Two days after AWS announced its $1 billion embedded-engineering program, Microsoft went bigger. On July 2, it launched Frontier Company, a $2.5 billion unit that will put 6,000 engineers and industry specialists inside enterprise clients to build and run AI systems on their behalf. Then, four days later, it cut 4,800 jobs.
The two moves together tell the same story. Microsoft is not simply adding a new service. It is redirecting the company toward it. The job cuts hit Xbox and commercial sales hardest, and Microsoft's own chief people officer confirmed they build directly on the Frontier Company announcement. Resources are moving from what the company is winding down to what it is building up.
Microsoft's stock has dropped roughly 21 percent so far in 2026, and the company carries a contracted but not yet recognized revenue backlog of $627 billion. Frontier Company is partly an answer to investors asking why that backlog is not converting faster. A dedicated unit with measurable outcome targets is a more direct answer to that question than another model release.
The total commitment now on the table across the four major players is striking. OpenAI's Deployment Company is backed by more than $4 billion in private equity. Anthropic put $1.5 billion into a joint venture with Goldman Sachs and Blackstone. AWS committed $1 billion from its own balance sheet. Microsoft's $2.5 billion is the largest of the four. Added together, the industry has pledged more than $9 billion in roughly eight weeks to solve the same problem: most companies cannot get AI from a working demo to something that runs in production.
Research from MIT found that 95 percent of enterprise AI pilots produce no measurable impact on profit and loss. The problem is almost never the AI itself. It is connecting the AI to old internal systems, satisfying security teams, changing the way people actually work, and keeping the system running after the vendor's team leaves. That last part matters more than most clients realize.
Gartner has already flagged a warning worth reading before signing any of these engagements. Its analysis suggests that by 2026, fees for this kind of embedded engineering work could run between $200,000 and $400,000 per quarter per project, before platform and integration costs. More pointedly, it predicts that 70 percent of enterprises will eventually abandon the AI projects started through these engagements, because the vendor's team becomes a permanent fixture and internal staff never develop the skills to take over.
That is the real risk here, and every one of these vendors says it is managing it. AWS describes leaving clients with documentation, trained staff, and working software after its 45-day sprints. Microsoft says its model is designed so that a client's internal intelligence compounds over time and does not walk out the door when the engineers leave. Both are promises worth testing in a contract before work begins.
One detail that separates Microsoft from AWS is the talent situation. Microsoft listed more than 1,200 AI-related open positions as of early July, and it is pulling staff from the consulting partnerships it already has with Accenture, Capgemini, EY, KPMG, and PwC. AWS is building its corps from scratch alongside AI agents that reduce the number of human engineers needed per engagement. The engineering talent for this kind of work is genuinely scarce: industry data puts US base salaries for senior engineers in these roles at between $215,000 and $310,000, with total compensation at the top firms regularly exceeding $500,000.
For anyone evaluating whether to engage with one of these programs, the practical questions are now more specific than before. Which vendor is already embedded in your infrastructure? What does the contract say about who owns what when the team leaves? What internal staff will be trained, and how is that measured? The era of buying a software licence and hoping it works is over. The era of buying an engineering team and hoping you can keep what they build is what comes next. That transition has real risks too, and it pays to understand them before the engineers arrive.