OpenAI's The Deployment Company and Anthropic's enterprise services joint venture with Blackstone / H&F / Goldman Sachs were announced in the same week—behind these two strategic declarations lies the same judgment: the peak of the pure model API business model may have passed, and the next wave of growth lies in the enterprise implementation layer, which will directly squeeze the middleware integrators and traditional consultancies that help enterprises connect APIs.
From selling models to selling services—APIs are the infrastructure layer; enterprise implementation is the differentiated profit layer. Both companies are simultaneously bypassing the middle layer to build delivery capabilities directly.
Middleware integrators and consultancies bear the brunt—when OpenAI and Anthropic directly provide implementation services, the "helping clients connect APIs" business faces head-on competition from the top vendors.
Procurement decision power is shifting from IT to the CFO—Uber burned through its full-year Claude Code budget in four months, showing that "metered billing" is creating a new type of budget overrun. Enterprises need measurable ROI, not the most advanced model.
"Verifiable AI" becomes the entry barrier for highly regulated industries—Kepler uses Claude to trace every computation result back to SEC source files. The core design principle is "the model cannot be the entire system."
v2 update (2026-09)—OpenAI Deployment Company officially launched on May 11 and acquired consultancy Tomoro (the first-published version of this article had incorrect funding/valuation figures, now corrected to $4B/$10B); the Anthropic side was named Ode with Anthropic and secured its first named client; Uber has set monthly spending caps on agentic coding tools; IDC predicts CFOs will dominate 65% of enterprise AI spending decisions by 2027—the §03 procurement power shift judgment is further validated. See the new section at the end for details.
The outside world often interprets these two announcements as "AI companies also want to make consulting money." But the more accurate reading is: model capabilities have become commoditized to the point where they can no longer form a moat on their own; the real competition has shifted to "who can deliver results into enterprise workflows"—which means traditional software and consulting companies face not a partnership opportunity, but a structural threat.
OpenAI forming "The Deployment Company" is not an ordinary product launch—it is a systemic reshaping of OpenAI's business model: raising over $4 billion at a $10 billion valuation (Update 2026-09: the first-published version of this article mistakenly stated these figures as $400B/$1000B, now corrected per OpenAI's official announcement), focused on helping enterprises deploy OpenAI tools. This positioning is far more similar to the logic of McKinsey and Accenture than to AWS.
Almost simultaneously, Anthropic announced a partnership with Blackstone, Hellman & Friedman, and Goldman Sachs to create an enterprise-grade AI services company, packaging Claude's model capabilities into managed services for large enterprises.
The strategic logic of the two announcements overlaps almost completely: from selling models to selling services (APIs are the infrastructure layer; enterprise implementation is the differentiated profit layer), from technology company to solutions provider (large enterprises don't buy compute—they buy results), bringing in traditional capital (Blackstone and Goldman Sachs bring not just funding, but relationship networks that open Fortune 500 procurement decision chains).
a16z's formulation cuts to the structural heart of this trend: in the AI era, only two survival paths remain for software companies—become deeply vertical AI-native tools, or become infrastructure for general AI agents; the middle ground is disappearing. OpenAI and Anthropic's choice is to explicitly side with "infrastructure + vertical services," bypassing the middle layer entirely.
The most direct casualties of this shift are the middle layer: the integrators and AI consultancies that make a living by "helping enterprises integrate OpenAI/Anthropic APIs into their business systems." Another a16z analysis describes the technical foundation of this logic: as AI capabilities permeate, the abstraction layers encapsulated by traditional SaaS products are being penetrated layer by layer—when AI can pierce abstraction layers to directly manipulate underlying data, the value of "I'll encapsulate it for you" disappears.
This has direct implications on the enterprise side as well: small and mid-sized AI integrators—when OpenAI and Anthropic directly provide implementation services, the "helping clients connect APIs" business faces direct competition; the survival space of traditional enterprise software—the argument of "Workday's Last Workday?" is reinforced here, as AI agents interface directly with underlying databases, bypassing the middle-layer SaaS logic of HR/ERP; large consulting firms—McKinsey and Accenture are already feeling the pressure. OpenAI's "The Deployment Company" is an explicit declaration of war.
Data from actual research (a16z: The Real State of Enterprise AI Adoption) paints a picture far more conservative than vendor narratives: customer service automation, code assistance, and content generation are the top three scenarios by adoption rate, while highly anticipated "AI agent autonomous decision-making" applications still have low adoption rates—trust, compliance, and process integration are the three real-world barriers.
This cross-validates with reporting from The Information: enterprise AI ROI is moving from "does it work?" to "is it worth the money?"—technology procurement decision power is shifting from IT departments to the CFO level. Uber burning through its full-year Claude Code budget in four months is a microcosm of this shift: the "metered billing" model of AI tools is creating a new type of budget overrun. Enterprises need to establish clear mapping mechanisms between AI usage volume and business output, rather than treating AI as an open-ended R&D experiment. Update (2026-09): Uber has codified this judgment into specific rules; see the new section at the end.
This is precisely the opportunity for OpenAI's "The Deployment Company" and Anthropic's enterprise services company: when enterprise AI needs to convince the CFO, "measurable business results" matter more than "the most advanced model."
Large enterprises don't buy compute—they buy results.
The model cannot be the entire system—it must be verifiable. — Kepler Financial AI Architecture Design Principle
Kepler's case provides a concrete implementation path for success in a highly regulated industry: embedding Claude into the financial research pipeline, making every computation result traceable to SEC source files, with the core design principle that "the model cannot be the entire system."
This architectural paradigm—AI proposes conclusions, original data is verifiable—precisely solves the problem that worries CFOs and compliance departments most: the results produced by AI black boxes cannot be audited. "Verifiable AI" is the passport for AI to enter highly regulated industries such as finance, law, and healthcare—not merely a capability threshold.
For AI decision-makers: the decision unit for AI procurement is shifting from "which model is best" to "which service provider can deliver measurable business results"; for AI startups: the true moat lies in unique industry data + deep business process integration, not in model capabilities themselves; for traditional software and consulting companies: AI model companies extending into enterprise services is a structural threat, not a partnership opportunity—the window of speed is closing.
OpenAI Deployment Company and Anthropic's enterprise services company will directly compete for large enterprise clients; meanwhile, the "verifiable AI" architecture (model + auditable original data) will become the entry barrier for highly regulated industries. Before OpenAI Deployment Company fully establishes its distribution system, building vertical moats is an urgent priority.
On the OpenAI side, the "declaration" has materialized into an entity: on May 11, 2026, OpenAI officially announced the formation of the Deployment Company, led by TPG with Advent, Bain Capital, and Brookfield as co-leads, joined by 19 investment institutions and consultancies/system integrators including McKinsey, Bain, and Capgemini. The first-round funding exceeded $4 billion, valuing the company at $10 billion, with OpenAI retaining majority control. On the same day, OpenAI announced the acquisition of AI consultancy Tomoro, directly integrating its approximately 150 "Forward Deployed Engineers" into the Deployment Company—its most direct move to align with the Palantir-style delivery model.
On the Anthropic side, the consortium officially named its enterprise services company on July 15, 2026—Ode with Anthropic, led by Blackstone and Hellman & Friedman, with new investors including General Atlantic, Leonard Green, Apollo Global Management, GIC, and Sequoia Capital. TechCrunch reported its valuation at $1.5 billion, starting with approximately 100 engineers. In late July, Ode announced its first named client partnership—a collaboration with compliance technology company LogicGate, embedding Claude-driven agentic capabilities. Both companies have moved from "announcement" to "launch + first deal," validating this article's initial judgment: this is not a PR posture, but real organizational and capital commitment.
The Uber case mentioned in §03 also subsequently validated the "procurement power shift" judgment: in June 2026, TechCrunch reported that Uber set a $1,500 monthly per-employee spending cap on agentic coding tools (including Claude Code and Cursor). Uber President and COO Andrew Macdonald, in an interview with Fortune, admitted he found it difficult to draw a direct causal line between increased Claude Code usage and consumer-side product innovation, even implicitly struggling to prove it delivered "25% more useful consumer features."
This is not an isolated case. IDC predicts that by 2027, CFOs will dominate 65% of enterprise AI spending decisions; an August 2026 Forbes Finance Council column directly characterizes this phenomenon as a structural transformation of the CFO role—they must now understand a new unit of enterprise economics: the token. This article's §03 judgment about procurement decision power shifting from IT to the CFO has been validated by more systematic data four months later.