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 bulletins lies the same judgment: the peak of the pure model API business model may have passed, and the next wave of growth lies in enterprise implementation, which will directly squeeze the middle-tier integrators who help enterprises connect APIs and traditional consulting firms.
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 tier to build delivery capabilities in-house.
Middle-tier integrators and consulting firms are hit first—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 burning through its annual Claude Code budget in four months shows that "metered billing" is creating a new type of budget overruns; 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 the original SEC filing; the core design principle is "the model cannot become the entire system."
The outside world often interprets these two announcements as "AI companies also want to make consulting money." But a more accurate reading is: model capabilities have become commoditized to the point where they can no longer constitute a moat on their own; the real competition has shifted to "who can deliver results into enterprise workflows"—which means traditional software and consulting firms are facing not a collaboration opportunity, but a structural threat.
OpenAI establishing "The Deployment Company" is not an ordinary product launch—it is a systemic reshaping of OpenAI's business model: raising over $40 billion at a $100 billion valuation, focused on helping enterprises deploy OpenAI tools. This positioning is highly similar to the logic of McKinsey and Accenture, not 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 almost completely overlaps: from selling models to selling services (APIs are the infrastructure layer; enterprise implementation is the differentiated profit layer), from technology company to solution provider (large enterprises don't buy compute power; they buy results), bringing in traditional capital (Blackstone and Goldman Sachs bring not just funding, but relationship networks that open doors to Fortune 500 procurement decision chains).
a16z's framing cuts to the structure of this trend: in the AI era, there are only two survival paths left for software companies—become deeply vertical AI-native tools, or become the infrastructure for general AI agents; the middle ground is vanishing. OpenAI and Anthropic's choice is a clear alignment with "infrastructure + vertical services," directly bypassing the middle tier.
The most direct victims of this pivot are the middle tier: the integrators and AI consulting firms that make a living by "helping enterprises integrate OpenAI/Anthropic APIs into their business systems." Another a16z analysis describes the technical underpinning 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 through abstraction layers to directly manipulate underlying data, the value of "I'll encapsulate it for you" disappears.
This has direct implications for 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 in "Workday's Last Workday?" is reinforced here, as AI agents directly interface with underlying databases, bypassing the middle-tier SaaS logic of HR/ERP; large consulting firms—McKinsey and Accenture are already feeling the pressure; OpenAI's "The Deployment Company" is a clear declaration of war.
Data from actual research (a16z: The Real State of Enterprise AI Adoption) paints a far more conservative picture than vendor narratives: customer service automation, coding 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 triple real-world barriers.
This cross-validates with reporting from The Information: enterprise AI ROI is moving from "can 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 annual 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 overruns; enterprises need to establish clear mapping mechanisms between AI usage volume and business output, rather than treating AI as an unlimited R&D experiment.
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 power; they buy results.
The model cannot become the entire system—it must be verifiable. — Kepler Financial AI Architecture Design Principle
Kepler's case provides a concrete path for successful implementation in a highly regulated industry: embedding Claude into the financial research pipeline, making every computation result traceable to the original SEC filing, with the core design principle that "the model cannot become the entire system."
This architectural paradigm—AI proposes conclusions, original data is verifiable—precisely solves the problem that worries CFOs and compliance departments the most: the results produced by the AI black box cannot be audited. "Verifiable AI" is the entry ticket for AI into highly regulated industries such as finance, law, and healthcare, not merely a capability threshold.
For AI decision-makers: the AI procurement decision unit is shifting from "which model is best" to "which service provider can deliver measurable business results"; for AI startups: the real moat lies in unique industry data + deep business process integration, not model capability itself; for traditional software and consulting firms: AI model companies extending into enterprise services is a structural threat, not a collaboration opportunity—the speed window is closing.
OpenAI Deployment Company and Anthropic's enterprise services company will directly compete for large enterprise clients; and 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 barriers is an urgent priority.
First published 2026-07-15