OpenAI and Anthropic Pivot to Consulting: The Strategic Shift from Selling APIs to Enterprise Deploy-AI Topic

A Watershed Moment for AI Business Models: The Rise of the "deployment Consultancy"

May 2026 marks a pivotal turning point in the Artificial Intelligence industry. On a single day, the world's two most prominent AI model companies—OpenAI and anthropic—unveiled nearly identical strategic shifts.

The Timeline of Events (May 2026):

Morning (ET): Bloomberg reported that openai is rAIsing capital to establish an independent entity, "The Deployment Company," valued at $10 billion with $4 billion in funding.
Afternoon (ET): Anthropic officially announced a $1.5 billion Joint Venture with Wall Street giants Blackstone, HeLLMan & Friedman, and Goldman Sachs, dedicated exclusively to enterprise AI services.

Separated by just a few hours, these announcements signaled a departure from the nARRative of competing on parameters, context Windows, and training costs. Instead, both companies introduced the Same new corporate form: the AI Consultancy.

This raises a critical question: Why are model companies pivoting to consulting?

Two Joint Ventures, One Core strategy

On the surface, the structures APPear distinct. Anthropic has partnered with Wall Street capital (Blackstone, H&F, Goldman Sachs), while OpenAI has assembled a coalition of 19 investors. However, at their core, they are executing the exACT same play.

The New Service Model:

Not just selling API tokens: The traditional pay-per-call volume business continues through existing channels.
Not just selling SaaS subscriptions: The per-seat monthly revenue model is a game for companies like Sierra.
embedding Engineers: The new strategy involves embedding engineers directly into CLIent organizations to redesign workflows, integrate AI into core processes, and charge for project-based delivery.

This model has a familiar name: Consulting & Delivery. It is the same business model that Accenture, Deloitte, and IBM Consulting have operated for decades. The key difference lies in client acquisition: Anthropic’s venture explicitly targets the portfolio companies of its PE backers. With Blackstone alone holding over 250 large enterprises in its portfolio, the joint venture starts with a locked-in customer base.

Has the API Market Peaked?

One might assume API growth has plateaued. The numbers suggest otherwise. Anthropic disclosed that its Annual Recurring Revenue (ARR) surpassed $30 billion in early 2026 (up from roughly $9 billion at the end of 2025)—a threefold increase showing no signs of slowing.

However, high ARR growth does not necessarily equate to a "good" business. There are three hidden challenges:

Declining Margins: OpenAI and Anthropic are engaged in a price war. With token prices dropping over 80% since 2024, revenue relies heavily on volume.
FrAGIle Customer loyalty: Switching costs are virtually non-existent. An enterprise can migrate from claude to GPT to Gemini in hours; loyalty is limited to the cost of stitching Prompts together.
Enterprise Needs Solutions, Not Just APIs: Fortune 500 giants do not just want a "pool of tokens"; they demand "usable solutions." This is a lesson Salesforce proved repeatedly during the SaaS era.

While APIs are profitable, they do not build a strong moat. Consulting delivery does—once your engineers have rewritten a client's workflow, the client becomes deeply entrenched and unlikely to leave.

The End of the golden age for Traditional Consulting?

Over the past two years, the global consulting industry has faced an awkward reality: major clients are asking "How do we use AI?", yet consultancies lack the intrinsic AI Implementation capabilities to answer effectively.

Firms like Accenture and Deloitte are scrambling to hire AI engineers, but they face two critical disadvantages compared to model companies:

Lack of Model Ownership: When a consultancy delivers a project using Claude, they rely on public APIs and documentation. Anthropic’s embedded engineers, however, have access to internal model cards, proprietary toolchains, and unreleased features. A project that takes a consultancy a year might take Anthropic’s team just three months.
Obsolete Pricing Models: Traditional firms bill based on "man-months" (e.g., a senior consultant at $2,000/day). The AI era requires far fewer hands on deck. Anthropic’s model of "embedding a small team of engineers" effectively dismantles the traditional labor-heavy billing structure.

This explains why Anthropic partnered with financial institutions rather than other consultancies. They are building the consultancy themselves, with PE firms providing the clients and capital, and Anthropic providing the models and engineers—leaving no room for intermediaries like Accenture.

The Parallel Story in China

Chinese cloud and AI providers have long been walking this path, though under different nomenclature than "consulting joint ventures."

Equivalent Models Already in Motion:

Alibaba Cloud + Tongyi: Government/Enterprise "AI Appliance + Implementation Services."
Volcano Engine + Doubao: Industry solutions + on-site resident engineers.
Huawei Cloud + Pangu: Industry large models + end-to-end integration.
Baidu AI Cloud + Wenxin: Qianfan platform + industry consulting teams.
Tencent Cloud + Hunyuan: Industry models + ecosystem partner (SI) collaboration.

The common thread is packaging "Model Company + Consulting Delivery" for government and enterprise clients. The difference lies in the division of labor: in China, cloud vendors often act as the model provider, sales channel, and implementer simultaneously. In Silicon Valley, it is a clearer alliance between Model Companies, PE/Wall Street, and implementation partners.

Key Takeaway for the Industry:
Selling APIs and SaaS alone is not a "deep" enough business for model companies. The true moat lies in embedding engineers into client companies. Rewriting a client's workflow makes the model irreplaceable. While this is a "heavy" and less "sexy" Operation compared to scalable SaaS narratives, Anthropic’s move confirms that the most profitable moats are often built through the "heavy" lifting that others avoid.

What This Day Truly Signifies

Viewed on a longer timeline, the evolution of model companies is clear:

2023: Research labs publishing papers and dEMOs.
2024: API providers monetizing via tokens.
2025: Consumer product companies (ChatGPT/Claude.ai).
May 2026: System Integrators.

Each step represents a commercial下沉 (sinking deeper into the market), getting closer to the client, securing larger contracts, and building deeper moats. This is the exact path trodden by IT giants like Salesforce, SAP, and Oracle—evolving from "selling software" to "selling implementation."

The true significance of May 4th is not that model companies are "giving up." It is their realization that the winning battle is not about "training smarter models," but about "rewriting client workflows with models." The former is a tech company's obsession; the latter is an IT giant's common sense.

A Question for the Future:
Looking back from 2030, which category will be the most profitable: Model Companies, Next-Gen SaaS Companies, or Next-Gen Consulting Firms? Which one are you betting on?

★★★★★

Be the first to rate this article.

OpenAI and Anthropic Pivot to Consulting: The Strategic Shift from Selling APIs to Enterprise Deploy