The most important lesson in Anthropic's reported growth is not the size of the numbers. It is how the money is earned: by charging for work the product does, not for the number of people who log in. Every B2B company that prices per seat should take that seriously.
What SaaStr reported
In a SaaStr analysis, Jason Lemkin lays out Anthropic's disclosed and reported run-rate over the first half of 2026, from roughly $9 billion exiting 2025 to about $47 billion by mid-May, and projects $70 to $90 billion by the end of the year. At that level, he argues, only Microsoft's software and cloud business would still be larger among public software companies. He compares it with Salesforce, which he cites at around $41 billion in revenue, and notes that Oracle's larger headline figure is mostly cloud infrastructure rather than software.
Treat the year-end figure as a projection, not a result. The trajectory is the point: Lemkin attributes it largely to consumption pricing on tokens, and notes Claude Code alone reached a multi-billion-dollar run-rate within about nine months.
Related: AI Agents Struggle to Read B2B Pricing Report: A Study by @sejournal, @MattGSouthern
Why seat-based pricing is under pressure
Per-seat pricing made sense when software was a tool people used. The more work software does on its own, the weaker the link between value and headcount becomes:
- Fewer users, more output. If an agent does the work of several people, a customer needs fewer seats even as they get more value.
- Budget moves to outcomes. Buyers compare the cost of a tool with the cost of the work it replaces, not with other tools.
- Expansion stalls. Seat growth depends on hiring. Usage growth depends on value delivered.
That does not mean every company should switch to usage pricing tomorrow. It does mean the pricing conversation now starts with "what work does this do?"
Pricing on output: questions to ask
Before changing your model, work through these with your team:
- What unit of work does your product complete that a customer would recognise as valuable?
- Can you measure that unit reliably and show it to the customer?
- Does usage rise as customers get more value, or can they get value while usage falls?
- How will customers budget for a bill that varies month to month?
- What happens to your gross margin when usage spikes?
If you cannot answer the first two cleanly, output pricing will create billing disputes rather than growth.
Related: Demystifying Demand Generation Services Pricing: What You Need to Know
An example from visitor identification
Pricing on outcomes is not new in every category. At Happierleads, credits are consumed only for successfully identified visitors, and our Qualify step means you are only shown, and charged for, visitors who match your ideal customer profile. The customer pays for a result they can see, a named prospect, rather than for access to a dashboard. You can see how that works on our pricing page.
The broader lesson: when the unit of billing matches the unit of value, the sales conversation gets simpler.
What this means for go-to-market teams
Output-based products change how you sell, not just how you invoice:
- Sell the result. Demos should show work completed, not features clicked.
- Shorten time to value. Usage-based revenue only starts when the product is doing work, so onboarding becomes a revenue function.
- Watch consumption like pipeline. Falling usage is an early churn signal; rising usage is an expansion signal.
- Rethink compensation. Paying reps only on signed contract value ignores the usage that actually produces revenue.
The risks of chasing the headline
It is tempting to read a story like this and assume any AI feature will produce similar growth. It will not. Anthropic sells a general-purpose capability that a very large market is adopting at once. Most B2B products serve a narrower job. The useful takeaway is the pricing logic, not the growth curve.
There is also concentration risk for buyers. As budgets flow to a small number of AI platforms, SaaS vendors that sit on top of them need a clear answer to "why not just use the model directly?"
What to do this quarter
- List the jobs your product completes for customers and pick the one that best reflects value.
- Start reporting that number to customers, even before you price on it.
- Test an output-based add-on or tier alongside your existing plans.
- Align onboarding and customer success goals to getting that number up early.
How buyers will evaluate output pricing
Changing your pricing model also changes how customers buy. Procurement teams used to seat-based contracts will ask new questions, and your sales team needs answers ready:
- Predictability: can the customer cap spend, commit to a volume, or get alerts before a bill spikes?
- Proof: can they see exactly which units of work they paid for and verify them?
- Quality: what happens when an output is wrong or not useful? Do they still pay?
- Comparisons: how does the cost compare with doing the same work with people or with another tool?
Vendors that answer these clearly tend to close faster, because the buyer can build a business case from the numbers. Vendors that cannot answer them end up discounting to overcome uncertainty. A practical middle ground many B2B companies use is a committed platform fee plus a usage component, which gives customers a predictable base and gives the vendor upside when value grows. Whatever model you pick, publish the rules plainly and make usage visible inside the product, so the invoice never surprises anyone.
The takeaway
Whether or not the year-end projection lands, the direction is clear: software that does work will increasingly be paid for the work. Find the unit of value in your product, measure it, show it to customers, and let pricing follow. Companies that make that shift deliberately will handle it far better than those forced into it by a buyer's spreadsheet.
