Forward-Deployed Engineers: the New AI Moat the Mid-Market Can't Buy

Across the portfolio companies we deploy into, the pattern is the same and it is not a model problem. The model is fine. The order data is a mess, the ERP (enterprise resource planning system) hides half of what matters, and nobody on site has the time or the specialism to turn a capable model into a working system. The value was never locked inside the model. It was locked inside the gap between the model and the business. That gap is where deployment lives, and deployment is the whole game now.

The three largest artificial-intelligence platforms on earth just agreed with that, out loud, and put more than twelve billion dollars behind it in a single quarter. Not behind better models. Behind engineers who sit inside the customer and make the thing actually work.

The value was never locked inside the model. It was locked inside the gap between the model and the business.

The industry just voted, and it voted for deployment

Amazon Web Services (AWS) committed one billion dollars on 30 June 2026 to embed forward-deployed engineering pods inside customers, funded entirely from its own balance sheet. Small teams, roughly five or six engineers, working in cycles of around forty-five days, measured against shared business outcomes rather than billable hours. Its own vice president put the reason plainly: the currency customers keep asking for is speed.

That followed two nearly identical bets weeks earlier. OpenAI launched a Deployment Company valued at around ten billion dollars, raising over four billion from private equity partners including TPG, Advent, Bain Capital and Brookfield. Anthropic launched a roughly 1.5-billion-dollar joint venture with Blackstone, Goldman Sachs and Hellman & Friedman. Both copied the same playbook: put engineers physically inside the enterprise and hold them to a result, not a deliverable.

Read those three bets together and the message is unmistakable. The companies with the most to gain from selling you models instead have concluded that models are no longer the constraint. Blackstone's president said the joint venture exists to break one of the biggest bottlenecks in enterprise adoption: the scarcity of engineers who can implement frontier artificial intelligence at speed. When the people selling the models tell you the models are not the problem, believe them.

This is the same wedge we have argued from the start. The Massachusetts Institute of Technology (MIT) study that found roughly ninety-five percent of enterprise artificial-intelligence pilots produced little or no measurable impact did not blame the models. It blamed integration: legacy databases, authentication, data residency, the unglamorous plumbing. The bottleneck is deployment. Twelve billion dollars just confirmed it.

The pods are real, and they are pointed away from the mid-market

Here is the problem for a private equity portfolio. A forward-deployed engineer is now one of the most expensive people in technology. Compensation surveys put average total pay well into six figures, running to seven figures for principals at the frontier labs, while demand has grown roughly tenfold against a candidate pool that barely moved. Scarce, expensive people go to the highest-value accounts first. That is not cynicism, it is arithmetic.

So look at where the pods actually landed. The named launch customers for these programmes are the National Football League, the National Basketball Association, Cox Automotive, Southwest Airlines and the Allen Institute. A pod of six engineers on forty-five-day cycles is only economic against a very large contract. The maths points them at the enterprise, and it holds them there.

Meanwhile, only about seven percent of private-equity-backed companies have reached enterprise scale with artificial intelligence, according to FTI Consulting's 2026 survey of two hundred fund and operating leaders. The mid-market has exactly the problem the hyperscalers just spent twelve billion dollars solving, and none of the three solutions is built to reach a mid-market platform at a price that clears.

Scarce, expensive engineers go to the highest-value accounts first. That is not cynicism. It is arithmetic, and it points every pod away from the mid-market.

The honest counter-argument: one bet does aim at the mid-market

The sharpest objection to all of this is that one of the three bets is explicitly aimed at mid-size businesses, and specifically at private-equity-owned ones. Anthropic's joint venture names its partners' own portfolio companies as the first proving ground, across healthcare, manufacturing, financial services and retail, with engineers who sit down with clinicians and IT staff. So it is not true that every pod skips the mid-market. An informed reader will call that out, and they should.

But look at how that access is gated. Anthropic's mid-market reach runs through the balance sheets of Blackstone, Goldman, Hellman & Friedman and Apollo, into their portfolio companies. If a fund is not inside that club, the venture is not calling. It proves the mid-market is the prize. It does not make the capability available to a mid-market manager who is not one of those partners.

Two further things gate it. "Mid-size" in mega-fund language means the upper-mid-market, companies materially larger than a typical lower-mid-market platform, where US entry multiples sit around seven times earnings before interest, tax, depreciation and amortisation (EBITDA) on sub-billion-dollar enterprise values. And the tell is what a real mid-market fund did when it looked at the same choice. In June, a mid-market healthcare investor launched its own permanent, in-house artificial-intelligence and automation function, hiring two senior specialists into its value-creation team rather than waiting for a hyperscaler pod. It did not buy the capability. It built it, because nobody was selling it to a fund that size at a price that made sense.

So the framing is not "the pods skip the mid-market." It is sharper than that: the pods are gated, by account size for the hyperscalers, and by which sponsor's book you sit in for the joint ventures. A mid-market fund outside that club still has to solve deployment itself.

Build, attract, or borrow: the mid-market's real choice

That leaves a mid-market portfolio with three options, and elimination does most of the work.

Build in-house. Hire two or more senior specialists permanently, at six-figure all-in cost each, and carry them against a handful of portfolio companies. One fund just did exactly this. It works if you have the scale to keep them busy and the appetite to carry the overhead through a slow quarter. Most mid-market funds have neither. A permanent AI function amortised across five portcos is expensive insurance.

Attract a hyperscaler pod. Be large enough, and inside the right sponsor relationship, for a six-person pod on a forty-five-day cycle to be economic. A mid-market platform is, by definition, not that account. This option is closed before it opens.

Borrow it. Bring in a small, senior team that embeds on one portfolio company's own data for a few weeks, ships a working system, then moves to the next asset, without the permanent payroll or the minimum-account-size gate. This is the forward-deployed model, activated fractionally across the portfolio instead of dedicated to one enterprise account. It is the only one of the three a mid-market fund can actually buy.

Most mid-market funds cannot carry an in-house AI team and cannot attract a hyperscaler pod. The forward-deployed model still works. It just has to be borrowed, not owned.

What we do differently

This is the model we already run, and we ran it before the industry priced it at twelve billion. A small, senior team embeds on one asset's own data, works in weeks not quarters, and is measured on a business outcome the operating partner recognises on the P&L, not on a deck. We connect to the systems as they are, the messy order book, the legacy ERP, the spreadsheets, because that is what every industrial business actually has. No data-cleanup project on the client side, no eighteen-month programme, no army of junior consultants.

The difference from a dedicated enterprise pod is the economics. Because the capability is fractional, it clears against a single mid-market platform, and because it is fund-agnostic, it does not depend on which sponsor's book you sit in. Deploy on one portfolio company, prove the number, then replicate the foundation across the next. The intelligence compounds down the portfolio instead of resetting each time. That is how a mid-market fund gets the forward-deployed advantage the mega-funds just bought, without the overhead it cannot carry or the account size it cannot fake.

The proof

This is not a theory we are selling. It is the model we run.

We embedded on a European industrial distributor's own order data and surfaced forty-five million euros of at-risk revenue their sales team could not see, in six weeks. In the four weeks after launch, they reduced that at-risk revenue by over ten million euros. It is not a pilot sitting in a slide deck. It is a production system the sales team uses every day.

That is a forward-deployed engineering pod operating at mid-market scale, on one asset, against a number the operating partner can defend, delivered without the fund having to hire an AI team or attract a hyperscaler.

Forty-five million euros of at-risk revenue found in six weeks, over ten million of it reduced in the next four, in a system the sales team now uses every day. Not a pilot. A production system.

Why this ladders to DPI

Financial engineering is exhausted. With average holds now past six years and thousands of portfolio companies sitting unsold, returns come from the operating plan or they do not come at all. Bain and McKinsey frame it as "twelve is the new five": deals now need roughly ten to twelve percent annual operational EBITDA growth to deliver what about five percent gave when cheap leverage did the work.

Protected and grown revenue is EBITDA. A better-instrumented, better-run, de-risked business is a more sellable one, and it carries a higher exit multiple. Higher multiples convert to distributions to paid-in capital (DPI), the number limited partners now judge a fund on more than internal rate of return. Strong DPI raises the next fund. That is the chain, and deployment is the first link. The mega-funds have secured that link for themselves and for the companies inside their club. The forward-deployed model, borrowed fractionally, is how a mid-market fund secures it too.

If you want to see what that looks like on one of your portfolio companies, a rapid read of where revenue is at risk and where the growth sits, start here or book a call.