The frontier labs have solved capability faster than the world has learned to use it, and this post answers where the money actually sits for an entrepreneur who is never going to train a model.
The Biggest AI money this Month Was not Spent on Models. It Was Spent on Getting Models to Land.
Somebody asks me a version of this question almost every week. They read the AI headlines, see the billions moving around, watch another model release, and quietly conclude that the opportunity has already been claimed by companies with data centers and research teams. Then they ask what is left for a normal business owner.
Here is the direct answer. The money is in deployment, not in models. You do not make money from AI by building AI. You make money by closing the distance between what these systems can already do and what an ordinary business has actually implemented. That distance is enormous, it is measurable, and right now it is the most valuable real estate in the industry.
You do not have to take my word for how big that gap is. Look at where the largest companies in the world are putting their money this month. On July 2, Microsoft announced Microsoft Frontier Company, a new operating business backed by a 2.5 billion dollar commitment and 6,000 industry and engineering experts, according to TechCrunch. Its job is not to build a better model. Its job is to make Microsoft’s existing AI tools actually work inside customer organizations.
That is the whole thesis in one announcement. The most sophisticated AI company on earth looked at its own product line and concluded that the bottleneck was no longer capability. The bottleneck was implementation. So it spent 2.5 billion dollars on implementation.
The same gap exists in every small and midsized business in your city. Nobody is spending 2.5 billion dollars to close it for them. That is the opening.
Key Takeaways
- The constraint in AI has shifted from what the models can do to whether a business can put them into production, and the biggest AI investments announced this month were aimed at deployment, not capability.
- Gartner forecasts 585.5 billion dollars in AI services spending for 2026 against 32.6 billion dollars in AI model spending, which tells you where the commercial weight of this market actually sits.
- MIT’s Project NANDA found that roughly 95 percent of enterprise generative AI pilots deliver no measurable impact on profit and loss, which means the failure point is almost always implementation rather than technology.
- The frontier labs are solving this problem for the Fortune 500 through forward deployed engineers and global integrators, which leaves everything below the enterprise tier largely unserved.
- Building a deployment services practice requires you to understand workflows and change management, not to train anything.
Capability Outran Adoption, and Nobody Closed the Gap
I want to describe the situation plainly, because the hype in both directions makes it hard to see.
The models are good. They have been good for a while. The average business owner reading this has access, for a few hundred dollars a month, to systems that can read a contract, draft a proposal, answer a customer, reconcile a spreadsheet, and route a support ticket. That capability is real and it is sitting there, largely untouched, in millions of businesses.
Meanwhile, adoption looks nothing like capability. McKinsey’s State of AI research found that 88 percent of organizations now regularly use AI in at least one business function, but nearly two-thirds have not begun scaling it across the enterprise, and a similar share report no significant effect on the bottom line. Read that carefully. Almost everyone has touched AI. Almost nobody has operationalized it.
The sharpest version of this comes from MIT. Project NANDA’s report “The GenAI Divide: State of AI in Business 2025” found that roughly 95 percent of enterprise generative AI pilots produced zero measurable return. The researchers were specific about why. It was not model weakness. It was brittle workflows, missing context, and misalignment with the way people actually do their jobs.
Ravi Kumar S, the CEO of Cognizant, said it about as directly as anyone has in the July 27 announcement of his company’s expanded partnership with Anthropic: “AI capability is rising faster than enterprises can absorb it, and that gap is the defining problem of this moment.”
That sentence is worth sitting with, because it is not a complaint. It is a market description.
Here is what it means for you. Every business that bought an AI subscription and got nothing out of it is not a business that decided AI does not work. It is a business that has an unsolved implementation problem and a budget line it cannot justify. That is not a hard sale. That is a business quietly waiting for someone to show up who knows what to do next.
And the people who know what to do next are, at this moment, in extremely short supply relative to the number of businesses who need them.
Follow Where the Money Went this Month
Let me put the receipts on the table, because this is not a trend I am inferring. It is a trend the largest AI companies in the world announced out loud in the last four weeks.
Microsoft. On July 2, 2026, Microsoft launched Microsoft Frontier Company with a 2.5 billion dollar commitment and 6,000 industry and engineering experts, per TechCrunch. Commercial Business CEO Judson Althoff described it as “the largest, most capable, outcome-driven engineering organization in the industry.” Early named partners include the London Stock Exchange Group, Unilever, and Land O’Lakes. Not a model. A deployment organization.
Amazon. Two days earlier, on June 30, AWS announced its own one billion dollar internal commitment to a forward deployed engineering organization, also reported by TechCrunch. Same shape, same purpose.
Anthropic. On July 27, Cognizant became one of a small number of Global Premier Partners in the Claude Partner Network. More than 30,000 Cognizant associates have already completed Claude training, against a stated ambition to eventually reach the company’s 350,000 plus associates. Cognizant also committed to readying 5,000 Frontier Certified Engineers and 10,000 Frontier Business Operators credentialed directly by frontier-model companies, inside a certification pipeline reaching 40,000 professionals. The published results from that work are operational, not theoretical: contract review time cut by up to 40 percent with extraction accuracy above 88 percent for a biopharmaceutical client, and underwriting research reduced from hours to roughly a minute, saving each underwriter about eight hours a week in that deployment.
OpenAI. On July 22, OpenAI shipped Presence, an enterprise platform for deploying trusted voice and chat agents. OpenAI reports that Presence powers its own English-language phone support line and now resolves 75 percent of inbound issues without human assistance, and that a Codex-powered improvement loop cut human handoffs by 15 percentage points in ten days. BBVA, SoftBank, and IAG are named as early adopters. Here is the detail most people skipped: OpenAI states plainly that Presence “is not yet available as a self-serve product,” and that deployments are led by OpenAI Forward Deployed Engineers and select global systems integrators.
Read that last line again. The most capable AI company in the world shipped its flagship enterprise product and did not ship it as software. It shipped it as a service, because it knows the software alone does not land.
Now the number that ties it all together. Gartner’s May 2026 forecast puts worldwide AI spending at 2.59 trillion dollars in 2026, up 47 percent year over year. Inside that total, AI services are forecast at 585.5 billion dollars. AI models are forecast at 32.6 billion dollars. The services line is roughly eighteen times the model line.
Gartner’s John-David Lovelock added the framing: “Enterprises have yet to really flex their spending potential. That is coming and 2026 will be the inflection year.”
The money is not in the model. It never really was.
Sell the Landing, Not the Launch
So what do you actually do with this?
You build a services practice around implementation. Not around building AI, not around reselling AI, and not around explaining AI. Around getting one specific thing working inside one specific business, and proving it in numbers the owner cares about.
Here is why this is available to you and not just to Microsoft and Cognizant.
Every organization I named above is chasing the top of the market. Microsoft Frontier’s early partners are the London Stock Exchange Group and Unilever. Cognizant is deploying into biopharmaceutical and insurance clients. OpenAI Presence is being deployed into BBVA and SoftBank. Those firms will fight over the Fortune 500 for the next decade, and they will win it.
Underneath that tier sits the overwhelming majority of the economy: the seven-figure agency, the regional manufacturer, the fourteen-person law practice, the medical billing company, the property management firm. Those businesses have the same implementation problem and none of the same options. No forward deployed engineer is flying out to a fourteen-person firm. The economics do not work for a global integrator, and they work beautifully for an individual operator.
The second reason this is available to you is that the actual skill required is not technical. Go back to the MIT finding. Pilots failed because of brittle workflows, missing context, and misalignment with daily operations. Every one of those is a business problem, not an engineering problem. The person who can sit with an office manager, watch how the intake process really runs, notice the four places a human is retyping information that already exists, and design an AI-assisted workflow around that reality, is more valuable in this market than the person who understands transformer architecture.
You already have that skill if you have ever run anything. Most entrepreneurs discount it because it feels ordinary. In this market it is the scarce input.
The third reason is that the raw capability is now cheap, general, and stable enough to build a practice on. You are not betting on a research breakthrough. You are assembling tools that already exist into a workflow that already exists, for a business that already has revenue. That is a services business, and it does not require you to train anything.
What it does require is a willingness to take responsibility for the outcome rather than the deliverable. The consultants who lose in this market will be the ones who deliver a strategy deck. The ones who win will deliver a working process, a trained team, and a before-and-after number.
Practical Steps to Build a Deployment Practice
Here is how I would start if I were beginning this week.
1. Pick one workflow, not one industry. Choose a single repeatable process you understand well: inbound lead response, quote generation, intake, invoice reconciliation, support triage. Depth in one workflow beats shallow familiarity with a whole vertical, and it lets you get faster with every engagement.
2. Learn the deployment stack, not the model math. Your competence needs to be in connecting systems, writing clear instructions, setting guardrails, defining escalation rules, and testing outputs. Notice that OpenAI’s own description of Presence is mostly about policies, guardrails, escalation, and evaluation. That is the actual job.
3. Run the first one inside a business you already know. Your own company, a client, a friend’s operation. You need a real deployment with real messiness before you need a website. The first engagement is for the case study, not the invoice.
4. Measure the before and after in the owner’s language. Hours per week, days to close, cost per ticket, error rate. Cognizant did not sell “AI transformation.” It reported eight hours a week saved per underwriter. Numbers like that are what get you the second client.
5. Package the outcome, not the hours. Sell a defined result with a defined scope: “your quoting process, cut from three days to same day, live in six weeks.” Scope creep kills implementation practices faster than pricing does.
6. Build a handoff, not a dependency. Document the workflow, train the team, and leave behind something that runs without you. This feels like giving away leverage. It is the opposite. Businesses refer the consultant whose work still functions six months later.
7. Sell the next one while you finish this one. Implementation work is lumpy. Start the conversation for the following engagement before the current one closes out, so you are not restarting your pipeline from zero every time.
Frequently Asked Questions
Do I need to know how to code to sell AI implementation services?
No. The failure points MIT identified were brittle workflows, missing context, and misalignment with daily operations, which are business problems. Technical fluency helps you connect tools and test outputs, but the scarce skill is understanding how work actually flows through an organization and redesigning it.
What kind of businesses actually pay for this work?
Businesses with repeatable, high-volume processes and enough revenue to feel the cost of inefficiency. Agencies, professional services firms, regional manufacturers, medical billing companies, and property managers all qualify. The common trait is not size or industry; it is a process someone is currently doing by hand every single day.
How is this different from being an AI consultant who gives advice?
Advice ends at a recommendation. Implementation ends at a working process with a measured result. That distinction is the entire reason Microsoft, Amazon, and OpenAI are staffing engineers inside customer organizations rather than shipping documentation. Clients have plenty of advice already and almost no working deployments.
Will the big firms and frontier labs take all of this work?
They will take the top of the market. Microsoft Frontier’s early partners include the London Stock Exchange Group and Unilever, and OpenAI Presence deployments are led by its own engineers and global integrators. Those economics do not reach a fourteen-person firm, which is where the unserved demand sits.
How do I price implementation work with no track record?
Start with a small, tightly scoped engagement on one workflow and price it against the value of the outcome rather than your hours. Deliver it, document the before and after numbers, and let that result set your pricing for the next one. Evidence raises rates faster than confidence does.
The Gap is the Business
Step back and look at what actually happened this month.
Microsoft put 2.5 billion dollars and 6,000 people behind deployment. Amazon put a billion behind the same idea. Anthropic tied itself to a services firm training tens of thousands of people to implement Claude. OpenAI shipped its most important enterprise product and refused to sell it as software, because software alone does not land.
Four of the most capable technology organizations on the planet independently concluded the same thing in the same four weeks. The hard part is no longer building the intelligence. The hard part is getting it to show up inside a real business, on a real Tuesday, in the middle of real work that people are already behind on.
That is not a temporary condition. Gartner’s own forecast expects AI services spending to reach 759 billion dollars in 2027, which means the implementation problem is projected to get bigger before it gets smaller. The capability curve and the adoption curve have separated, and the space between them is where the work is.
You do not need a research lab to stand in that space. You need to understand how a business actually operates, pick one process that is bleeding hours, and make it work. Then do it again, a little faster, with a case study behind you.
This is exactly the kind of practical implementation work we walk entrepreneurs through inside White Beard Strategies. If you want to turn this shift into a real offer rather than an interesting article you read once, grab a training replay or join the membership, and let us help you build it.
Nobody is waiting for you to build a model. They are waiting for someone to make the one that already exists show up for work on Monday.
Jonathan Mast is the founder of White Beard Strategies, where he helps entrepreneurs move from buying AI to actually running on it. He is a speaker and builder of practical AI systems, and he is convinced that in this era the most valuable skill is not knowing what AI can do, but knowing how to get it working inside a business that already has customers to serve.
