Ten weeks separate the first of these announcements from the last, and AWS came last. On June 30, Amazon Web Services said it would put a billion dollars into a new Forward Deployed Engineering organization, embedding its own engineers inside customer teams to build agentic systems on a compressed timeline the company calls 45-45-45: forty-five minutes to define a use case, forty-five hours to validate it, forty-five days to put it into production. The announcement landed at the AWS Summit in Washington, in front of an audience of federal and public sector technology buyers.

Set next to what OpenAI, Anthropic, and Google Cloud have already done this spring, the AWS number is the smallest of the four, and it is also the only one funded entirely from one company's own balance sheet. The other three each solved the identical problem, too few engineers who can turn a language model into a working production system, by bringing in outside capital. The differences in how they did it say more about what a customer is actually buying than the headline dollar figures do.

At least, that was the read of the market until Thursday, July 2. Just two days after AWS made its move, Microsoft completely shook up the board, committing a massive $2.5 billion to launch the Microsoft Frontier Company. Backed by 6,000 industry and engineering experts, Microsoft explicitly positioned this unit not as traditional forward deployed engineers, but as highly specialized Operational Interpreters—a unified force working directly with enterprise customers to build custom AI platforms while safeguarding their proprietary data.

The Prevailing Read Treats This as One Race

The easy version of this story is that four vendors independently arrived at the same insight: enterprises don't need another model, they need people who can wire the model into a real workflow. That reading, that this is a copycat land grab modeled on Palantir's decade-old embedded-engineer playbook, isn't wrong. It's also not the part that matters to a buyer. Four vendors converging on the same tactic tells you the tactic works. It tells you nothing about what happens to your account, your data, or your recommendation path once the engagement starts, and that is where the four programs stop resembling each other.

With Microsoft's entry into the fray, the tactics have evolved even further. Rather than just wiring models into workflows, Microsoft’s stated ambition is to establish a continuous learning loop where an enterprise's "unique IQ" compounds over time—pushing the FDE playbook squarely into the realm of systemic business transformation.

Who Owns the Engineers Matters More Than Who Funds Them

AWS staffed its FDE unit from its own AI engineering organization, building upon the group that has spent three years running customer deployments through the AWS Generative AI Innovation Center. No outside investor has a stake in the unit or a claim on its output, offering a single, uncomplicated line of accountability. Importantly, this internal engineering force doesn't operate in a vacuum. AWS explicitly designed the unit to work in tandem with the AWS Partner Network (APN), augmenting the massive ecosystem of global systems integrators rather than attempting to replace them.

OpenAI took a different path. The OpenAI Deployment Company launched May 11 with more than four billion dollars in initial capital from nineteen investment firms, consultancies, and systems integrators, led by TPG, with Advent, Bain Capital, and Brookfield as co-lead partners. OpenAI keeps majority ownership and control. But the unit's engineering base didn't originate at OpenAI at all. It came from the acquisition of Tomoro, a London-based consulting firm whose roughly 150 forward deployed engineers already had client relationships with Tesco, Virgin Atlantic, Red Bull, and the NBA before OpenAI ever entered the picture. Bain & Company, Capgemini, and McKinsey aren't just referral partners in this structure—they're co-investors with a financial stake in how the engagements perform.

Anthropic went further still. Its $1.5 billion venture, announced May 4 with Blackstone, Hellman & Friedman, and Goldman Sachs as founding partners, is not majority-controlled by Anthropic. Anthropic, Blackstone, and Hellman & Friedman each committed roughly $300 million; Goldman put in $150 million. The venture's explicit target market is companies already owned by private equity, meaning the engineers recommending how deeply to embed Claude into a portfolio company's workflows are working inside a vehicle partly owned by the same firm that owns the company being advised.

Google Cloud's $750 million, announced April 22 at Cloud Next, is structured differently again. Google isn't building a direct-to-customer engineering organization at all. The fund embeds forward deployed engineers alongside existing consulting and systems integration partners, including Accenture, Capgemini, Deloitte, and TCS, rather than putting Google's own people in the room with the customer. The relationship a buyer signs up for runs through whichever systems integrator they already have a contract with. Google is financing the deployment layer, not staffing it.

Late breaking news after AWS Summit

Microsoft's new Frontier Company most closely mirrors the AWS approach by using its own balance sheet, but drastically outscales it. Led by President Rodrigo Kede Lima, the 6,000 embedded Operational Interpreters are Microsoft's own people. However, unlike AWS’s structure, Microsoft's Commercial Business CEO Judson Althoff aggressively emphasized an "open, heterogeneous AI platform." They are explicitly offering to deploy models from OpenAI, Anthropic, or open-source providers—not just Microsoft AI—while heavily leaning on Global SI partners like KPMG, EY, Capgemini, and PwC to handle the global delivery scale.

None of the Vendors Are the Same Bet

AWS Launched at a Public Sector Summit Amid Complex Procurement Realities

The reference customers named in AWS's own announcement—the NFL, the NBA, Southwest Airlines, Cox Automotive, Ricoh, and the Allen Institute—are entirely commercial and nonprofit. None is a government agency, though this is a standard reality of public sector procurement. Government agencies typically take 6 to 12 months to clear public case studies for press release use. Furthermore, forward deployed engineers in these environments require stringent security clearances, specialized on-premise or GovCloud access, and strict compliance with FedRAMP frameworks that differ vastly from commercial deployments. Choosing a public sector summit as the launch venue signals where AWS sees the highest ceiling for this unit, setting the stage for government reference architectures to surface over the coming quarters.

Self-Sufficiency, Platform Gravity, and Open Standards

AWS describes the goal of each engagement as leaving customers self-sufficient: deployed systems, runbooks, architectural documentation, and trained staff who can operate independently once AWS exits. In practice, what these engagements leave behind is a robust semantic layer—a governed knowledge graph that continues to grow. To AWS's credit, this infrastructure is increasingly built on open standards like Apache Iceberg, and the multi-model optionality of Amazon Bedrock ensures customers can swap foundation models as the landscape shifts. Still, a well-integrated knowledge graph natively optimized for a specific cloud ecosystem inherently creates platform gravity. Ultimately, all the vendors are building toward a similar outcome: durable, highly valuable systems that keep driving enterprise workflows long after the deployment engineers leave the building.