Earlier this week Anthropic confirmed what the WSJ had been reporting since Sunday: a new standalone AI services firm, backed by Blackstone, Hellman & Friedman, Goldman Sachs, General Atlantic, Leonard Green, Apollo, GIC, and Sequoia. And it's not just Anthropic — Bloomberg reported on the same day that OpenAI has finalised a similar JV with PE firms, apparently investing $4 billion at a valuation of $10B. This got me thinking — how does an investment of this size actually get turned into a company?
I posted a quick reaction on LinkedIn when this started circulating on Monday, and having had a few conversations over the past few days, I want to dig into it more — because I think there are some particularly interesting implications from each of the paths they may follow — and how this may impact the market.
Professional services businesses are human capital businesses. You can't simply capitalise one into existence. Blackstone's Jon Gray talked about "building a scaled, world-class company" — but scale in services means experienced people who can deliver, and those don't appear the instant you write the cheque.
Krishna Rao's line in the press release explains the pressure: "Enterprise demand for Claude is significantly outpacing any single delivery model." Anthropic can build relationships with Accenture, Deloitte, Cognizant — and they have, through the Claude Partner Network just launched in March with a $100M investment. Even measuring things in 'AI time' there's not yet been opportunity to assess the capacity these partners bring to Anthropic clients — but given these deals typically take many months to negotiate, the JV is best viewed as a parallel attempt to bridge the gap between demand and capacity. However, this is a tricky strategy to get right, as it risks disrupting this nascent ecosystem of certified partners before it has even been established.
Some context: the Avanade comparison
Before getting into the mechanics, it's worth framing what this kind of structure actually is. The closest precedent is Avanade — the JV Microsoft and Accenture formed in 2000 to build a scaled delivery vehicle for Microsoft enterprise technology. Historically, Microsoft's default posture, like that of many of the platform providers that came later (such as SAP, Salesforce and ServiceNow, or Google Cloud, and AWS), was to keep implementation at arm's length: certify and enable partners, but don't compete with them directly. Avanade was a deliberate departure from that — a dedicated vehicle that could operate more like the vendor's own team without Microsoft formally entering the services market itself. Anthropic is making the same structural choice.
Avanade became dominant in that space, but it didn't kill the competition in the independent Systems Integration (SI) ecosystem. Instead, it raised the floor. Firms competing on price or generalist capability got squeezed; firms with genuine vertical depth or proprietary IP continued to do well. It forced a strategy of differentiation and specialization.
Two markets — and why it matters for everything else
The JV's primary mandate is the PE portfolio companies of its investors. Blackstone alone has hundreds of portfolio businesses. Hellman & Friedman, and Goldman's AM book present the same picture. These companies need AI implementation; they have PE board members who can direct vendor selection; and they want a standardised, scalable playbook. That market is large enough to justify the capital structure on its own.
But the press release is careful to say the firm will also serve "independent companies that can benefit from the platform" — and that distinction is worth paying attention to, because those two markets want different things from a services firm. PE portfolio companies want speed, standardisation, and a clear line back to the investor relationship. Independent Fortune 500 companies, regulated industries with vendor neutrality requirements, organisations doing genuine R&D co-development — they want deep and specialised expertise and, often, the absence of entangled financial interests in the outcome.
The market dynamic should shape the approach, and guide Anthropic's strategy — if they try and serve it all through the JV, it leaves little room for the broader partner ecosystem to flourish. By instead making clear choices, there is a better chance of meeting all the demand. This would also recognise a tension that Anthropic will have to manage: the more the JV is seen as a PE-aligned delivery vehicle, the less credible it is as a neutral partner for enterprises that aren't in a Blackstone or Goldman portfolio. Those companies will still need someone — and that's the space they should be encouraging independent partners to be building towards.
So, with the JV as a focused capacity play, the how — how you actually build that capacity — now becomes the crux of the whole thing. My best read is a classic buy / borrow / build combination, and each approach again has interesting implications.
Buy
The Claude Partner Network now looks like something else in addition to a partner programme: a catalogued, certified acquisition list. Boutique firms with 20–100 engineers, genuine Claude expertise, and proprietary implementation patterns and tools are exactly what a JV needs to acquire to get off the starting blocks.
But there's a real danger in cannibalising the ecosystem. A market that is saturated by talk of acquisitions can be distracting and unsettling. If all the most promising partners are absorbed by the JV, this also risks sending signals about those partners that aren't included. Employees from those remaining companies may also see the JV as a more attractive option, given the direct access to Anthropic.
The two-market framing can be useful, as it also provides a constraint on which acquisitions actually make sense. A firm that's built deep expertise in, say, clinical trial optimisation or sovereign-grade government AI is probably not the right fit — the JV doesn't need to go there, and frankly there's more value for both those firms and Anthropic if they remain independent. The interesting targets are firms with strong horizontal implementation capability, repeatable delivery methodology, and ideally a track record across the mid-market or PE-adjacent sectors where the JV will spend most of its time. Firms that have specialised into the exact verticals the JV won't prioritise could be, perhaps counterintuitively, better positioned than they were before.
The next question is then one of size and number. Smaller companies will have less established operating systems — and likely will appreciate the support that the JV puts in place. However, bringing lots of small acquisitions together at once is not without integration challenges. Each company will have established cultures and ways of working — and the last thing the JV would want is to spend time, effort and resources focused internally, as people negotiate and jostle for position within the new entity. Acquiring a larger organisation would avoid some of the micro-negotiation, but there would be more baggage in terms of systems and tools, and it would come with a culture that would be even more established, and harder to shape if it's not already what the JV wants.
Acquisition has to be on the table — as it's the only way to deploy the capital in a relatively short timeframe. In terms of how best to do it, I'd be inclined to suggest a mixed approach — a mid-sized anchor, with smaller tuck-ins — is the way to balance the factors above. However, the wrong selections could lead to a compounding of errors, and drag things out. It may be worth considering extending the JV — in a more direct parallel with Avanade — where a larger SI is brought into the fold to help set things up well from the start, without it being a full acquisition.
Borrow
The press release confirms that "Anthropic engineering and partnership resources" are to be embedded directly within the new firm. This is what many folks refer to now as the Palantir model, which popularised the now-trendy "Forward Deployed Engineer" label — technical staff who live inside client engagements rather than operating at arm's length — and it's become something of a template for AI-native firms trying to bridge the gap between model capability and enterprise deployment.[1]
The tight coupling between the vendor and the delivery creates a feedback loop that more distanced partner models don't. When your implementation team is talking to the product team every week, the implementations get better faster. Clients get access to early capability. The vendor gets real-world signals about what breaks in production.
What makes this interesting — and slightly uncomfortable — for the broader partner ecosystem is that it means the JV won't just be better resourced than many independent partners; it will structurally have earlier access to model changes, tighter integration with Anthropic's roadmap, and a feedback channel that independent firms simply can't replicate. The firms that will feel this most acutely are the larger partners currently building CPN practices on the assumption that certification and co-marketing support puts them ahead of the field. They may discover that they're not on a level playing field — and the FDE model will be why. This may, in turn, push them towards seeking to join the JV themselves — which would solve the capital deployment problem but doesn't expand market capacity.
Furthermore, the FDE advantage points towards the JV being better equipped to take on the bigger and more technically challenging work than partners. This is directly contrary to the approach of focusing first on more standardised work for mid-market and portfolio clients — and puts pressure on the specialised boutiques. Anthropic will need to figure out a way to temper their desire to do it all, if they truly want to expand the services market.
Build
This is the hardest approach, and the one I'm most uncertain about. The obvious play is recruiting from the pool of engineers displaced in the 2026 Big Tech layoffs. There are genuinely talented people available, many of whom have been working with these models since before most consultants had heard of them.
But implementation consulting isn't the same discipline as building products inside a tech company. Enterprise delivery requires stakeholder navigation, discovery, scope management, and a tolerance for organisational friction that most Big Tech backgrounds haven't tested in quite the same ways.
At the same time, the services sector itself hasn't been immune to the same pressures — restructuring, layoffs, pricing model challenges. So the available talent pool is itself shaped by the forces the JV is trying to get ahead of. Build is the slowest path to deploying the capital, and probably the one most dependent on the other two working first.
What I'm watching
There's a defensive logic to this move that the press release doesn't state but that's hard to ignore. If the large SIs — Accenture, Deloitte, Cognizant, all quoted as CPN partners — become the primary relationship owners for enterprise AI, Claude risks becoming just another interchangeable model input. The JV is partly about Anthropic owning enough of the last mile that it can't be disintermediated. That makes the tension with those same SI partners less awkward and more structurally adversarial — and how Anthropic manages it will say a lot about whether the CPN was a genuine ecosystem bet or positioning ahead of this announcement.
The OpenAI $4B figure, if accurate, sharpens the picture further. This is a race — not just for talent and PE portfolio pipeline, but for the enterprise relationships that calcify over time. First-mover advantage in professional services is real; clients don't switch implementation partners lightly.
But the whole thesis depends on one thing neither JV can control: Claude — and GPT — staying meaningfully better than the alternatives long enough for those relationships to form and stick. If the models converge and enterprise buyers stop caring which one they're on, these are stranded assets. The services bet is also a bet on the research roadmap.
If you're building in this space — as a boutique partner, a buyer, or someone working out what this means for your own practice — I'd be interested to hear how you're reading it.
Sources: Anthropic/Blackstone/H&F/Goldman press release | Claude Partner Network announcement | WSJ | Bloomberg on OpenAI JV
But it's an older idea than Palantir. IBM pioneered the approach with its resident engineering programmes, which have roots back to hardware installation specialists in the 1950s. More recently, companies like Pivotal Labs and Slalom Build have approached this from the services side, building their entire reputation on embedded, opinionated delivery: co-located teams, pair programming, and a repeatable methodology that feeds back into the product. Engineers who embed deeply rather than advise from a distance. ↩︎