Meta is reportedly launching a new Enterprise Solutions unit that will embed engineers and product managers inside corporate customers to push them toward its AI tools, according to The Information. The unit, led by head of product Naomi Gleit, is the latest sign that the most expensive job in enterprise AI is no longer the researcher building frontier models but the engineer flying to a customer site to make those models work.
The move makes Meta the latest major AI company in a 13-week run of deployment-heavy enterprise announcements. OpenAI announced its Frontier Alliances on February 23, pairing its Forward Deployed Engineering team with McKinsey, Boston Consulting Group, Accenture and Capgemini. On May 4, Anthropic announced a joint venture, reportedly valued at more than $1.5 billion, with Blackstone, Hellman & Friedman and Goldman Sachs that embeds its engineering resources inside a standalone enterprise services firm. One week later, OpenAI announced The OpenAI Deployment Company, capitalized at more than $4 billion and led by TPG with Advent, Bain Capital and Brookfield as co-lead founding partners, and acquired Edinburgh-based Tomoro for its roughly 150 deployment engineers. The two ventures together put about $5.5 billion behind AI deployment and enterprise services.
The Forward Deployed Engineer, or FDE, is a senior software engineer who sits inside a customer’s environment, learns its data and politics and ships production code against the mess that legacy IT systems present. Palantir pioneered the title more than a decade ago and called its early embedded engineers “Deltas.” The model long looked like a margin disaster to traditional software investors, and most of the industry ignored it until deployment became the binding constraint. Palantir’s Q1 2026 results, with 85 percent year-over-year revenue growth and 133 percent growth in U.S. commercial business, gave the market fresh proof points for a model already gaining momentum.
Why Model Access Stopped Closing Deals
The shift is happening because raw model access is no longer enough to win an enterprise account. A 2025 MIT report found that about 95 percent of enterprise generative AI pilots showed no measurable profit and loss impact, a failure the researchers traced to flawed integration rather than weak models. The bottleneck was not model capability but the gap between a polished demo and a working integration inside legacy systems, change advisory boards and compliance regimes that move at their own pace.
OpenAI executives have been blunt about this. Enterprise customers already account for more than 40 percent of OpenAI’s revenue, according to CNBC reporting on commentary by chief revenue officer Denise Dresser, with the company expecting parity with consumer revenue by year end. That mix only holds if pilots convert to production, and the labs have concluded that conversion does not happen without engineers on the ground.
Anthropic has been on this track for longer. The company rolled out Claude to more than 470,000 Deloitte employees in October 2025 in what remains its largest enterprise deployment, then expanded its Snowflake partnership in December into a multi-year $200 million agreement that puts Claude in front of more than 12,600 Snowflake customers across AWS Bedrock, Google Vertex AI and Microsoft Azure. That cross-cloud detail matters because the hyperscalers are not bystanders. They own the data planes where workloads run, gatekeep procurement and in many accounts distribute the models, so any deployment firm has to work through them.
The Collision With System Integrators
The strategic question for buyers is who actually does that hands-on work. Before 2026, most of it flowed through global consultancies, offshore-led integrators, hyperscaler professional services and specialist boutiques, with Accenture, Deloitte, TCS, Infosys and Capgemini positioning their AI services lines to absorb the next wave.
That positioning is now contested. The Anthropic joint venture is framed in its announcement as a way to expand the supply of skilled implementation partners, and Anthropic is careful to call it additive to its existing integrator relationships rather than a replacement for them. Even so, by deploying Claude across portfolio companies of Blackstone, Hellman & Friedman and the other backing investors, the new firm creates a parallel channel that can bypass traditional integrators in those accounts.
The OpenAI Deployment Company is an even more direct challenge. It is a majority-owned OpenAI subsidiary that employs engineers, bills clients and acquires firms. OpenAI also runs the Frontier Alliances program with the four largest consultancies, which, on any honest reading, makes the same vendor both a partner and a competitor, since McKinsey, BCG, Accenture, and Capgemini are certifying practice groups in OpenAI technology while OpenAI builds an in-house alternative.
The Indian system integrators face the sharpest squeeze. Infosys, TCS, Wipro and HCLTech have competed on cost-effective delivery of engineering talent at scale, while the lab-owned firms are positioning on proximity to the model and direct access to the roadmap. A Forward Deployed Engineer at the OpenAI Deployment Company can route a customer request straight into the product in a way a third-party engineer cannot.
What The Labs Cannot Yet Do
The economics of the FDE model are unproven outside Palantir. Industry trackers report a sharp rise in postings over the past year, though figures vary widely by methodology, and senior technical roles at OpenAI and Anthropic can reach the high six figures, even if it is unclear whether FDE pay scales that high across thousands of deployments. Those numbers only work if the labs can productize repeatable patterns out of each engagement. Palantir spent a decade building Foundry and AIP to turn artisanal FDE work into reusable platform IP, and the frontier labs are betting they can do the same far faster.
Coverage is another open question. The lab-owned firms are starting with marquee accounts, private equity portfolio companies and existing enterprise relationships, not the mid-market and regional customers where Indian integrators have built deep practice areas. The hyperscaler consulting arms complicate things further, since AWS, Google Cloud and Microsoft both compete with and partner alongside the new firms.
What CXOs Should Take From This
For technology buyers, the implication is concrete. The implementation market is sorting into rough tiers rather than clean categories. At the premium end sit lab-owned firms whose engineers have direct access to model behavior and roadmaps. Global consultancies sit in the middle, with multi-platform certification and change-management muscle. Offshore-led integrators compete on scale, cost and incumbency. In practice, a single program still pulls in a model lab, a hyperscaler, a data-platform vendor, and often a boutique shop at once.
That makes ownership the question worth pressing. CIOs negotiating new AI engagements should ask whether the deployment team has a feedback channel into the model owner, who controls architecture and data integration, and what happens if the underlying model is deprecated or repriced. The labs are building moats out of implementation labor, and the open question is whether buyers end up locked into a single model family, as enterprises were once locked into a single ERP vendor.
Meta’s entry is a useful signal. If the company with arguably the strongest open weights story in the industry has decided that selling Llama through partners is not enough, the conclusion is hard to avoid. Model access has become a commodity input, and the labs with the most to gain are spending billions, roughly $5.5 billion between the OpenAI and Anthropic ventures alone, on the engineers who install those models inside customer environments.
