Anthropic S-1: What It Reveals About Enterprise AI Adoption

Key takeaways

• More than half the Fortune 500 pay Anthropic at least $100,000 a year, but that can still be a very small deployment
• Average spend is under 0.1% of total expenses, and even the top 10 customers stay below 1%
• Cowork, Anthropic’s tool for non-technical staff, is described as growing faster than Claude Code did
• Cheaper AI per task has so far led to more use, not less
• The big unknown is how much growth depends on hands-on deployment help

Anthropic, the company behind Claude, filed its S-1 on 26 October. That is the document a company files with US regulators before it lists its shares publicly, and it gives outsiders their first proper look inside the business.

Most headlines have focused on whether AI is a bubble, the size of the losses and how much Anthropic spends on computing power. In a new video, AI strategist Nate B Jones deliberately looks at something else: what Anthropic’s customers are actually doing with Claude, and what that says about where AI spending is heading.

We covered the filing as a news item in this week’s AI news round-up. This article goes deeper into the customer side, explains Nate’s argument in plain English and pulls out what it means for a small business.

What does Anthropic’s S-1 show about its customers?

The S-1 shows wide reach but shallow use. According to the filing, as Nate describes it:

• Anthropic reported $16.3 billion in revenue for the first half of 2026
• More than half the Fortune 500 generate at least $100,000 a year in revenue for Anthropic
• Among the listed companies in that group, spending with Anthropic averages under $1 for every $1,000 of total expenses in the September quarter
• The customer list includes the 10 largest banks and nine of the 10 largest pharma companies in the Fortune 500

Nate’s headline point is that people are asking the wrong question. Instead of “is AI a bubble?”, he challenges the assumption that today’s AI budgets define the size of this market. In his view the market is small now, not big, because most companies have barely started.

He also notes that the large net loss in the filing mostly reflects financing and accounting charges, which is not the same thing as the cash it costs to run Claude.

Why is winning a customer different from becoming important to one?

Getting a customer in the door is not the same as becoming essential to them. For a Fortune 500 company, $100,000 a year can be a pilot in a couple of teams. Anthropic’s own filing says most enterprises using Claude have deployed it across a small number of teams and use cases compared with what Claude could handle.

So a well-known logo on the customer list tells you Anthropic has been tried. It does not tell you whether that company will spend ten times more next year. That depends on whether the business keeps finding new work worth handing to AI after the first project.

Why do the biggest buyers still spend under 1%?

Because AI is still a tiny line item even for the keenest buyers. Nate highlights that Anthropic’s 10 largest Fortune 500 customers spend more than eight times the group’s average share of expenses, yet even they average below 1%.

For Nate, that is mostly opportunity. A company does not need to replace much other spending for Anthropic to grow a lot. It only needs AI to help the people, software and information a business already pays for do more.

What is the bear case for enterprise AI spending?

The bear case is that big companies are adopting AI because they feel they must, not because they have found deep value. Nate plays devil’s advocate here. AI is mostly used by individuals, and businesses struggle to turn that individual help into measurable returns unless they plan for it very deliberately.

Many large companies, he says, simply saw everyone else adopting Claude Code (Anthropic’s coding agent), gave it to their engineers and waited to see what happened. If that is all that is going on, the impressive growth could be a one-off: engineering departments are where Anthropic tops out.

He draws a clear line between two stages:

1. Table stakes: “we need AI because everyone has it”, which he thinks most companies have reached
2. Deep embedding: “we are rebuilding how we work around AI because it gives us outsized value”, which is the next step and not yet proven

What is the bull case beyond Claude Code?

The bull case is that non-technical teams are starting to use Claude too. Nate says the S-1 shows Cowork, Anthropic’s product for non-technical work, growing faster than even Claude Code did. If that holds, coding was only the starting point and other departments are next.

The filing describes a familiar pattern: an employee uses Claude for their own work, someone finds a useful application in analysis or research, and use spreads. Nate points out that this is the general story of AI adoption, not something unique to Anthropic.

His own expectation is that a $100,000 customer could become a million-dollar customer for many firms within a year or so, as agents get good enough for everyday non-technical tasks. Treat that as his forecast, not a figure from the filing.

Why does deployment work set the pace?

Because putting AI into a real process takes real work. Someone has to connect the data and systems, give the AI the right context, decide how its work will be checked and make the whole thing repeatable. That is why both Anthropic and OpenAI are investing in forward-deployed engineers: specialists who help customers set AI up inside their workflows.

Nate makes a point that matters for any size of business: these investments add up. Once your data, systems and instructions are connected and usable, the next AI project is easier. The filing even says Claude can use the context it builds up inside a company to suggest the next workflow worth automating.

Anthropic is also putting money behind this. Nate says it has committed $100 million to a Claude Frontier Academy, aiming to train 10,000 engineers by the end of next year to lead Claude projects in their own organisations.

How does Claude reach customers through other products?

Claude is sold directly, inside other companies’ products and through cloud marketplaces. Nate highlights that firms such as Salesforce and Legora (a legal AI company) build products on top of Claude, so a business can end up paying for Claude twice: once directly and once inside a vendor’s tool.

• In the September quarter, more than half of customers paying at least $100,000 used both Anthropic’s platform and one of its own apps such as Claude Code, up from about 30% at the end of 2025
• 48% of first-half revenue came through cloud marketplaces
• More than 80% of September-quarter revenue came from model versions released in the previous six months

This is why Nate thinks a normal software lens misses the picture. Counting staff licences ignores AI that runs processes or sits inside something a company sells. We explored the same idea in per seat vs per token AI pricing.

Why can cheaper AI tasks still grow total spending?

Because when a resource gets cheaper and more efficient, people tend to find more uses for it. This is known as the Jevons paradox. According to the filing, Opus 5.5 cut customer cost per task by about 40% compared with Opus 5, through lower prices and better task efficiency.

Nate checked his own usage and found he now uses fewer tokens per useful task, but spends as much or more overall, because he finds more things worth doing. He admits that is one person’s experience, but says it matches what he hears from others. Anthropic also says customers tend to use more after releases that are both more capable and cheaper per task.

He adds that a lot of agent usage is cached content, meaning repeated information that is far cheaper to process, so raw token counts can be misleading. The real question for Anthropic is whether customers buy enough extra work to outweigh the lower price per task.

What does the S-1 answer about margins and deployment?

Back in June, Nate asked five questions about AI labs. Here is how he answers them now the S-1 is out:

What does this mean for small businesses?

You do not need to be a Fortune 500 company for this to matter. The pattern in the S-1 is the same one a small business goes through: try AI on one task, find it useful, then look for the next job it can take on. A few practical lessons:

• Treat your first AI project as a start, not the finish. Plan what you will try next once it works
• Do the groundwork once. Tidy, connected information makes every later project easier (see our data readiness guide)
• Pick workflows with care. Nate warns that real processes are messy and poorly documented, so choose the high-value step and make sure it succeeds
• Watch total spend, not just price per task. Cheaper models tend to lead to more use
• Measure the return. Use our ROI method before expanding

This fits what we believe at Prodify Digital: simple systems you own and understand, built step by step, usually beat big leaps. Your own data, email list and documented processes are what make AI useful, whoever’s model you use.

Frequently asked questions

The bottom line

Anthropic’s S-1 shows a company that has won a place in most big businesses but still takes a tiny share of their budgets. Nate B Jones reads that as early days, with growth depending on companies doing the slow work of putting AI into real workflows. Your next step: pick the next workflow in your own business and get its information ready.

Related posts

• GPT-6.1 Terra: OpenAI’s Cheaper Fast Model, Anthropic’s IPO Filing and This Week’s AI News
• Per Seat vs Per Token AI Pricing: Why It’s the Wrong Question (and What to Ask Instead)
• How to Measure the ROI of an AI Tool in a Small Business

Garry Knight

I'm Garry Knight, the person behind Prodify Digital. I write about email list building, email marketing, SEO, AI search and the tools that connect them. My aim is to make online marketing easier to understand, so creators and small business owners can make informed decisions about building an audience and keeping people engaged. Here you'll find straightforward guides and product reviews that explain what something does, where it fits and which limitations matter. The focus is on clear explanations and useful next steps—not hype, shortcuts or promises of easy earnings.

See Full Bio