The AI 100: benchmarks for the third supercycle

The internet took distribution cost to near zero. Cloud did it for compute. Each produced a trillion-dollar supercycle, and now intelligence is multi-trillion. As Bessemer Partner Sameer Dholakia likes to say, “intelligence is going to be bigger as a market than oil.”

In partnership with Newcomer Media, we’re unveiling the inaugural AI 100 to recognize the top 100 private AI companies. The AI 100 represents a completely new cohort of business: faster shipping velocity, different time-to-scale, capital-intensive, and new talent dynamics. The benchmarks behind this first cohort of the AI 100 are the evidence.

For a decade, the Cloud 100 was Bessemer Venture Partners’ definitive ranking of the world’s top private cloud companies, and within a decade that cohort grew from $99 billion to $1.1 trillion in that time. In 2023, AI-native entered the list as a category, and within two years, it was the most valuable one on it. The list didn't evolve by choice; this cohort simply outgrew the measuring stick.

TL;DR: the inaugural AI 100 by the numbers

The AI 100 is Bessemer Venture Partners’ annual ranking of the world’s 100 most important private AI companies, published in partnership with Newcomer Media. The inaugural list was announced on November 12, 2026, at the Cerebral Valley AI Summit in San Francisco, and succeeds the Cloud 100, which Bessemer published annually from 2016 to 2025. Companies were scored on market leadership (35%), estimated valuation (30%), operating metrics (20%, and people and culture (15%).

Here’s the TL;DR of the 2026 AI 100 by the numbers:

• The 2026 AI 100 is worth $3.6T, all of which is private value and more than three times the $1.1 trillion 2025 Cloud 100 it succeeds. A single year of AI-native company formation added more private value than a decade of cloud did.
• The average AI 100 company is valued at $36B, against $11.2B in the 2025 Cloud 100 cohort. The median tells the same story at a different scale: $9.3 billion this year, $5 billion last.
• Anthropic tops the inaugural list, ahead of OpenAI, Databricks, Stripe, Waymo, Cognition, Anduril, Ramp, Fireworks, and ElevenLabs. These 10 companies average $250B each and account for 70% of the list’s total value.
• 9% of the AI 100 holds 55% of its value. Frontier labs and foundation model builders represent $1,974B in aggregate, making it the most concentrated category in a decade of Bessemer’s benchmarking.
• AI 100 companies carry a 57x average ARR multiple, a new high above the Cloud 100’s 2021 peak of 34x. The median is a more measured 23x, against 20x for the 2025 Cloud 100.
• The average AI 100 company grew 480% year over year, and the top ten’s average is 306%. The median grew 275%, which is still faster than any year the Cloud 100 ever recorded.
• Each AI 100 company raised an average of more than $3.9B in capital, excluding Anthropic and OpenAI. The median capital raised is $1.09 billion across the AI 100.

The 2026 AI 100 at a glance

The inaugural AI 100 is worth $3.6 trillion and maps who is driving value creation in the AI ecosystem. The average company on the list is valued at 36 billion, more than 3x the $11.2 billion average of the 2025 Cloud 100. The median valuation is $9.3 billion, and a handful of giants are dragging the mean up. The median is what a normal company on this list looks like, and it’s nearly doubled from last year’s $5 billion. While most of the money is concentrated at the top, the opportunity is spread across the AI application layer. If you’re building outside the labs, the median indicates you’re in a good position.

Across 14 different categories, data and agent infrastructure is the top-represented category with 16 companies. But the labs demonstrate the power law at play: frontier labs and foundational models only represent 9% of the list, and yet hold 55% of the list value.

The top 10—Anthropic, OpenAI, Databricks, Stripe, Waymo, Cognition, Anduril, Ramp, Fireworks, and ElevenLabs—average $250 billion valuation and account for 70% of the list’s total value.

Last year the average Cloud 100 revenue multiple was 20x. The AI 100 is a record 57x. Well past the 2021 ZIRPera multiple peak of 34x. There is a simple reason for these big multiples—companies are growing faster than ever.

Average ARR growth is 480%, about 6x the Cloud 100. Even the median is 275%, and these are companies at significant scale.

For nine years, the Cloud 100 never left the 55% to 100% range. You might think you should pay a premium for that growth, but here’s the twist: on a growth-adjusted basis, the AI 100 is the cheapest cohort in recent history, with an average 12x multiple on a growth-adjusted basis.

Looking ahead: trends across the AI ecosystem

The AI 100 is a new type of list with varying dynamics across business models, categories, and more. Some companies have effectively no revenue alongside companies approaching $100 billion in ARR. For example, Anthropic went from founding to roughly $70 billion in annualized run rate in just five years. And #45 Legora reached Centaur status within 18 months, earning its second $100 million ARR in under six.

As part of the application process for the AI 100, we surveyed over 300 AI-native companies. These answers give us insight into forward-looking trends:

1. 79% believe frontier model quality converges in the next 2-3 years
2. 85% believe compute demand will continue to outstrip supply
3. 87% believe value will accrue to the application layer, not to model providers

Most believe model quality will converge. Today 57% run the majority of workloads on closed frontier models, but there are plans to diversify as those rates may fall to 36%. Enterprises aren’t just shifting to open source, rather a mix across closed, open-weight, and custom models, including fine-tuned or trained models. Many are acting on the belief that the future belongs to those who own their intelligence.

And the companies making that possible are all on the list: #9 Fireworks, #23 Together, and #11 Baseten for serving open and custom models. #94 Applied Compute and #21 Thinking Machines’ Tinker for post-training. #15 Mercor for expert training data. #16 Fal for media generation. #3 Databricks for putting models next to enterprise data. #32 Mistral and #51 Cohere for buyers who need the weights on their own soil.

85% of respondents expect demand to keep outstripping supply. When there’s that much demand, the money follows. In the last year, an entire financing market has sprung up around GPUs. Today, we’re also seeing more companies move to owning their own fleet instead of renting, as owning is increasingly more attractive than unpredictable rental rates and expensive prepays. 87% of the companies we surveyed expect value to accrue to the application layer rather than to model providers. The list already reflects it.

Value is accruing to the application layer

The application layer on the AI 100 is worth $629 billion across roughly 40 companies, at an average valuation of $15.7 billion. That average sits well above the $9.3 billion list median—the typical application company here is worth more than the typical honoree overall.

The labs are one story, but the application layer is thousands of them: payments, spend management, drug discovery, law, medicine, HR, security, insurance, defense, and beyond.

The next frontier: cybersecurity, biotech, and physical AI

We believe cybersecurity (3), biotech (3), and robotics (13) are the biggest AI opportunities ahead. Both cybersecurity and biotechnology only have three companies represented, and together represent <3% of the list’s value, so consider this a call for startup innovation!

Physical AI is the counterexample showcasing that AI doesn’t stop at software. 13 companies have a cumulative valuation of $332 billion, which is about 9% of the list’s value.

Some of these companies are already driving millions in revenue. Some are effectively pre-revenue: Physical Intelligence, Mind Robotics, Bedrock. The move from lab to commercial deployment is already in motion: Waymo is on the road. Saronic is building vessels to defend our seas. Bedrock is building data centers. We expect a lot more here. Next year’s list will tell us how fast the pre-revenue companies cross over.

Capital concentration in the AI 100

Taking a closer look at how capital-intensive these businesses are: six companies have each raised more than $10 billion, with OpenAI at $224 billion, Anthropic at $130 billion, and then Databricks, Waymo, Prometheus, and Anduril. Together, OpenAI and Anthropic have raised $354 billion—about 80% of what the six have raised combined.

Excluding the two labs, the average AI 100 company has raised $3.9 billion, with a median of just over $1 billion. To make this list, you’ve already raised more than a billion dollars, which was rare just five years ago.

Great businesses beget great founders

The AI 100 is a powerful list not just because of the innovation and value it drives, but because of the next generation of companies it fosters. 14 of the 100 companies were founded by someone who previously worked at another company on the list.

Famously, OpenAI trained and inspired the founders behind Anthropic as well as #21 Thinking Machines, #37 Safe Superintelligence, #94 Applied Compute, #52 TypeSafe, and #86 Periodic Labs. It’s clear that great businesses beget great founders.

#4 Stripe is another AI finishing school. Three of its alumni went on to found #97 Render, #24 Physical Intelligence, and #88 Chai Discovery. And more AI 100 companies are spinning out the next generation, such as Waymo alum building Bedrock Robotics. We imagine part of the next AI 100 will be built within this first list, so if you’re hoping to preview next year’s winners, you may want to take a closer look at this year’s companies.

Ten years ago, we published the first Cloud 100. In 2016, the cumulative value of the list was $99B. That Cloud 100 cohort grew to $1.1T by 2025. And now The AI 100 is worth $3.6 trillion. That’s 36x in a decade.

The AI 100 is now the bellwether for what’s driving innovation and where human intelligence is choosing to build their careers. We look forward to next year to see how the AI economy grows given what’s on the horizon.