Microsoft, AWS and Anthropic are spending billions - and not on better models

On July 2, Judson Althoff, CEO of Microsoft’s commercial business, announced the formation of the Microsoft Frontier Company. The new operating unit will embed 6,000 industry and engineering experts inside customer organizations to design, deploy, and run AI systems. Microsoft is backing it with $2.5 billion. Rodrigo Kede Lima, most recently president of Microsoft Asia, will lead it.

The announcement landed just two days after AWS committed $1 billion to a forward-deployed engineering organization of its own. Anthropic and OpenAI had already launched service ventures on the same pattern in May. Within a fortnight, every major AI platform vendor has moved toward the same conclusion.

The limiting factor for enterprise AI has shifted from the model itself to the engineering resources needed for deployment. Such a pivot fundamentally reshapes the industry, forcing a recalibration of corporate talent acquisition, the competitive dynamics of global systems integrators, and the underlying unit economics for all stakeholders.

Inside the Microsoft Frontier Company

The Frontier Company is not a separate legal entity. Most of its people already work at Microsoft and are part of existing forward-deployed engineers, technical consultants, support staff, and industry-focused sales teams. Headcount will grow through internal moves and outside hiring. Its initial customers include Unilever and Novo Nordisk. Microsoft has also lined up forward-deployed engineering partnerships with global system integrators, including Accenture, Capgemini, EY, KPMG, and PwC.

Althoff positioned the unit as more than a services arm: “This goes beyond what has been labeled as Forward Deployed Engineering and will be the largest, most capable, outcome-driven engineering organization in the industry,” he writes in the announcement.

Microsoft calls the offering Frontier Transformation, which combines industry knowledge, change management, and enterprise AI engineering. The company commits not to use customer data or intellectual property for model training.

The financial details deserve a closer look than the headline number. Microsoft declined to say whether the $2.5 billion is fresh capital or repurposed from existing consulting budgets. It has not specified the spending period either. Microsoft already operates Industry Solutions Delivery, a sizeable consulting organization, alongside its FastTrack deployment programs. Whether the Frontier Company is genuinely a new capability or a rebranding with an increased focus on AI will become clear from customer engagements.

The Palantir playbook goes mainstream

To understand what these vendors are building, we need to look at the origins of the forward-deployed engineer role. Palantir created the role after its founding in 2003. It placed software engineers inside intelligence agencies and enterprise clients whose data could not leave the building and whose requirements changed faster than any specification document. The engineer worked within the customer’s environment, understood the workflows firsthand, and shipped working systems rather than slide decks. The model remained unique to Palantir for over a decade, and its public filings still describe forward-deployed engineers as staff who help customers identify use cases and modernize data architectures.

The frontier labs adopted the playbook first. In May, Anthropic launched an AI services venture with Blackstone, Hellman & Friedman, and Goldman Sachs, aimed at midsized businesses deploying Claude. Days later, OpenAI unveiled its Deployment Company, backed by TPG, Advent International, Bain Capital, and Brookfield. Both labs chose joint ventures, which bring outside capital and partner firms into the deployment relationship rather than funding it from their own balance sheets.

The pattern goes beyond the platform giants and frontier labs. Cursor runs its own forward-deployed engineering team under Pauline Brunet. She describes the role at the AI Engineer World’s Fair as going on-site, working inside a customer’s systems, and deploying customized agent workflows across the software development lifecycle. A startup selling a code editor and the world’s largest software company have agreed on the same delivery model within a quarter. The embedded engineer closely follows the enterprise AI playbook rather than a niche tactic.

AWS vs. Microsoft: Two designs for the same job

There are quite a few similarities between the AWS and Microsoft programs. Both embed vendor engineers within customer teams to build production AI systems on the customer’s own data and under the customer’s governance - both fund the effort from their own balance sheets rather than through joint ventures.

Sprint versus embed

The key difference lies in the engagement design: AWS optimizes for short, self-liquidating deployments, whereas Microsoft builds a persistent presence within the account.

Francessca Vasquez, vice president of frontier AI engineering and services, announced the AWS organization on June 30. Engagements run in roughly 45-day cycles, with each customer hosting a pod of five or six engineers. The unit describes itself as agentic-first, meaning the teams use purpose-built agents to build agentic systems. Success is measured by whether the customer becomes self-sufficient, with knowledge graphs, runbooks, and trained internal staff left behind at the end of an engagement. AWS says teams are already embedded with Cox Automotive, the NBA, the NFL, Ricoh, and Southwest Airlines.

Microsoft, by contrast, describes continuous co-design and improvement with measurable business outcomes, a model that looks closer to an embedded transformation partner than a 45-day sprint. That approach is a calculated risk, since the same depth that strengthens the relationship also deepens the customer’s dependence on Azure tooling.

The four initiatives fundamentally differ in structure, capital, and intent.

Each row carries a trade-off rather than a ranking. The joint ventures shift cost and risk to partners but dilute control over the customer relationship. The hyperscaler models keep control while absorbing the full cost of thousands of embedded engineers.

Why every vendor reached the same conclusion

The economics behind the convergence paint a different picture. Research from MIT, McKinsey, RAND, and Gartner arrives at broadly the same conclusion that most enterprise AI pilots fail to produce a measurable return. Enterprises bought subscriptions to ChatGPT, Claude, Gemini, and Copilot at scale. They then realized that impressive demos rarely succeed when integrated with proprietary data and legacy workflows, and that, due to organizational inertia, deployment rather than model quality becomes the critical factor.

Althoff himself supplied the most candid admission of the quarter. He said that Microsoft made a mistake by binding the original Copilot exclusively to OpenAI models. Customers, according to him, care more about the combination of their data and the models than about any particular model. With the Frontier Company, Microsoft enables customers to mix OpenAI, Anthropic, Microsoft AI, and open-source models in a single engagement. Model choice becomes just a configuration while the embedded engineers become the durable relationship.

What buyers should ask

Customers can use this approach to their advantage. An enterprise negotiating an AI engagement can now demand outcome-based pricing and insist that the vendor’s engineers train internal staff rather than create dependency.

The first question for enterprise buyers is ownership. Which parts of the deployed system are proprietary logic the customer keeps, and which are wrappers around the vendor’s managed services?

The second question for buyers is exit cost. When an AWS pod leaves after 45 days, or a Microsoft team concludes its rotation, what does it take to run and evolve the system without them?

Global system integrators will be under pressure. Accenture, Capgemini, EY, KPMG, and PwC are listed as partners in Microsoft’s announcement. Yet the Frontier Company competes for the same transformation budgets those firms have owned for decades. A vendor with thousands of embedded engineers and direct access to its own product roadmap can undercut an SI on both speed and depth. The integrators that thrive will be those that move up to industry-specific process knowledge that platform vendors cannot replicate.

The forward-deployed engineering race marks a milestone in enterprise AI where distribution dominates capability as the competitive front. Model quality gaps between frontier vendors have narrowed, and enterprises now hedge across providers as a matter of policy. Proximity has now become the new focus of competition. Vendors now fight over who sits closest to the customer’s data, workflows, and budget.

Microsoft’s answer is compelling and convincing on paper, provided the Frontier Company turns out to be new capacity rather than a fresh coat of paint on its existing consulting business. If the early engagements with Unilever and Novo Nordisk turn into referenceable outcomes, the Frontier Company becomes the template against which AWS, Google Cloud, and the frontier labs are measured.

The next signal to watch is Google. A formal forward-deployed organization from Google Cloud would complete the hyperscaler set and confirm that the embedded engineer is now as strategic to the AI platform wars as the model itself.