If you're building an AI services company, this week handed you a lot to think about. Both Anthropic and OpenAI announced partnerships with private equity firms to launch new entities variously described as "AI Services Company" and "Deployment Company." Every founder conversation we had this week touched upon it in some way. Some read the announcements as market validation. Others are worried they're about to compete with the very companies whose models they're building on.

Both announcements have a few things in common: enormous headline numbers, deliberately vague descriptions of what will actually be built, and Goldman Sachs somewhere on the cap table. But the structures are meaningfully different. Axios did the best job summarising the specifics:

Both companies describe a similar operating model: small teams of their forward-deployed engineers working alongside customers' own engineering staff to develop custom implementations. As Anthropic describes it, a typical engagement starts with understanding where Claude can have the biggest impact, then building out Claude-powered systems tailored to each organisation's operations. The basic bet is that AI adoption is no longer mainly a model-quality or compute problem. The bottleneck is now that companies lack the workflows, data access, security rules, operating discipline and, frankly, the imagination to install AI safely inside real business processes.

Both announcements were explicit on some details and deliberately vague on others. Anthropic said its entity will focus on mid-market companies. OpenAI indicated it might use the capital to acquire consulting firms. Neither mentioned who will lead the new ventures, how they'll handle competitive situations with existing partners, or what happens to the change management side of workforce transformation.

The first question that comes to mind: why do you need these amounts of capital to build what is, at its core, a consulting business?

Building a consulting firm is not typically capital-intensive. Find a client, bill a client, deliver work, pay your people, pocket a decent margin, repeat. The history of the industry bears this out in an almost embarrassing way.

McKinsey and BCG never raised outside capital. Bain did raise some money, but only in response to partners cashing out roughly $200 million and saddling the firm with the debt, which required a financial rescue. Slalom, the largest stand-alone pure-play technology consultancy, was bootstrapped from its 2001 founding. Its first outside round came in 2011 - ten years in. Total capital raised across four rounds sits at around $7.67 million, which is a rounding error against its revenue. Accenture and Capgemini have raised money in public markets, but at a much later stage. 

What Anthropic and OpenAI are doing - raising vast amounts of capital for a consulting entity that is pre-revenue and pre-product - is genuinely unprecedented in professional services.

One possible answer could be inveting in go-to-market. Enterprise software vendors have notoriously large and expensive GTM teams, often spending more money on selling their product than future R&D. But this is where PE involvement is supposed to help the most. PE-backed US companies employ around 13.3 million workers and contribute approximately 7 percent of US GDP, per the EY/American Investment Council report. Harvard Law professor John Coates estimates private equity controls between 15 and 20 percent of the entire US economy when you account for indirect influence and AUM. Having large, highly incentivised shareholders who can open portfolio company doors to this cohort of companies removes the need for a conventional enterprise sales machine. In theory.

But I think the real answer here is that a large chunk of the $5.5bn raised by Anthropic and Open AI will go towards each other.

It is increasingly clear that competition at the frontier model layer is no longer primarily about model accuracy.

According to Technobezz, as of early 2026, Anthropic's Claude - which entered this market as the underdog - is winning roughly 70% of new business matchups against OpenAI in the professional sector. Viewed alongside the 17.5% guaranteed coupon in the OpenAI deal, the OpenAI warchest starts to read less like a strategic initiative and more like a defensive move: an attempt to catch up to Anthropic's enterprise traction through sheer speed and capital deployment.

The bet both companies are making is that once a frontier model is implemented and embedded inside enterprise workflows, it becomes extremely difficult to displace. That bet is timely, given what is happening with open-source models, which we'll come back to later on.

An interesting detail that hasn’t gotten it’s fair share of attention is the acquisition question. Reuters reported that Open AI is in discussions to acquire existing consulting firms, and is in advanced stages on three deals. I can’t imagine that Anthropic is far behind. Multiple people involved in the JVs have cited the scarcity of engineers who can implement frontier AI systems as "one of the most significant bottlenecks to enterprise AI adoption" – a characterisation that came directly from Blackstone President and COO Jon Gray. If you can’t hire and train FDEs fast enough, acquiring an existing team is smart move. 

But who's actually available to buy? The candidate list is thinner than the JV’s deal sizes would imply:

The realistic acquisition landscape is narrower than the headlines suggest.

Another aspect that hasn’t received much attention in the mainstream commentary is the competitive threat both companies are facing from open source models.

An AI product founder I spoke with recently described open-source models as "a few months behind" the frontier, and said he and other founders are watching them very closely. Cost is going to become a much larger selection factor as model quality converges. My colleague Dylan recently visited China to compare notes with local entrepreneurs and investors. One pattern that emerged from those conversations: companies there are using Western models for strategic or sensitive projects, then switching to open-source and local models later in the deployment cycle to optimise cost. I suspect a variation of that pattern will play out more broadly.

And then there is the fundraising question. Much has been written about the disconnect between the capital raised by frontier model developers, the infrastructure investments made, and the actual pace of enterprise adoption. Bain estimates that feeding AI's compute appetite will require more than $500 billion per year in global data-centre investment by 2030 - which in turn requires roughly $2 trillion in annual revenue to be viable. Even under generous assumptions, Bain estimates the AI industry will come up approximately $800 billion short.

Both OpenAI and Anthropic are heading toward IPOs. Both will face the scrutiny of a much broader investor base - one that, unlike the current crop of late-stage venture funds, will not simply take the growth story at face value. The last thing Dario Amodei and Sam Altman need is a running battle with short sellers. Going into IPO roadshows with concrete enterprise case studies and durable, low-churn revenue - as opposed to high-turnover consumer subscriptions - makes the narrative considerably more credible.

Private equity firms involved in the JVs are definitely not passive observers here; they are under their own pressure. The shift from financial engineering to operational value creation has left many funds struggling. Software company valuations have come under sustained pressure from the AI threat, and many funds have had to mark down holdings in their existing portfolios - holdings their limited partners are already questioning. All else considered, the 17.5% guaranteed return in the OpenAI deal is, to put it plainly, rather a good outcome for the funds involved. It is not an especially good outcome for OpenAI and it is worth asking why OpenAI agreed to it.

A compelling narrative about how the funds are going to use AI to rescue its current portfolio - while also participating in the trend as an investor - is potentially the difference between a successful fundraise and a difficult one.

Most of the commentary so far has centred on whether these announcements represent a replication of Palantir's playbook. I think that framing is wrong, and it's worth being precise about why.

Palantir is a technically complex product that primarily interfaces with technical users. Frontier models have much broader applications and touch far more non-technical people. The most widespread current use case currently is customer support; sales is likely next. In that context, technical complexity is not the dominant variable. Intimate knowledge of specific workflows and, critically, change management capability are equally important.

The English textile workers who destroyed machinery at the start of the industrial revolution - the Luddites - did so out of fear that automation would take their livelihoods. You can feel a version of that same anxiety among today's front-line workers. Without a carefully designed change programme, the organisation will reject the transplant. That is not a problem Palantir had to solve. It is a problem OpenAI and Anthropic will have to solve, and nothing in either announcement addresses how they plan to go about it.

Palantir also does not work with external system integrators and cannot be purchased off the shelf. It has never had to navigate the tension between in-house delivery, partner collaboration, and encouraging clients to eventually handle deployment independently. I suspect Open AI and Anthropic’s instinctive answer is that deploying powerful technology in the real world justifies any approach that might stick. But there is no precedent for another company having done this successfully at scale, so that instinct remains untested.

The more instructive precedent is Avanade. Launched in April 2000 as a joint venture between Microsoft and Accenture, Avanade is now majority-owned by Accenture and, 26 years on, remains the primary delivery vehicle for the Microsoft ecosystem (incl. Copilot, agentic AI, and security). It has won the Microsoft Global SI Partner of the Year award 20 consecutive times. By any durability and revenue measure, it is a success: roughly 60,000 staff, multi-billion-dollar revenue, two decades of channel dominance. The harder question - whether Microsoft would have been better served by owning more of it, or by building services capability in-house rather than letting Accenture capture most of the value - is genuinely debatable. And unlike the PE funds now backing OpenAI and Anthropic, Accenture was already in the business of technology consulting on day one.

I doubt many of you reading this would have been aware of it - I definitely wasn’t before researching this story. Avande aside, the prevalent strategy for other other software giants such as SAP and Salesforce has been to partner with third-party firms rather than build implementation capacity in house. For every dollar of software those companies sell, partners generate between five and eight dollars in adjacent services revenue - we covered it in detail in a previous issue. That model works because it draws on decades of relationship depth and process intimacy that no software vendor builds overnight. Both OpenAI and Anthropic have formal relationships with large system integrators and top-tier strategy consultancies. The announcements were vague on how those relationships will be affected. Anthropic's explicit focus on the mid-market has been read by some - including AI Daily Brief - as a deliberate signal that the enterprise segment remains the domain of existing SI partners.

Taken together, this gives me less the feeling of a deliberate, thought-through strategy, and more the feeling of Werner Herzog's "Every man for himself and God against all."

If you're building an AI enablement or services business, this is not the moment to be intimidated. What you are building matters, and it matters a great deal. Let me be direct about why.

The math alone argues against a zero-sum reading of these announcements. Bain's estimate of $2 trillion in annual revenue required to justify the infrastructure build - against an industry currently far short of that - means that there is far more demand to be activates than the new JVs alone can capture. It is not in the interest of OpenAI or Anthropic to shut partners out. The economics of AI infrastructure adoption require a large, distributed ecosystem of implementation capacity.

In more practical terms:

Let’s keep building.

– Daria