deployed, as in: your engineers are now operating from inside the client’s walls.
The insight behind it is deceptively simple. The Palantir engineer knows everything about how the software works. The JP Morgan engineer knows everything about JP Morgan — the data structures, the compliance requirements, the internal politics, the specific problem they’re actually trying to solve. Neither one can succeed alone. The FDE model forces those two knowledge sets to collide in the same room until something actually works.
This approach is especially well-suited to customers with what you might call weird and complicated problems: hospitals, banks, defense agencies, large financial institutions. These organizations have requirements that off-the-shelf SaaS can’t accommodate. They have legacy systems, regulatory constraints, and internal processes that were never designed with AI in mind. The FDE doesn’t sell them a product and leave. The FDE builds the thing that works for them specifically.
Why the AI Labs Have a Deployment Problem
Here’s the tension that makes the FDE model so relevant right now. The models are genuinely capable. The benchmarks keep improving. Anthropic’s ARR reportedly exploded from $9 billion to over $44 billion in 2026, with SemiAnalysis reporting it’s been doubling roughly every six weeks. Analyst Ming Li calculated that Anthropic is adding $96 million in ARR per day. Inference margins have jumped from 38% to 70% in roughly a year. AWS took 13 years to reach $35 billion in annual revenue. Salesforce took over 20 years to pass $20 billion. The model capability story is not the problem.
The deployment story is.
Anyone who has actually tried to implement AI agents inside a real business knows the gap between “this demo is incredible” and “this is running reliably in production.” The harness — the scaffolding, the integrations, the databases, the tooling that wraps around the model and makes it actually do something useful — is where most enterprise AI projects stall. It’s not that the models can’t do the work. It’s that connecting the model to the work requires skills that are genuinely scarce right now, and that most enterprise IT teams don’t have.
This is the deployment gap. And it’s the specific problem that the FDE model was designed to solve.
Not a coding agent. A product manager.
Remy doesn't type the next file. Remy runs the project — manages the agents, coordinates the layers, ships the app.
The irony is that the AI labs have spent years building increasingly capable models while the bottleneck has quietly shifted downstream. You can have the best model in the world and still fail to deploy it usefully inside a hospital or a bank, because the hospital’s data is in three different legacy systems, the bank has compliance requirements that change the architecture entirely, and neither institution has the internal talent to bridge the gap between what the model can do and what the institution actually needs. The real-world performance differences between frontier models matter far less than most enterprises expect — the deployment gap swamps the capability gap in practice.
What the New Ventures Are Actually Building
Anthropic’s joint venture is targeting the financial sector first, which makes sense given the investor lineup. Blackstone is one of the largest alternative asset managers in the world. Goldman Sachs is Goldman Sachs. These aren’t passive investors — they’re also the first customers, and they’re bringing their own networks of portfolio companies and clients into the deployment pipeline.
The structure is worth paying attention to. This isn’t Anthropic licensing Claude to Blackstone and wishing them luck. The $300 million founding commitment and the joint venture structure suggest something closer to a shared deployment operation — a machine for getting AI actually installed and running inside complex financial institutions, with Anthropic’s technical expertise combined with Blackstone’s institutional access and operational knowledge.
OpenAI’s “development company” is operating at a larger scale — $4 billion from 19 investors at a $10 billion valuation — and appears to be targeting a broader range of industries including manufacturing and healthcare, not just finance. The zero investor overlap between the two ventures is interesting. It suggests the two companies are carving out different institutional networks rather than competing for the same limited pool of enterprise relationships.
Both of them are, in different ways, trying to solve the same problem Palantir solved: how do you get a genuinely capable but technically complex system actually deployed and running inside organizations that weren’t built for it?
For builders thinking about what this means for their own work, the strategic differences between Anthropic, OpenAI, and Google on agent deployment are worth understanding — because the FDE model is really a bet on one particular theory of how enterprise AI adoption happens.
The Non-Obvious Thing Buried in This
The FDE model is sticky in a way that most SaaS isn’t.
When a company installs a CRM, they can theoretically migrate to a competitor. It’s painful, but it’s possible. When a forward deployed engineering team spends six months building a custom AI system that’s deeply integrated into your internal data, your workflows, your compliance architecture — that system becomes load-bearing infrastructure. You don’t rip it out. You depend on the lab that built it for continued maintenance, updates, and improvements.
This is the strategic logic that makes the FDE model so attractive to Anthropic and OpenAI right now. The enterprise AI market isn’t just about selling tokens. It’s about becoming the infrastructure layer that large institutions can’t easily remove. The FDE is how you get there.
One coffee. One working app.
You bring the idea. Remy manages the project.
There’s also a timing dimension. The hyperscaler CapEx numbers suggest the infrastructure buildout is accelerating, not slowing. Morgan Stanley raised its forecast for the five major hyperscalers to $805 billion in CapEx for 2026, and $1.1 trillion for 2027. The Mag 7 companies spent over $400 billion in CapEx in Q1 2026 alone, with a reported and projected backlog of around $1.3 trillion. That backlog number is the tell — demand is substantially outpacing supply, which means the constraint isn’t model capability or compute availability in the long run. It’s deployment.
The companies that figure out how to deploy at scale, inside complex institutions, with the kind of custom integration work that actually makes the systems useful — those companies capture the value that the infrastructure buildout is creating. The FDE model is a bet that deployment expertise is the scarce resource, not model capability.
This also reframes the shift from seat-based to token-based pricing in an interesting way. In the FDE model, you’re not selling seats. You’re selling a deployed system that generates ongoing token consumption as the institution uses it. The stickiness of the deployment is what makes the token revenue durable.
The Palantir Parallel Is More Specific Than It Looks
It’s tempting to read the Palantir comparison as a loose analogy — “enterprise software company figures out how to sell to big institutions, AI labs copy the playbook.” But the parallel is more specific than that.
Palantir’s core insight wasn’t just “embed engineers in customer organizations.” It was that the problems worth solving — the ones with the most money attached to them — are the ones that are too weird and too specific for off-the-shelf software to handle. The FDE model is optimized for exactly those problems.
The AI labs are now making the same bet. The financial institutions, hospitals, and government agencies they’re targeting are precisely the organizations with the most complex, most custom, most compliance-heavy requirements. They’re also the organizations with the most money and the highest stakes. A hospital that successfully deploys AI in its clinical workflow isn’t going to switch vendors because a competitor offers slightly better benchmarks. A bank that has Anthropic’s engineers embedded in its risk management systems isn’t shopping around.
For builders trying to understand where the enterprise AI market is heading, this is the pattern to watch. The labs are not just building better models. They’re building deployment machines designed to capture the most complex and most valuable institutional customers — and to make those relationships structurally difficult to unwind.
Platforms like MindStudio are relevant here for a different reason: they give builders the orchestration layer — 200+ models, 1,000+ integrations, and a visual builder for chaining agents and workflows — that lets teams prototype and deploy AI applications without writing the orchestration infrastructure from scratch. That’s a different market than what Anthropic’s FDE venture is targeting, but it’s the same underlying problem: the gap between model capability and working deployment.
What Atlassian’s Earnings Reveal About the Broader Pattern
One data point that doesn’t get enough attention in the context of enterprise AI deployment: Atlassian’s stock was up nearly 30% on its most recent earnings report. Revenue grew 32% year-over-year, up from 23% the prior quarter. The driver wasn’t just growth — it was the adoption of Rovo, their AI search tool built natively into Jira and Confluence.
CEO Mike Cannon-Brookes noted that customers using Rovo were growing their own ARR at twice the pace of those who weren’t. The reason Rovo works better than a generic RAG-based AI search tool is that Atlassian has spent 20 years capturing structured relationships between work, teams, people, code, and knowledge inside Jira and Confluence. When Rovo needs context, it does a graph lookup instead of a vector dump. It uses far fewer tokens to get to a better answer.
This is the same logic as the FDE model, applied at the software layer. The value isn’t in the model. It’s in the integration — the deep, structural connection between the AI and the institution’s actual data and workflows. Atlassian built that integration over 20 years. Anthropic’s FDE engineers are trying to build it in months, for institutions that don’t have a Jira.
The Atlassian case also suggests something about what successful enterprise AI deployment looks like from the outside: it shows up in the customer’s growth metrics, not just in the vendor’s revenue. Customers using Rovo are growing faster. That’s the proof point that makes enterprise AI sticky — not benchmark scores, but measurable impact on the customer’s own business. The coding performance differences between frontier models that dominate benchmark coverage are largely irrelevant to this dynamic; what matters is whether the deployed system moves the needle on the institution’s actual KPIs.
What to Watch For
The FDE model has a scaling problem that Palantir navigated slowly and that the AI labs will need to navigate much faster. You can’t just hire unlimited forward deployed engineers. The value of the FDE is that they’re genuinely expert — they understand both the model capabilities and the customer’s specific environment. That expertise takes time to develop, and it doesn’t scale the way software does.
The interesting question is whether the AI labs can use AI to accelerate the FDE model itself. If an FDE can use AI tools to compress the integration and deployment work that used to take months into weeks, the model scales differently. Tools like Remy point at one version of this: you write a spec — annotated markdown describing what the system needs to do — and Remy compiles it into a complete TypeScript stack with backend, database, auth, and deployment. The spec is the source of truth; the code is derived output. If FDEs can work at the spec level rather than the code level, the deployment timeline compresses significantly, and the scaling constraint on the FDE model loosens.
The other thing to watch is whether the zero investor overlap between Anthropic’s and OpenAI’s ventures holds. Right now, they appear to be carving out different institutional networks — Anthropic in finance, OpenAI broader. If those networks start to overlap, the competition shifts from model capability to deployment relationships. That’s a different kind of moat, and it’s one that’s much harder to evaluate from the outside.
Palantir spent years being dismissed as overvalued, complicated, and too dependent on government contracts. Then it returned 640% in five years. The AI labs are betting that the same model — embed your best people inside the most complex institutions, build systems that are structurally difficult to remove — works even faster when the underlying technology is improving exponentially.
They’re probably right. The question is who builds the deployment machine fast enough to matter.
