· AI
Goldman Sachs Is Financing Both Sides of the AI Deployment Land Grab
When the same bank funds both sides of a new market, the thing it is buying is not a winner. It is the toll booth.
In the first half of May 2026, Anthropic and OpenAI each stood up a multibillion-dollar outside venture built around the same idea: stop selling a model and a login, and start sending your own engineers to sit inside the client's building and build the thing yourself. Anthropic went first, forming a roughly $1.5 billion firm with Blackstone, Hellman & Friedman, and Goldman Sachs that places Anthropic engineers inside mid-size businesses to run their AI transformation, Fortune reported. Days later, OpenAI launched its own version, the OpenAI Deployment Company, with $4 billion of committed capital at a $10 billion valuation and backing from Goldman Sachs, Bain and Company, Capgemini, and McKinsey, tech.co reported. OpenAI bought its way into the talent with an acquisition of Tomoro, a UK applied AI consultancy, bringing about 150 forward-deployed engineers over on day one.
My position: the detail that should worry a buyer, not reassure them, is Goldman Sachs sitting on both cap tables. When one bank finances the two companies racing to own the same emerging category, the bank is not betting on which model wins. It is betting that whoever wins, it collects a fee on the delivery layer either way. That is a sign the category is being built to extract rent, not to compete it down.
Steelman: this solves a real, documented problem
Before I argue against it, the case for this move is genuinely strong, and I should state it plainly. The bottleneck is not invented. Christian & Timbers, an executive search firm that tracks this market, surveyed more than 250 executives at 180 companies and estimates demand for forward-deployed engineers will grow roughly 2,100 percent by the end of 2026, while putting the number of engineers with the sector knowledge to reliably deliver enterprise AI ROI at around 2,000 in the entire country. That figure comes from a firm with a direct commercial stake in the talent shortage it is describing, so treat the precise multiple as directional, but the underlying shortage shows up elsewhere too: the same reporting found the share of companies planning to hire for the role jumped from under 10 percent at the start of 2026 to 70 percent by mid-year. Blackstone's Jon Gray framed his firm's venture as addressing "one of the most significant bottlenecks to enterprise AI adoption," and Goldman's Marc Nachmann called it a way to "democratize access to forward-deployed engineers." If you cannot hire or retain the handful of engineers who can actually wire a frontier model into your real systems, renting one bundled with the model you already use removes a real coordination problem: no more pointing fingers between the software vendor and the integrator when something breaks.
Why the bundle should still worry you
Here is what the steelman glosses over. A forward-deployed engineer who is employed by the model vendor has an incentive structure that points away from your independence, not toward it. Their employer's business model depends on you staying a customer of that specific model, indefinitely. An engineer you hire or contract directly has the opposite incentive: finish the integration, document it, hand it to your team, and the engagement ends on a date you can name. Bundling the person with the subscription erases that distinction on purpose, because the whole point of a multibillion-dollar venture is to make switching costly.
Look at who supplied the capital. Blackstone, Goldman Sachs, Bain, Capgemini, and McKinsey are not builders competing on price. They are the institutions that have spent decades selling enterprises exactly the kind of long, expensive, hard-to-exit engagement that a free market in software was supposed to make obsolete. Billions in private equity and bank capital do not sit in a venture for free. That capital needs a return, and in a structure that bundles the engineer with the model, the only place that return can come from is the client's recurring bill. I want to be clear that this is my own inference about the economics, not a figure either company has published, but it follows from how private capital works: nobody commits $1.5 billion or $4 billion expecting to merely break even.
What this means before you sign anything
If a vendor offers you an embedded engineer bundled with their model, ask two questions before you sign: who does the engineer answer to when their advice and your interest diverge, and what do you own, in writing, the day the engagement ends. A clean answer to both is a sign you are buying engineering. A vague answer to either is a sign you are buying tenancy. That is the whole reason MojoSoftware builds engagements that end with your team holding the system, not a renewal notice from us or from whichever model we used to build it. If you are weighing a bundled offer against building that capability yourself, that conversation is worth having before you sign a multiyear contract to solve a problem that may only need one.
Sources
References used in this article. Links also appear alongside the relevant claims.
Let's make it happen.
You don't need a polished brief. A couple of lines about where you want your company to go is plenty, and we'll come back with what we'd tackle first.
We scope the work around your goals and timeline before quoting anything, so you know exactly what you're getting.