Sequoia's 2026 AI forecast of delays and fast adoption

Sequoia Capital partner David Cahn published his 2026 outlook in December 2025. His thesis is that 2026 will run on two tracks. Infrastructure and AGI timelines will slip. Adoption of AI by end users will keep rising. For founders, the second track is the one that matters, and it is the one Cahn sees no reason to slow.

Data centre delays as the first track

Cahn argues that demand for AI capital spending from Big Tech is stronger than ever, but the supply chain behind it has not scaled to match. He points to chip foundries and lithography suppliers with near-monopoly positions that cannot be forced to add capacity. He cites analyst Ben Thompson's observation that TSMC had grown revenue by 50% since 2022 while growing capital spending by only 10%.

He also flags industrial inputs that rarely get attention. Generators, cooling units and skilled construction labour can each hold up a data centre. Cahn notes that the average AI data centre takes roughly two years to build. Projects announced in 2024 and started in 2025 will either come online in 2026 or start to run late. He names one warning sign to watch: hyperscalers storing new chips instead of installing them.

AGI timeline delays

The second delay Cahn describes is in expectations. Talk of AGI arriving in 2027 was common in Silicon Valley. Cahn writes that a walk-back began in mid-2025 and that the new consensus places AGI in the 2030s at the earliest. He expects that view to spread beyond the Bay Area during 2026. His main risk for investors is that today's hyperscaler spending ends up outdated.

AI adoption as the second track

On adoption, Cahn is direct: he does not expect delays. He lists the facts he sees entering 2026:

• End revenue from AI is still on the order of tens of billions per year, against data centre and energy investment on the order of trillions over five years.
• Coding and ChatGPT are the two clear killer applications.
• Nearly a dozen more startups are on a path to cross $100M in revenue.
• Large enterprises are struggling to implement AI in-house, which is causing fatigue.

He predicts that 2025's "$0 to $100M" club of fast-scaling AI companies becomes a "$0 to $1B" club in 2026.

Revenue per employee as the efficiency signal

Cahn writes that many of the best AI startups earn more than $1M in revenue per employee. He reads that as evidence of market pull rather than a pushed sale. He also notes that these companies use AI agents for their own legal, recruiting and sales work, which compounds their efficiency. Application companies also ride a falling compute cost curve as new data centres come online through 2030.

The enterprise fatigue point matters for startups. When large companies fail to build AI tools in-house, they become buyers of finished products. Cahn sees that dynamic adding momentum to startups.

Founder takeaways

• Build at the application layer. Cahn's delays hit infrastructure and AGI timelines. His forecast for adoption is uninterrupted.
• Sell to enterprises that tried to build it themselves. Failed internal projects create budget and urgency for a working product.
• Track revenue per employee from day one. It is the metric Cahn uses to separate pull from push.
• Run your own company on agents. Use AI for internal legal, recruiting and sales tasks before you hire for them.
• Plan for cheaper compute. Model gross margin with falling inference costs, but do not depend on a specific date for new capacity.

Cahn closes by arguing that the next phase of AI will take hard work and endurance rather than a sudden leap. For a founder in Vancouver or anywhere else, that is a practical brief. The opportunity is in shipping products that customers adopt now, not in waiting for the next model.

Sources

1. AI in 2026: A Tale of Two AIs
   Sequoia Capital (David Cahn) · Dec 3, 2025