Anthropic’s ad bill is soaring, according to IPO filing

That still leaves the ad budget small in Anthropic’s overall costs. Sales, marketing and partnership costs reached $3.17 billion in the first half of 2026 but most of that went to cloud companies that resell Claude, whose fees accounted for $2.7 billion of the increase.

Nevertheless, it’s a long way from 2023, when Anthropic’s ad spend was so small the S-1 filing called it “not material.” Claude had only gone on sale commercially in the first quarter of that year Anthropic made $41 million in revenue across the whole of it. The company has far more to sell these deals and most of it has arrived since the start of 2025. Claude Code, a coding agent, came out that February. A run of new models followed over the rest of the year, ending with Opus 4.5 in November. Claude Cowork, which brings the Claude Code approach to everyday office work, landed in January 2026.

More launches meant more to advertise and Anthropic’s ad budget grew to match. The S-1 filing attributes the 2025 rise in ad spend to “higher media and public relations spend to support product launches”, and gives the same reason, without the PR, for this year’s outlay. Sales and marketing costs rose in every quarter the filing covers, with the biggest increases in the first half of 2026. 

The move appears to have paid off. The filing lists Opus 4.5 among the reasons quarterly revenue growth went from 48% in the last three months of 2025 to 132% in the first three months of 2026. Revenue for the first half of this year came to $16.3 billion, compared with $1.2 billion in the same period of 2025. 

Why the spend keeps growing

Anthropic gets paid when people use Claude. Most of its revenue comes from customers billed like an electricity meter, or by how much work Claude does for them and which model does it. A business that builds Claude into its own app pays for the usage that app generates, so a busy month means a bigger bill. Those usage fees brought in $13.1 billion of the $16.3 billion Anthropic made in the first half of 2026. Subscriptions, where customers pay a fixed fee for a set amount of use made up the rest. The filing expects usage to remain “the substantial majority of total revenue for the foreseeable future.”

That makes Anthropic’s growth a question of getting people to lean on Claude harder. Each extra coding job run through a metered account adds to the bill. Subscribers who hit their limits can trade up to a pricier plan or pay for more. With so much riding on it, the ad budget isn’t going to shrink any time soon. 

The S-1 filing says as much. “The promotion of our brand requires us to make substantial expenditures, which we anticipate will increase as our market becomes more competitive,” it reads. Anthropic also worries it “may fall behind competitors with more established or widely recognized brands.”

That worry has run underneath the AI race all year. The models are close enough in ability that what people think of each company counts for a lot, and Anthropic knows it doesn’t fully control that. The filing warns it could be “grouped with other industry participants and judged similarly, regardless of differences in our approach or conduct.” It also flags the public perception that its products “contribute to job displacement or economic description could generate significantly negative media coverage.”

“This document makes it exceptionally clear that they consider their brand one of their greatest assets,” said eMarketer’s principal analyst, AI in marketing and commerce, Nate Elliott. “They use the word ‘brand’ more than 50 times just in the risks section. (Although, to be fair, that section is almost 100 pages long. It turns out AI vendors carry lots of risk.)”

Anthropic has spent 2026 trying to show it’s a different kind of AI company, and its place has cost it. In February, it said publicly that its tech shouldn’t be used for mass domestic surveillance or fully autonomous weapons. President Trump responded by ordering federal agencies to stop using Claude, and the Department of War labelled Anthropic a “supply-chain risk to national security”. Anthropic sued. In August a federal court in California ruled largely in its favor. A month later, though, an appeals court in Washington upheld the supply-chain risk in a 2-1 decision and Anthropic has since asked the full court to look at it again. 

Two months after the Pentagon fight began, Anthropic held back its most powerful model. It released Mythos in April only as a gated preview for a small group of organizations working on cybersecurity, citing its “advanced capabilities in cyber vulnerability discovery.”

Keeping ads out of Claude is part of the same pitch. The filing lists “monetization strategies that differ from those of certain competitors” among the ways Anthropic sets itself apart. It doesn’t name the competitor and it doesn’t need to. OpenAI has started running ads in ChatGPT in February this year, and Anthropic used part of this year’s Super Bowl to tell viewers Claude would stay ad-free. Its argument is that serving businesses pays better anyway. It says enterprise customers deliver “higher monetization than consumer AI.”

“And that makes sense in an industry with so much market share left to claim,” Elliott said. “Most U.S. internet users still don’t use AI chatbots in any given month; even for most of those who do use AI the cost of switching from one chatbot to another isn’t as high as, say, abandoning your email address. Under current market conditions, they’re right to consider maintaining a positive brand image paramount.”

The ad money, in other words, has two jobs. One is getting Claude in front of more people. The other is making sure Anthropic’s name, and not the industry’s general reputation, is what comes to mind when they think of it. 

Anthropic did not respond to Digiday’s request for comment.