Elad Gil on AI market crystallization: 'We know who the finalists are' in foundation models and coding

Summary

Elad Gil argues that the AI market has entered a phase of competitive crystallization after roughly six to nine months of clarity emerging from what had been an unusually opaque landscape. The field of foundation model contenders is now identifiable: OpenAI, Anthropic, Google, xAI, Meta, Mistral, and a small number of others. In coding, Gil pins the finalists as Cognition, Cursor, the foundation model companies, and Microsoft. Healthcare has consolidated around Abridge and potentially Comir. By contrast, financial tooling, sales enablement, and accounting remain open races.

The TAM Shift Nobody Is Pricing Correctly

Gil's most pointed structural argument is that AI markets are being systematically undervalued because analysts are still applying seat-based SaaS pricing frameworks. The real opportunity is displacement of the labor market, not software licensing. Gil's team estimates the addressable services economy at approximately $5 trillion in US labor that AI can augment or automate. That reframes customer support, legal, and finance from small software TAMs into massive labor replacement opportunities — a distinction he says most investors are missing.

Forward integration by foundation model companies is the central durability risk for application-layer startups. The historical pattern is clear: Microsoft used Windows to dominate the Office suite, Google used search to kill vertical search players. Gil expects the same dynamic to play out, particularly in coding, and possibly in customer support and sales.

AI Roll-Ups: Three Requirements, Few Real Ones

Gil has reviewed several dozen AI roll-up teams and backed two. His framework requires three distinct capabilities: a strong AI operator, a disciplined M&A acquirer with a defined acquisition profile, and an operational leader capable of reorganizing workforces around new tooling. Most teams claiming to be AI roll-ups are, in his view, executing traditional PE buyouts — buying at PE prices, raising at AI valuations, and arbitraging the gap without deploying meaningful AI. He describes the genuine version as potentially "an AI Danaher" that converts services-margin businesses into software-margin businesses.

Brainco and the Enterprise Opportunity

Gil recently launched Brainco alongside Jared Kushner, Eric Wu (former CEO of Opendoor), and Louis Videgaray (former Finance and Foreign Minister of Mexico), targeting AI transformation for the world's largest institutions. The model involves a shared infrastructure platform handling data, evaluations, and enterprise AI foundations, with vertical-specific applications built on top. Target clients are organizations at the scale of JPMorgan, which is reportedly spending $2 billion annually on AI to recover a comparable $2 billion in savings — a break-even that Gil views as just the baseline of what's coming.

Adoption Will Take a Decade, Despite Real Revenue Now

Gil pushes back on both bubble skeptics and hype. Cursor is rumored to be in the high hundreds of millions in annual revenue. Microsoft Azure added an estimated $2–3 billion in AI revenue per quarter from a near-zero base two to three years ago, implying a $10 billion-plus annualized AI run rate on that platform alone. Yet Gil maintains full adoption will take roughly a decade, with organizational inertia — not technology readiness — as the binding constraint. He compares current enterprise AI pilots to the early BofA mobile app: clunky and crash-prone, but on the same trajectory that eventually produced mature mobile banking.

Competitive Durability Beyond AI

On companies like Rippling, Gil argues that deeply embedded HR and workforce platforms are more durable precisely because AI cannot easily displace them with a greenfield competitor. The primary risk is a secular decline in headcount that shrinks the seat count those platforms monetize — offset by the possibility that AI enables more companies to form at smaller team sizes, expanding the total customer base.

Gil describes himself as still underexposed to AI broadly, drawing a parallel to social media: after MySpace and Friendster failed, observers called social saturated, then came Facebook, LinkedIn, Twitter, Instagram, WhatsApp, and TikTok in succession. He views AI as materially earlier in that compounding cycle.