Posts

Anthropic filed their S-1¹ this week, the largest AI listing yet.

Founded in 2021 by Dario & Daniela Amodei, the company has grown from a research lab into the second-largest model developer, selling Claude to more than 11,900 customers who spend over $100,000 a year. The comparisons below set the reported OpenAI figures² against Anthropic’s derived gross margin³, adjusted operating margin⁴ & compute commitments⁵.

Anthropic generated $16.3B of revenue in the first six months of 2026, a fourteen-fold increase over the $1.2B of the first half of 2025. That half-year alone was more than three times all of 2025, when Anthropic grew 11.9-fold to $4.6B, more than four times OpenAI’s 253%.

Every major technology build-out requires two phases, according to Carlota Perez¹: the infrastructure to power it, & then the applications to pay for it.

Between 1996 & 2005, American telecom companies spent $1.5t building the internet in today’s dollars.² Meanwhile, applications riding on that infrastructure booked $180 billion of revenue.³

That’s an interesting ratio, isn’t it? Application revenue to infrastructure spend. It matters because the shorter the gap between the two, the sooner the infrastructure pays for itself. Let’s compare this to the AI era.

There are two markets in AI : applications & infrastructure. Bardia Shahali has built sales teams both : first at Fireworks AI from roughly $3m to roughly $400m, & now VP of Sales at Granola.

In our conversation at Office Hours earlier this week,¹ we talked about the differences in building modern sales organizations in each domain. Here are three takeaways :

Watch the full conversation on YouTube : Office Hours with Bardia Shahali

Yesterday brought two data points about the health of inference businesses : SpaceX’s third-quarter report¹ & DeepSeek’s inference post.²

Overall, they demonstrate that inference businesses can be quite healthy even at current prices & efficiencies. Plus there is a tremendous amount of upside assuming demand persists at higher prices per card hour.

DeepSeek calculates a $38 maximum per GPU card-hour.³ It describes a GPU running under optimal conditions : maximum utilization for external customers rather than internal work, & 100% of its output usable & billable, sold at full list price.

We all consider Anthropic an AI company.

But what if it’s a cloud?

Anthropic is growing much faster than the other clouds : 15x last year. GCP’s growth is far behind at 82%. But at current growth rates, Anthropic is closer to GCP’s pace, doubling next year.¹

Five frontier models now sit within five points of each other in terms of intelligence, but their cost per unit differs by 3.8x.

Google’s Argon prices at the bottom of the cluster at $1.99 a task¹, 74% below the most expensive model, Fable 5.1. It is also 67% below Opus 5.5, 56% below Muse Wonder & 39% below Astra.

The 2026 Theory AI Survey surprised us, just as the 2025 edition did. In 2025, we found that context, limited AI performance. In 2026, that seems to be solved.

455 senior technical builders replied to give us a cross-section of the AI ecosystem today.¹

Security is the biggest barrier to adoption

Security rose from sixth to first among pain points in twelve months. 43% of respondents rate security & governance the most important pain point, & evals come in second at less than half, 22%.

On Tuesday, October 20th at 10 AM Pacific, Office Hours will host Bardia Shahali, VP of Sales at Granola, for a conversation on what the sales organization should look like in the AI era.

We’ll talk about :

• Why AI-era sales teams run smaller, and how far quotas have moved from traditional SaaS benchmarks.
• How faster ramps and larger deals change who you hire and how you compensate reps.
• How to structure go-to-market when the product spreads on its own.

Register here for Office Hours : The Leaner Sales Org with Bardia Shahali.

What if AI models have just reached commodity status? In early 2025, we debated whether labs would become the airlines of the AI industry : high cost, low margin businesses with undifferentiated products.

Over the past month, the evidence has mounted :

• OpenAI cut prices on Luna by more than 80% to win share.¹
• Frontier models’ share of tokens slipped from 53% in August into the mid-40s as companies defaulted to smaller, cheaper tiers.²
• Open source demand has shifted decisively from frontier-dominated to medium-&-small tiers.³
• Spending among the top 1% of adopters fell 9.7% in August to $7,205 per employee per month, cooling from its July peak.⁴
• Decision models like TypeSafe’s Jev are winning share of LLM calls at $0.04 per million input tokens, roughly 75x cheaper than standard LLMs for routing, gating, & classification.⁵

Throughout all this, token usage has surged 50% since July, while prices have fallen 41%.⁶

Jeanne DeWitt Grosser, COO of Vercel, joined me on Office Hours to walk through how she & her team actually built the agent that now runs the top of Vercel’s sales funnel.

It started with one engineer.

Vercel founded a go-to-market engineering team & handed the problem to a single engineer spending roughly 20% of his time on it. The first version of the agent was a prompt of about 125 lines, written by the best SDR on the team, encoding Vercel’s rules for qualification.