The Gizin Dispatch #36

The real story: AI model companies have pivoted to 'deploying people.' This isn't the disruption of consulting — it's the redesign of PE's investment recovery model.

OpenAI's $10B subsidiary and Anthropic's $1B JV. The scale differs by 10x, but the structure is identical. The PE side is buying an AI adoption channel into their portfolio companies, while the AI side gains the 'last mile' of enterprise sales.

■ Three Points a CFO Should Watch

1. The equity structure reveals the strategy
OpenAI is offering preferred equity; Anthropic is offering common equity. OpenAI's structure is designed to 'protect PE's $4B' — with downside protection, drawing TPG in as anchor investor at a $10B valuation. Anthropic's common equity is a 'shared risk' design. With a smaller $1B scale, it preserves partnership parity.
This difference signals that OpenAI prioritizes scale, while Anthropic prioritizes relationships.

2. PE's real objective isn't 'AI adoption' — it's 'accelerating value-up'
TPG, Advent, Bain Capital, Brookfield — all mega-funds with hundreds of portfolio companies. Their rationale for committing $4B is straightforward: improve portfolio company EBITDA, and fund returns go up. AI adoption joins the traditional value-up toolkit (cost reduction, M&A, management overhaul). Forward-deployed engineers are the execution squad — this isn't 'AI consulting' but 'AI-ification of the PE investment recovery pipeline.'

3. GIZIN already has a structural lead
The 'deploy AI people into companies' model that OpenAI is now building is what GIZIN has been doing as Gizin. The difference: OpenAI deploys human engineers; GIZIN deploys AI employees. OpenAI's model hits a scaling bottleneck at headcount. GIZIN's doesn't. That's the fundamental structural gap.

■ A Question for Readers
Both OpenAI and Anthropic have conceded that selling model APIs alone can't capture the enterprise market. Their decision to bet hundreds of billions on 'deployment support' proves it. When your company adopts AI, outcomes will be determined not by model performance but by 'who walks alongside the implementation.' Whether you turn to human consultants or AI employees for that role — that's the next fork in the road.

Bottom line: OpenAI's strategic pivot isn't a 'failure of diversification.' It's the 'triumph of focus' that Anthropic proved.

Fidji Simo declared at an all-hands meeting: 'We can't afford to be distracted by side quests and miss this moment,' calling it 'code red.' What's being shelved: Sora (video generation), Atlas (browser), hardware devices, and ChatGPT's e-commerce features.

The real signal isn't what OpenAI is abandoning — it's what they're doubling down on. Coding tools and enterprise customers — precisely the domain where Anthropic led the way. The Ramp Index figure we reported last edition — 'Anthropic wins ~70% of new enterprise contracts' — was shared internally at OpenAI as a 'wake-up call.' The numbers visible from the outside carried the same weight on the inside.

A strategist should see three structural shifts here.

1. Focus beat diversification.
Anthropic never touched voice, image, or video generation. They concentrated on Claude Code and Cowork, embedding themselves in the enterprise workflow. OpenAI, under what Sam Altman described as 'betting on multiple startups internally,' ran a diversification strategy that ultimately scattered resources, resulting in 'compute frequently shuffled between teams' and 'products shipped without clear use cases.' The focused player captured the market; the diversified player is now chasing.

2. OpenAI's follow-through is the strongest validation of Anthropic's thesis.
When a competitor starts copying your strategy, consider your hypothesis market-validated. OpenAI's declaration of 'total focus' on coding + enterprise is, by itself, an endorsement that Anthropic's market read was correct.

3. 'Consumer AI dominance' and 'enterprise AI capture' are separate competitions.
OpenAI remains dominant in consumer AI. But as Simo explicitly stated, 'We especially need to lock down enterprise productivity' — the enterprise market is recognized as a distinct battleground. Consumer brand power doesn't determine enterprise adoption. Enterprises care about one thing: does it work in practice?

Read alongside this edition's PE × BigAI JV race (OpenAI $10B vs Anthropic $1B) and Meta × Nebius $27B infrastructure deal, and the full picture of the AI industry emerges. Capital is flowing in at unprecedented scale, but what separates winners from losers isn't the volume of capital — it's the quality of focus.

■ A Question for Readers
Reflect on your own AI adoption. Did you choose 'AI that can do everything' or 'AI that fits your specific workflow best'? OpenAI's pivot proves that even AI providers can no longer get away with 'doing it all.' Focus is demanded of the buyer, too.

Bottom line: BigTech is abandoning the build-it-all-yourself approach. 'Neocloud' isn't an infrastructure subcontractor — it's the power utility of the GPU era.

Meta × Nebius: a 5-year deal worth up to $27B (dedicated $12B + additional $15B). What matters isn't the dollar amount — it's the structural shift. Meta's 2026 AI capex is projected at $115B–$135B, and the decision to outsource $27B of that to an external provider says it all. There isn't enough time to build their own data centers — demand velocity has outrun infrastructure build speed.

Nebius is the infrastructure division of the former Yandex (Russia's largest search engine), spun off as an independent company. NASDAQ-listed (NBIS), headquartered in Amsterdam, with data centers in Finland and France. Revenue went from $91.5M in FY2024 to $529.8M in FY2025 — 479% growth. In March 2026, NVIDIA made a $2B strategic investment, selecting Nebius as one of the first large-scale deployment partners for NVIDIA Vera Rubin.

The essence of 'Neocloud' as a new business model
Traditional cloud (AWS/Azure/GCP) is general-purpose. Neocloud is different: GPU-native, AI-workload-dedicated, with long-term contracts for large customers. This is closer to a power utility model than a cloud model. Build power plants (GPU clusters), sell to large consumers via long-term PPAs (power purchase agreements), and allocate surplus to smaller customers. In fact, the Nebius contract structure has Nebius first selling GPU capacity to third-party customers, with Meta taking the remaining capacity. Meta is the largest consumer while simultaneously functioning as the backstop (buyer of last resort).

We see the same structural shift in GIZIN's own infrastructure operations. We run 30 AI employees on a Mac Studio + MacBook two-machine setup: 4 persistent processes + 25 scheduled jobs on the Studio make it the 'mothership,' while the MacBook serves as the CEO's operational terminal — roles have naturally separated. The scale is completely different, but the structural logic of 'consolidate compute and push it outward' is the same. It's faster to centralize on dedicated infrastructure than to do everything in-house.

Numbers an infrastructure engineer watches
Nebius's 2026 capex plan: $16B–$20B. GPU capacity expanding from 170MW to 800MW–1GW. First EBITDA profit in Q4 2025 ($15M, 24% margin), targeting 40% margin by end of 2026. Long-term contracts secure utilization rates and guarantee depreciation recovery. This is what separates Neocloud from pure cloud — infrastructure investment is justified not by spot demand but by 5-year contracts.

■ A Question for Readers
The AI capex explosion isn't just a BigTech story. Total BigTech AI investment for 2026 is estimated at $650B. Neoclouds like Nebius are emerging as the recipients of that massive capital flow. When your company goes all-in on AI, have you considered GPU-native Neocloud as an option alongside AWS/Azure/GCP? 'Where you compute' is becoming an executive decision that directly determines AI performance and cost.