Recently, OpenAI officially announced the establishment of the OpenAI Deployment Company (hereinafter referred to as “Deploy Co”), led by TPG with a total investment of over $4 billion from 19 investors, valuing the company at $14 billion. The company’s core business is to deploy FDE (AI Frontline Deployment Engineers) to clients’ companies and embed the models behind ChatGPT into the company’s data, processes, and workflows. Also in May, Anthropic announced a joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs, promising capital of about $1.5 billion, to do the same thing, namely sending engineers to clients’ offices.

The two investments totaling approximately $5.5 billion are the most structural events in the global AI sector since 2026. They collectively mark a point where leading model companies begin to acknowledge that relying solely on selling APIs can no longer support their valuation and that they must transform into something akin to a consulting firm, as defined by Palantir in the mid-2000s “frontline deployment” model. This article will dissect the capital structure, motivations, and labor implications of this pivot.

$4 billion, 17.5% guaranteed return

According to an official announcement from OpenAI, Deploy Co is controlled by OpenAI, with external investors committing over $4 billion, with TPG leading the round, and Advent International, Bain Capital, and Brookfield acting as co-founding partners. The remaining 16 investors include SoftBank Corp., Goldman Sachs, Warburg Pincus, BBVA, B Capital, Emergence Capital, Goanna, WCAS, and other private equity and strategic capital firms.

What is truly unusual is the detail of the capital structure. According to Axios, citing informed sources, external investors are receiving preferred shares rather than common stock, with a structure that includes two key terms, OpenAI promising investors a minimum 17.5% return and setting a profit cap. In other words, this is not a routine equity financing but a structured deal resembling subordinated debt, where investors have a downside protection and an upside limit.

This arrangement is uncommon in the private equity sector. SaaStr pointed out in its April analysis that “PE firms typically target an internal rate of return (IRR) above 20%, but rarely is it guaranteed contractually by the portfolio company”. MarketWise interprets this structure as indicating that PE investors are cautious about the valuation and cash burn of OpenAI’s core entity; they prefer to hold preferred shares of the subsidiaries with guarantees rather than ordinary shares of OpenAI. Given that the valuation of OpenAI’s core company has reached about $852 billion (according to StartupHub.ai), this arrangement of “the core company cannot refinance and therefore uses subsidiaries with structured terms” serves as a signal in itself.

Another detail disclosed by Axios at the same time is that the pre-money valuation of Deploy Co is $10 billion, with a post-financing valuation of about $14 billion. This means that OpenAI is simultaneously packaging “future enterprise-level AI service revenues” as measurable cash flow assets to sell to 19 institutions.

The implementation is facilitated by the acquisition of Tomoro. Tomoro is an AI consulting and engineering company formed “in alliance” with OpenAI and registered in London in 2023, headquartered in London, with offices in Edinburgh and Manchester, and has established its regional headquarters in Singapore over the past year, with branches in Sydney and Melbourne. Its client list includes Tesco, Virgin Atlantic (for which it built an AI travel concierge), Supercell (which launched an in-game support agent servicing 110 million users within 12 weeks), Fidelity International, Red Bull, Mattel, and the NBA. Tomoro claims to have quadrupled its employee count in the past 12 months, with global monthly revenues growing over tenfold and this acquisition will bring in about 150 “experienced frontline deployment engineers and deployment specialists” to Deploy Co.

Bain, McKinsey, Capgemini invest simultaneously

Among the list of 19 investors, the most unusual aspect is not the private equity firms but three consulting companies: Bain & Company (the twin consulting firm of Bain Capital), McKinsey & Company, and Capgemini.

Axios columnist Dan Primack provided two interpretations for this arrangement. The softer interpretation is that these three consulting firms will use this opportunity to gain deeper insights into OpenAI's capabilities and roadmap, which they can then relay to their own clients. The sharper interpretation, however, suggests that OpenAI has convinced these traditional consulting firms to invest in a company that could disrupt them.

This dynamic is reflected in a more subtle manner in the $1.5 billion joint venture on the side of Anthropic. According to the Wall Street Journal, the capital structure of the joint venture includes Anthropic, Blackstone, and Hellman & Friedman each contributing about $300 million, with Goldman Sachs as a founding investor contributing about $150 million, and the remaining capital filled in by Apollo Global Management, General Atlantic, GIC, Leonard Green, and Sequoia Capital, totaling approximately $1.5 billion.

The positioning of the joint venture was described by Blackstone's COO and President Jon Gray as “breaking one of the critical bottlenecks in enterprise AI deployment,” by “expanding the engineer workforce capable of actual implementation.” Goldman Sachs’ Global Asset and Wealth Management Chief Marc Nachmann stated in the announcement that this joint venture will “enable mid-sized enterprises to access Anthropic’s solutions, democratizing access to the highly scarce frontline deployment engineers.”

Notably, the target customers of both joint ventures, Deploy Co and Anthropic JV, are locked on the companies invested by PE firms. Blackstone, Apollo, TPG, Bain Capital, Brookfield, Advent, Warburg Pincus, and others collectively manage over 2,000 companies, forming a vast, contract-bound internal distribution channel. Embedding the models into the operations of these invested companies serves as a source of returns for LPs and a tool for PE partners to reduce costs and improve profitability.

Anthropic's Reversal is the True Motivation Behind OpenAI's Bet

Venture capital firm Menlo Ventures has published a semi-annual enterprise-level LLM market share report since 2023, and its year-end version for 2025 shows that Anthropic currently holds a 40% share of the enterprise LLM API market, a significant increase from 24% last year and 12% in 2023; OpenAI’s market share dropped from 50% in 2023 to 27%, nearly losing half its enterprise share; Google rose from 7% in 2023 to 21%.

The gap in coding market share is even more pronounced. Anthropic holds about 54% of the programming market share, while OpenAI has 21%; Anthropic has consistently ranked first in programming evaluation rankings for 18 months since the release of Claude Sonnet 3.5 in June 2024. Menlo Ventures partner Deedy Das stated, “Anthropic is sweeping the enterprise market, OpenAI has given up nearly half of its share.”

This reversal in market share creates direct pressure on OpenAI’s management. In March of this year, OpenAI’s application business CEO Fidji Simo referred to Anthropic’s progress as a “wake-up call” during an internal all-hands meeting, characterizing OpenAI's response as “code red.” According to the Wall Street Journal citing meeting notes, Simo told employees, “We must not get distracted by marginal requests and miss this moment” and urged the company to “deliver results in productivity, especially enterprise-level productivity.”

The timeline thus becomes clear. In March, Simo sounded the internal alarm; in April, OpenAI entered advanced negotiations for a $10 billion joint venture with TPG, Advent, Bain Capital, and Brookfield; on May 4, Anthropic was the first to officially announce a $1.5 billion joint venture; and on May 11, OpenAI announced Deploy Co and the acquisition of Tomoro. The entire process was driven by Anthropic’s market share data, pushed into rhythm by the penetration speed of Claude Code.

The Internal Mapping of the White-Collar Flip: 800% Surge in FDE and Simultaneous Decline in SWE Demand

Deploy Co and Anthropic JV need to address the issue of human resources, specifically the supply issue of FDEs.

According to publicly available data from Indeed, the number of FDE job postings in the U.S. surged from 643 to 5,330 in the past 12 months, a year-on-year increase of 729%. LinkedIn data shows that from January to September 2025, FDE job postings in the U.S. surged over 800% year-on-year, making it one of the fastest-growing categories within tech jobs. In terms of regional distribution, New York has replaced San Francisco as the top hiring location for FDEs, accounting for about 35%, while San Francisco accounts for about 11%, driven by New York’s absorption of FDEs in financial services and regulated industries.

Salary ranges are significantly higher than traditional software engineers. According to PitchMeAI citing Anthropic’s public job postings, the base salary range for Applied AI FDE positions at Anthropic in the U.S. is $280,000 to $320,000; for mid to senior level FDEs at OpenAI and Anthropic, total compensation (TC) has stabilized in the range of $350,000 to $550,000, with some staff-level positions approaching $600,000. The average TC for FDEs at Palantir is about $238,000, with staff-level positions exceeding $630,000. New graduates typically start at a TC of $180,000 to $250,000.

In stark contrast to the explosive growth of FDE roles is the continuous decline in traditional software engineering positions. According to Indeed via FRED data, the number of software engineer job postings nationwide has decreased by between 35% to 45% from the peak in mid-2022, reaching a five-year low by early 2025. Research from the Stanford Digital Economy Lab based on ADP payroll showed that the employment of early software engineers aged 22 to 25 has slid nearly 20% from the end of 2022 peak. An industry observation by Pragmatic Engineer pointed out that AI-related infrastructure roles and engineering positions in regulated industries are still expanding, but demand for traditional SWE roles in most other industries is in a retraction phase.

Rowan, the co-founder and CEO of Apollo Global Management, referred to the AI wave as “without a doubt the biggest technology cycle of our careers” during Apollo’s quarterly earnings call, predicting that “nearly every job will be enhanced or replaced, and we will see a complete flip with blue-collar jobs rising and white-collar jobs under pressure.” Concurrently, Blackstone’s President Jon Gray made similar judgments at the Milken Conference, believing that AI will drive blue-collar employment into a “great boom.”

The rise of FDEs within Silicon Valley mirrors the macro logic described by Rowan, though in a more subtle way. FDEs are “engineering blue-collar workers” packaged with high salaries; they are required to travel 50%, work on-site at client offices, manage outdated client systems and compliance audits, debug within data silos, and respond to the political demands of client CIOs. This contradicts the long-held Silicon Valley creed of “zero marginal cost, pure software, remote work.” The FDE model essentially pushes engineers back to the frontlines, next to clients and into specific business contexts.

The Pragmatic Engineer's editor Gergely Orosz stated in an analysis in May: “In the arrangements of OpenAI and Anthropic, FDEs are categorized as independent subsidiaries, meaning that newly hired FDEs are likely to receive equity in Deploy Co or Anthropic JV rather than shares of the parent company.” In other words, the valuation premium at the model level and the labor premium at the deployment level are structurally separated. The parent company is selling a “future income”, while the subsidiary is engaged in a “labor-intensive business far exceeding SaaS”, with both linked through structured terms.

The Model Layer is Commercializing, the Deployment Layer is Capitalizing

Connecting the four threads reveals a relatively complete narrative of a turning point. The differences at the model layer are narrowing, with OpenAI, Anthropic, and Google collectively occupying 88% of the enterprise API market, with model quality assessments increasingly converging. However, the success rate of enterprise AI deployment is estimated by the industry to be between 5% to 20%, with the difficulty of implementation being the real bottleneck for AI commercialization. Anthropic has proven in 18 months that in a path of homogenized models, a more focused product and more stable enterprise deployment capabilities can surpass early movers.

OpenAI's response is to use capital structure to chase time. The $4 billion Deploy Co is not a routine financing, but a “future enterprise income securitization” arrangement with guaranteed returns and profit caps, circumventing the awkwardness of the core company’s valuation being hard to refinance. Anthropic’s $1.5 billion joint venture transforms the network of investment firms within PE into its own channel. The total amount of these two transactions expands the boundaries of AI giants from “model APIs” to “on-site deployments,” bringing the core profit pool of the traditional consulting industry into the competitive landscape.

The simultaneous investment by Bain & Company, McKinsey, and Capgemini gives this pivot special significance at the level of financial gamesmanship. Whether these three consulting firms approach this with a mindset of “understanding the opponent” or are prepared to be partially disintermediated, they are essentially providing capital for their potential future competitors, a pattern that is exceedingly rare in the history of the consulting industry over the past 20 years.

The explosive growth of FDE positions alongside the decline of traditional software engineer roles is not contradictory. They are two sides of the same structural flip; after the intelligence spillover from the model layer, companies are no longer willing to pay for “just writing another piece of software,” but rather “to ensure this AI can actually run in my business.” The former is increasingly approaching commodification, while the latter increasingly approaches a high-premium service. Using Rowan's words to describe it, it is a complete flip, and the world is not prepared for it.

The next point of observation is whether PE giants such as Carlyle, KKR, and EQT, which have not yet entered the market, will follow suit, and whether Meta's announced Enterprise Solutions department will follow up with a similar structure. If they do follow, this will define the capital narrative of enterprise AI from 2026 to 2027; if not, then this $5.5 billion bet will merely be a self-rescue of leading AI companies.