Beyond the Model: Why Anthropic and OpenAI Are Betting the Next Trillion Is in Implementation

The most revealing AI story of mid-July 2026 was not a new model. It was two frontier labs, independently, deciding that better models are not enough to win the enterprise. Anthropic launched Ode with Anthropic, a roughly 1.5 billion dollar implementation venture backed by Blackstone, Hellman and Friedman and Goldman Sachs. OpenAI has stood up its own equivalent, The Deployment Company. Both are built on the same wager: that the next enormous pool of AI value sits not in the model layer but in the messy, human work of wiring models into how real businesses actually operate.

For an industry that has spent three years treating model quality as the whole game, this is a notable turn. This piece explains what these ventures are, why the labs are building them, and what the move implies about where AI money is really made.

What these ventures actually are

Both are services businesses that put engineers inside customer companies. Ode with Anthropic was created as a joint venture with major financial backers, employs around 100 engineers, and grew out of Anthropic's acquisition of a services startup, Fractional AI, whose co-founders Chris Taylor and Eddie Siegel now lead it. Its job is not to sell API access. It is to embed teams that build, integrate and operate AI inside enterprises that lack the in-house capability to do it themselves.

OpenAI's Deployment Company follows the same logic from the other side of the rivalry. The shared model is often called forward-deployed engineering, and it is not new: it is the playbook Palantir spent years refining, sending engineers to live alongside a customer's problem rather than shipping software and hoping the customer figures it out. What is new is that the two leading model labs have decided they need that muscle directly, rather than leaving it to the consulting incumbents such as Deloitte and Accenture that already field large forward-deployed teams.

Why the labs are doing this now

The move answers a problem the last two years exposed. Enterprises have not struggled with AI because the models are weak. They have struggled because nobody inside the building can connect a powerful general model to decades of accumulated, idiosyncratic workflow, data and process. The gap between a capable model and a captured business outcome is an integration gap, and it is wide.

Three pressures make owning that gap attractive right now. First, differentiation is compressing at the model layer. As several labs cluster near the frontier and capable open-weight models improve, the raw model is a smaller share of the value and a harder place to sustain a premium. Second, services revenue is durable and expandable in a way API resale is not: once your engineers are embedded in a customer's operations, the relationship deepens and the switching cost rises. Third, it converts a support cost into a profit centre. Labs were already doing heavy hand-holding to land large accounts; formalising it as a venture turns that necessary work into a business, and pulls in outside capital to fund the scale-up.

Reading the strategy without cheerleading or cynicism

There are two tidy narratives on offer, and both are too clean. The optimistic one says the labs are simply meeting customers where they are, and that embedded expertise is the honest way to make AI deliver. The cynical one says the model business is commoditising, so the labs are retreating into lower-margin consulting because the software moat is thinner than advertised.

The more accurate reading holds elements of both. It is genuinely true that implementation is where a lot of enterprise value is unlocked, and that customers benefit from expert help bridging the integration gap. It is also true that a rush into services is exactly what you would expect if model-layer differentiation were narrowing, because services are where you go when the product alone no longer commands the premium. Both can be true at once, and the interesting tension is that the same move signals confidence about total addressable value and caution about model-layer defensibility in a single decision.

There are risks worth naming too. Services businesses scale with people, not code, so they carry lower margins and heavier operational drag than software. And a lab that both sells models and deploys them for customers sits close to a conflict: the embedded team has an obvious incentive to standardise every engagement on its parent's model, whether or not that is the best fit for the task in front of it.

What this means for how teams buy AI

For companies on the receiving end, the ventures are a useful signal even if they never hire one. The signal is that implementation, not model selection, is the hard part, and that budgeting and planning should reflect that. The lasting cost and value of enterprise AI lives in integration, change management and workflow design far more than in which flagship model sits underneath.

That reframing carries a quieter warning about lock-in. Accepting a single lab's forward-deployed team is efficient, but it tends to standardise an organisation on that lab's stack, which is comfortable until the frontier shifts, a price changes, or a different model turns out to be better for a given job. The teams that keep the most leverage are the ones that treat the underlying model as a component they can swap, and keep the integration layer, the connections to their data and workflows, independent of any one provider.

This is the unglamorous case for model-agnostic infrastructure. A platform such as Metir AI sits at the integration layer rather than the model layer, bringing the leading models from multiple labs into one workspace so a team can wire AI into its work once and then route each task to whichever model fits, without re-plumbing every time the frontier or the pricing moves. Owning that layer is how a company captures the implementation value the labs are chasing, while keeping the freedom the labs have a natural incentive to narrow.

The bigger picture

The launch of two lab-owned implementation ventures in the same week marks a maturing of the AI market. The story is moving on from whose model is smartest to whose model actually gets deployed and used, and that is a different, more operational contest. It rewards integration, services and the boring work of fitting a general tool to a specific business, and it explains why the labs are willing to build lower-margin services arms to own it.

For everyone else, the takeaway is not which venture to hire. It is that the value has shifted toward implementation, that implementation creates real gravity toward a single vendor, and that the durable advantage belongs to teams who capture the integration layer themselves while keeping the model underneath replaceable. The best models will keep changing. The workflows they plug into are what compound.

Own the integration layer, keep the model replaceable

The labs are right that value lives in implementation. The teams that keep the leverage are the ones that wire AI into their work once and stay free to swap the model underneath. Metir AI brings the leading models into one workspace at the integration layer, so you build on your workflows and data, not on a single provider's stack. Try Metir AI free and keep the freedom the value is worth.

Sources:

• Anthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just models | TechCrunch
• Anthropic raises $65B Series H at $965B valuation | Anthropic
• Anthropic tops OpenAI as most valuable AI startup | CNBC
• Microsoft's Frontier push aims to turn AI spending into measurable returns | Fortune