Elad Gil on which AI markets have winners — and which are still wide open

Solo VC capitalist extraordinaire Elad Gil said on signifier astatine TechCrunch Disrupt that AI has been 1 of the slightest predictable tech booms he’s ever seen.

Gil is connected the headdress array of virtually each deed institution of the past decade, including galore of today’s starring AI companies.

Still, helium thinks that implicit the past year, definite AI markets look to beryllium astir sewn up by marketplace leaders. Beyond these areas, a immense swath of AI remains anyone’s game.

“I started investing successful generative AI successful 2021 … astatine the time, not precise galore radical were paying that overmuch attraction to it,” Gil said. But helium had seen the monolithic leap successful capableness betwixt GPT 2, launched successful 2019, and GPT 3,  launched in 2021. “The measurement betwixt 2 and 3 was truthful ample that if you conscionable extrapolated retired the scaling laws, oregon the curve, past you could truly presume that this was going to beryllium incredibly important,” helium said.

That convinced him to commencement backing early-stage startups gathering products powered by ample connection models. His bets included some foundational exemplary makers similar OpenAI and Mistral, arsenic good arsenic exertion companies similar Perplexity, Harvey, Character.ai, Decagon, and Abridge. Yet passim 2024 and overmuch of 2025, the capabilities of foundational models leaped with each release, upending AI each fewer months.

“I utilized to accidental astatine the clip that AI was the 1 marketplace wherever the much I learn, the little I know. Usually, the much you larn astir something, the amended you cognize it, the easier you tin foretell the future, etc. But AI was conscionable hazy. There’s conscionable excessively overmuch uncertainty. And I deliberation there’s inactive markets similar that successful AI,” helium said.

However, he’s besides present seeing markets with wide winners. The astir evident illustration is with foundational models themselves. Even though hundreds of models exist, and immoderate countries similar South Korea are still moving present to make sovereign models by section companies, leaders person emerged. “Google, Anthropic, OpenAI, possibly xAI, possibly Meta, possibly Mistral — it’s similar a handful,” helium predicts of the winners.

After models, helium thinks AI-assisted coding has runaway winners that volition marque it hard for caller entrants to drawback up. Not lone person the foundational exemplary makers moved successful (Anthropic with Claude Code, OpenAI with Codex) but startup leaders similar Anysphere’s Cursor and Cognition’s Devin (which acquired Windsurf) volition beryllium hard to beat. And determination are well-funded startups like Magic (whom Gil called a imaginable “outlier”) oregon Poolside connected their tails.

He sees aesculapian transcription arsenic being cornered, with Abridge a beforehand runner and a fistful of others similar Ambiance being “important.”

He names lawsuit enactment – which was an aboriginal people of some accepted AI and the caller harvest of AI cause startups – arsenic having hard-to-catch marketplace leaders, specified arsenic his portfolio institution Decagon. (It raised $131 cardinal astatine a $1.5 cardinal valuation in June.) OpenAI president Bret Taylor’s startup, Sierra, competes successful this space. This is besides an country wherever the incumbents — Salesforce, Hubspot, and galore others — are adding AI offerings.

So which markets look wide open? Gil says fiscal tooling (fintech), accounting, AI security, and “other markets that we cognize are by default precise interesting. We conscionable don’t cognize who’s going to bash it.”

Ironically, accelerated maturation isn’t the awesome it erstwhile was that a institution is going to beryllium a breakout hit. “The CEOs of each large institution are fundamentally telling their teams, hey, we person an edict. We request to fig retired our AI strategy,” Gil said. “These elephantine enterprises are consenting to effort things that 2 years agone they ne'er would person tried, and it’s lone due to the fact that of AI.”

So caller AI markets tin onshore a batch of gross from big-name, endeavor customers quickly, “but that doesn’t mean they’re going to stick,” Gil points out.

It is lone aft a marketplace goes done its trial-phase roar rhythm that a startup and investors tin spot if this gross volition enactment and grow. “There’s mendacious signal, and past there’s worldly that is conscionable working,” Gil said. He calls retired ineligible AI startup Harvey arsenic 1 of the market-leaders that’s “just working.” It raised 3 monolithic rounds successful 2025, leaping from a $3 cardinal valuation to $5 billion to $8 billion, successful conscionable a fewer months.