For two years the big consulting firms were among the frontier labs’ biggest customers. Over one week in May 2026, the labs set up as their competitors. On May 4, Anthropic stood up an enterprise services firm with Blackstone, Hellman & Friedman and Goldman Sachs. Seven days later, OpenAI announced The Deployment Company, a business it controls that raised over $4 billion from 19 investors co-led by TPG, Advent, Bain Capital and Brookfield. TechCrunch covered the two as one story.

A normal supplier sells a product and leaves. These suppliers embed their own engineers inside your work, and the thing they most want is the one asset you cannot take back once it has been seen, which is how the work is really done. The firm that partners to gain an edge can be the firm that gives it away.

For eleven years I worked out the largest distressed corporate loans a major bank held, across dozens of industries, from airlines to steel mills. The recoverable value was rarely the equipment on the books. It was whether anyone still understood how the business actually ran. That understanding is the asset. It is what a consulting firm sells, and it is what these new arrangements are built to observe.

Both at once, and the easy half

Call the alliances a Trojan horse or a survival strategy. They are both, and the survival half is the easy half. A firm rents reasoning it could never build in-house and finally has an answer for the client who keeps asking what it is doing about AI. The half that gets missed is the Trojan horse. The same arrangement can quietly turn a firm’s proprietary data and hard-won expertise into the lab’s own product, because the lab that learns how your engagements work ends up holding most of what it needs to run them without you. What keeps your edge yours is not the lab’s goodwill. It is the structure of the contract you sign.

How deep the alliances already go

The alliances are deep, which is what makes the risk real rather than theoretical. In May 2024, PwC became OpenAI’s first ChatGPT Enterprise reseller and its largest enterprise customer, with more than 100,000 seats. In October 2025, Deloitte began putting Claude in front of about 470,000 employees, Anthropic’s largest enterprise deployment. The firms that went furthest fastest are the ones with the most of their own method now running on someone else’s model.

The labs are short of exactly what you hold

To see why that is a problem, follow what the labs are short of. The open web that trained the first models is largely spent, and the scarce input now is fresh expert knowledge. Watch what investors pay for it. Meta paid $14.3 billion for 49% of Scale AI. Mercor, which pays professionals to generate training data, reached a $10 billion valuation in October 2025, eight months after it was valued at $2 billion. Bloomberg reported the bootstrapped Surge AI in talks to raise at $25 billion or more. Those are prices for labelled data produced to order.

The unlabelled version, the live operating knowledge inside a working business, is the thing a lab would most want and the one thing it cannot buy in a funding round. My read is that it is worth more than any of these deals, and that a services engagement is the only way to reach it.

The engineer inside your walls

The way in is the forward-deployed engineer, embedded with the client to make the models work on real tasks. The role can sit on either side of the table. Deloitte advertises for an “Anthropic Forward Deployed Engineer”, while OpenAI embeds its own.

When the engineer works for the lab, the channel is not even hidden. OpenAI’s own post on X describes it: its forward-deployed engineers pair with your team, and “a tight feedback loop to OpenAI Research” means that “as you deploy agents, we learn what needs to improve around the model.” The lab presents that as a promise to improve the product. After eleven years of watching operating knowledge move between companies, I read the same loop as carrying what the post does not mention, which is how your business actually works.

What a platform learns, it can sell

The pattern predates AI. A Wall Street Journal investigation reported that Amazon staff used non-public data from independent sellers to develop competing house-brand products, despite a policy that said they would not. The labs are already doing the AI version in the open. Anthropic’s Claude for Financial Services, launched in July 2025, sells into the analyst workflows that banks’ own data vendors have long served. Policy is what a company says it will not do. A contract is what it agreed to. That distinction is the whole game.

The reasonable case for signing

The strongest case against all of this is also the most reasonable. A lab partnership hands you reasoning you could not build in a decade, and the labs have no obvious reason to compete with the customers who pay them. Not adopting is not a real option, and in my experience the firms that wait lose ground they do not get back.

Half of that is right. You do get capability you could not build alone, and standing still carries its own cost. On the other half, weigh what the labs do against what they say. OpenAI was selling $10 million consulting engagements in mid-2025, as reported by The Information, and by May 2026 it had anchored a $4 billion consulting venture. The habit of absorbing an adjacent business is already on the record. Jasper raised $125 million at a $1.5 billion valuation in October 2022 and was cutting staff by July 2023, once ChatGPT arrived weeks after its round. Sam Altman has said on the record that OpenAI will “steamroll” whatever is built in the model’s path, “not because we don’t like you, but because we have a mission.” The competition needs no bad intent, and that is exactly why you cannot manage it with trust. You manage it with structure.

Four questions before the signature

So the useful question is not whether to partner. It is how to partner without handing over the edge. Before a firm signs with a lab or its services arm, four questions decide whether its data and its accumulated judgment stay its own property. A private equity partner can run them on a portfolio company or an acquisition target, an advisory or Big Four partner on the firm’s own build or a client’s, an owner on their own business. A vague answer to any one of them is itself the finding.

  1. What never leaves the perimeter. Name every category of data the system sends to the lab, and where it is processed and stored. Sensitive material that stays inside your own environment is a pass. Where data physically sits is not a detail. Microsoft found it necessary to commit in writing to European governments that it would contest in court any order to switch off their cloud.
  2. What the embedded engineer can see. For any forward-deployed engineer, write down which workflows, data and decisions they can reach, and what reports back to the lab. A documented boundary is a pass. OpenAI’s own account of the feedback loop is reason enough to assume it exists and to bound it.
  3. Training and derived products. Put in the contract whether your data and interactions train the lab’s models, and who owns the workflows and tools built during the engagement. Written terms are a pass. A link to a policy page is not, as Amazon’s sellers learned.
  4. The model-agnostic fallback. Name what happens to the rebuilt process if the model is repriced or withdrawn, and whether a second model can be swapped in. A costed, model-agnostic path is a pass. Price and access sit well outside your building. Nvidia now pays the US government 15% of its China chip revenue for export licences, and a price set this year can be reset next.

Four specific answers mean a firm has found where its edge could leak and closed the main routes. Four vague ones mean it is already moving onto someone else’s platform. This is a check on ownership rather than a full security review, and the harder terms, the audit rights and the exit assistance, still have to be drafted. It tells you only whether the edge is yours to keep before you sign it away.

Come back to the week both labs put out their own shingle. For two years the consultancies taught their clients to trust the model, and the labs were taking notes. Adopting is not the mistake, and the firms that hang back will lose to the ones that move. The mistake is adopting in a shape that leaves your method sitting on a platform its owner can turn into a product. Run the four questions before the signature, so that what you were paid to build is still yours the day your partner decides to compete.

What that competition does to the consulting pyramid itself, the wide base of juniors under a thin layer of partners, is the next question, taken up in Will AI replace consultants, or just the consulting pyramid?

If you are adopting AI and mean to keep your expertise as your own asset, see how we build inside client environments.

The short version

  • Within one week in May 2026, both frontier labs opened their own consulting arms: Anthropic with Blackstone, Hellman & Friedman and Goldman Sachs on May 4, and OpenAI’s Deployment Company, which raised over $4 billion, on May 11.
  • A lab partnership can turn a firm’s proprietary data and expertise into the lab’s own product. The scarce input for frontier models is now fresh expert knowledge: Meta paid $14.3 billion for 49% of Scale AI in June 2025, and Mercor reached a $10 billion valuation in October 2025 paying professionals to produce it.
  • The channel is the embedded engineer. OpenAI’s own post describes a “tight feedback loop to OpenAI Research” from its forward-deployed engineers, and Deloitte itself advertises for an “Anthropic Forward Deployed Engineer.”
  • What a platform learns it can turn into a product. Amazon was reported in 2020 to have used third-party seller data to build competing products, and Anthropic’s Claude for Financial Services, launched July 2025, sells into workflows that banks’ incumbent data vendors long served.
  • A firm keeps its edge as its own asset by structure rather than trust: data that never leaves its own environment, a documented boundary around any embedded lab engineer, contract terms on training and derived products, and a costed model-agnostic fallback against repricing or cut-off.

Sources

  1. Blackstone, “Anthropic Partners with Blackstone, Hellman & Friedman and Goldman Sachs to Launch Enterprise AI Services Firm,” May 4, 2026. https://www.blackstone.com/news/press/anthropic-partners-with-blackstone-hellman-friedman-and-goldman-sachs-to-launch-enterprise-ai-services-firm/
  2. CIO Dive, “OpenAI’s Deployment Company raises over $4B for AI consulting and integration,” May 11, 2026. https://www.ciodive.com/news/openai-deployment-company-4-billion-ai-consulting-integration/819942/
  3. TechCrunch, “Anthropic and OpenAI are both launching joint ventures for enterprise AI services,” May 4, 2026. https://techcrunch.com/2026/05/04/anthropic-and-openai-are-both-launching-joint-ventures-for-enterprise-ai-services/
  4. CNBC, “PwC to become OpenAI’s first reseller and largest enterprise user,” May 29, 2024. https://www.cnbc.com/2024/05/29/pwc-to-become-openais-first-reseller-and-largest-enterprise-user.html
  5. CNBC, “Anthropic lands its biggest enterprise deployment ever with Deloitte deal,” October 6, 2025. https://www.cnbc.com/2025/10/06/anthropic-deloitte-enterprise-ai.html
  6. Forbes, “Meta invests $14 billion in Scale AI to strengthen model training,” June 23, 2025. https://www.forbes.com/sites/janakirammsv/2025/06/23/meta-invests-14-billion-in-scale-ai-to-strengthen-model-training/
  7. CNBC, “AI startup Mercor now valued at $10 billion with new $350 million funding round,” October 27, 2025. https://www.cnbc.com/2025/10/27/ai-hiring-startup-mercor-funding.html
  8. Bloomberg (via X), Surge AI fundraising talks, July 2025. https://x.com/business/status/1950642743723634828
  9. Deloitte careers, “Anthropic Forward Deployed Engineer (GPS),” accessed July 23, 2026. https://apply.deloitte.com/en_US/careers/JobDetail/Anthropic-Forward-Deployed-Engineer-GPS/350534
  10. OpenAI on X, post on forward-deployed engineers and the research feedback loop, accessed July 23, 2026. https://x.com/OpenAI/status/2019413717754991091
  11. CNBC, “WSJ: Amazon uses data from third-party sellers to develop its own products,” April 23, 2020. https://www.cnbc.com/2020/04/23/wsj-amazon-uses-data-from-third-party-sellers-to-develop-its-own-products.html
  12. Anthropic, “Claude for Financial Services,” July 15, 2025. https://www.anthropic.com/news/claude-for-financial-services
  13. Forbes, “OpenAI’s $10M AI consulting business: deployment takes center stage,” July 16, 2025. https://www.forbes.com/sites/solrashidi/2025/07/16/openais-10m-ai-consulting-business-deployment-takes-center-stage/
  14. Voicebot.ai, “Jasper AI laying off staff 9 months after $125M raise,” July 17, 2023. https://voicebot.ai/2023/07/17/jasper-ai-laying-off-staff-9-months-after-125m-raise/
  15. The Decoder, “Sam Altman explains why OpenAI might steamroll your AI startup,” 2024. https://the-decoder.com/sam-altman-explains-why-openai-might-steamroll-your-ai-startup/
  16. Microsoft, “European Digital Commitments,” April 30, 2025. https://blogs.microsoft.com/on-the-issues/2025/04/30/european-digital-commitments/
  17. CNN, “Nvidia and AMD will give US 15% of China sales,” August 11, 2025. https://www.cnn.com/2025/08/11/china/us-china-trade-nvidia-chips-intl-hnk