English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

OpenAI and Anthropic Turn 'Selling Intelligence' into a Business Transformation Play: May 4, 2026 Analysis

Forum topic · ✨步子哥 · 2026-05-06

Summary

On May 4, 2026, OpenAI and Anthropic announced consulting-style delivery businesses within hours of each other, signaling a strategic shift from selling model access to embedding AI directly into enterprise workflows. Bloomberg reported OpenAI was secretly raising $4 billion for an independent 'The Development Company' at a $10 billion valuation, while Anthropic officially announced a $1.5 billion joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs focused on enterprise AI services. The post argues that API revenue—despite Anthropic's ARR tripling to over $30 billion in 2026—offers weak margins and fragile customer loyalty, since token prices keep falling and switching costs are near zero. By deploying elite engineers into client companies and rewriting their workflows for project-based fees, model labs aim to build the deep moat that traditional consultancies like Accenture and Deloitte cannot match, lacking access to internal model tooling. The piece also notes Chinese cloud vendors (Alibaba Cloud, Volcano Engine, Huawei Cloud, Baidu, Tencent) already run similar model-plus-implementation plays, and traces AI's evolution from 2023 research labs to 2026 system integrators.

On May 4, 2026, the two biggest names in AI—OpenAI and Anthropic—announced consulting-delivery-style companies within hours of each other. The author's takeaway: model companies have finally realized the real money isn't in the cloud, but in every line of code and every process inside client companies. The dual announcement, seemingly coordinated, shattered the business of traditional consulting giants and taught a lesson to players still competing on parameters and context length: the AI moat was never how smart your model is, but whether you can embed it into the heart of a client's company and completely rewrite their workflows.

🌟 May 4: Bloomberg leak in the morning, official announcements by afternoon

In the morning (US Eastern time), Bloomberg broke the news: OpenAI was secretly raising funds to establish an independent company, "The Development Company," valued at $10 billion and planning to raise $4 billion, with 19 investors already committed most of the capital. A few hours later, Anthropic officially announced a $1.5 billion joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs, focused on enterprise-grade AI services.

The two events were only hours apart, yet felt pre-orchestrated. After two years of arguing over whose model had more parameters, longer context, or cheaper training, what emerged that day was not a new model but the same corporate form—embedding engineers directly into client companies, redesigning business processes, and charging per project delivery.

This is exactly the business traditional consultancies—Accenture, Deloitte, IBM Consulting—have run for thirty years: the "person-month model." The only difference: this time the players hold the most cutting-edge models and internal toolchains.

🔧 The new companies sell neither API nor SaaS, but "embedding AI into your company's DNA"

Many people's first reaction: is API selling failing? Not at all, by the numbers. Anthropic itself disclosed that 2026 ARR had surpassed $30 billion, up from $9 billion at the end of 2025—more than 3x growth, still accelerating. But high growth doesn't mean a good business, let alone a good moat.

The new venture's playbook is clear:

  • Not selling tokens (the usage-based channel continues)
  • Not selling SaaS subscriptions (what companies like Sierra do)
  • But dispatching a small number of top engineers to embed within client companies, helping them integrate AI into core business processes, charging by project milestones
  • What does this resemble? You bought a supercar (the model); the API just sells you fuel (tokens). The new company provides a professional pit crew, tunes the chassis, teaches you to drift, and drives the car onto the F1 track—the client can never leave, because under the hood everything is now your code and processes.

    The author has seen many enterprises that bought Claude, GPT, and Gemini APIs and stacked up prompts, only to find the real difficulty is making AI part of daily work. Security compliance, data privacy, legacy system integration, employee training—these are the real pitfalls. The new companies exist to fill those pits in one go.

    📉 Strong API growth, but three hidden problems make the moat flimsy

    Why are model companies doing the "heavy lifting" themselves? Because APIs, despite looking good, have three fatal weaknesses:

    1. Persistently declining gross margins. From 2024 to 2026, token prices halved again and again, with Sonnet 4.6 and GPT-5.4 cutting prices in turn, like the smartphone market squeezed to the bottom. Revenue maintained by volume looks impressive, but margins get squeezed like toothpaste. 2. Alarmingly fragile customer stickiness. A client happy with Claude today can switch when Gemini cuts prices tomorrow—technically done in hours. Switching cost is as low as changing a food-delivery app. There's no lock-in spell. 3. Most critically: large enterprises never wanted "a pool of tokens." They want "a solution I can deploy immediately." Salesforce proved this by conquering the market with "implementation + subscription." The Fortune 500's biggest spenders never care how smart the model is—they care how much their KPIs improve.

    API is a good business, but not a moat. Once your engineers rewrite the client's workflow, the model becomes irreplaceable DNA. That's the real lock.

    🏛️ The good days of the Accentures are over—traditional consulting's AI anxiety fully erupts

    For two years the global consulting industry has suffered an awkward collective silence: every major client asks "how do we use AI," yet consultancies can't deliver workable solutions.

    They scramble to hire AI engineers, but two hard flaws remain:

    1. They can't get the models' "internal scripts." When delivering Claude-based projects, consultancies only get public APIs and documentation; Anthropic's own engineers get model cards, internal toolchains, even unreleased features. The former takes a year to deliver; the latter runs in three months. 2. Their labor-pricing model is being wrecked by AI. A senior consultant costs $2,000/day; a project runs 50 person-months. Anthropic saying "we'll embed a few engineers" flips the person-month model on its head.

    No wonder Anthropic partnered with Wall Street PE firms rather than other consultancies—it wants to do the work itself: PE provides the client pool and capital, Anthropic provides models and engineers, with no need for Accenture as the "middleman."

    It's like Tesla suddenly appearing in the taxi industry: you're still selling tickets while someone else builds the intelligent transportation system.

    🇨🇳 Chinese cloud vendors have long run the "same script," just under different names

    Many assume this is a Silicon Valley invention, but China has been ahead, just with more down-to-earth names:

  • Alibaba Cloud + Qwen: "AI large-model all-in-one machine + implementation services" sold to government and enterprise clients—hardware + model + on-site teams
  • Volcano Engine + Doubao: industry solutions + embedded engineers for gaming, e-commerce, and content
  • Huawei Cloud + Pangu: end-to-end integration in industrial and telecom sectors
  • Baidu Intelligent Cloud + ERNIE: Qianfan platform + industry consulting teams
  • Tencent Cloud + Hunyuan: ecosystem-partner SI collaboration
  • The essence is identical to what Anthropic/OpenAI just did: bundling model company + consulting delivery. The difference: in China it's an internal "three-in-one" within cloud vendors (model + sales + implementation), somewhat organizationally tangled; in Silicon Valley it's a cleaner alliance of "model company + Wall Street + implementation partners" with more professional division of labor.

    The biggest lesson for Chinese peers: a model company doing only "API + SaaS" isn't a deep enough business. The real moat lies in that step of "embedding engineers into client companies"—once you rewrite the client's workflow, the model can never be swapped out. But it's also the heaviest, least sexy step. Anthropic's choice tells everyone: the moat that really makes money is often exactly the move others consider too heavy.

    🌍 From 2023 research lab to 2026 system integrator—the trajectory of AI business is now clear

    On a longer timeline, the path is obvious:

  • 2023: Model companies were research labs, only publishing papers and demos—like Einstein's theoretical physics institute
  • 2024: They became API providers, charging per token—like power plants selling electricity
  • 2025: They became consumer product companies like ChatGPT and Claude.ai—like Apple selling phones
  • May 4, 2026: They fully became system integrators—like SAP and Oracle once sold ERP plus implementation services
Each step is a descent in business model: closer to the customer, larger contracts, deeper moats. Salesforce, SAP, and Oracle walked this road countless times. Now it's the AI companies' turn.

The real significance of May 4 is not "model companies admitting defeat." It's that they finally recognized: the winning move is not training smarter models, but using models to rewrite client workflows. The former is a tech company's obsession; the latter is the common sense of IT giants.

Imagine looking back from 2030: the companies still grinding on parameter leaderboards may have become "model-selling tool vendors," while the players who dared to embed engineers into client companies and rewrite workflows will have become the true "digital stewards" of the AI era.

A question for you:

Looking back from 2030, which will be the most profitable AI company: a model company, a next-gen SaaS company, or a next-gen consultancy? Which of the three do you bet on?

The author's answer is already hidden in the article. What's yours?

------ References 1. Bloomberg. (May 4, 2026). OpenAI to Launch Independent "The Development Company" with $4B Funding Round. 2. Anthropic Official Press Release. (May 4, 2026). Anthropic Forms $1.5 Billion Joint Venture with Blackstone, Hellman & Friedman and Goldman Sachs to Deliver Enterprise AI Services. 3. TechCrunch. (May 4, 2026). Anthropic and OpenAI are both launching joint ventures for enterprise AI services. 4. Fortune. (May 5, 2026). Why the biggest AI labs are becoming the new Accenture. 5. Blackstone Press Release. (May 4, 2026). Blackstone Partners with Anthropic to Accelerate Responsible AI Adoption Across Portfolio Companies.

Tags

#openai#anthropic#ai-consulting#enterprise-ai#business-model#accenture#system-integration#ai-moat

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177619522