Anthropic’s Demand Problem Is a Good Problem—Until It Isn’t

You’re reading an issue of "The AI Economy," my newsletter exploring the forces shaping the AI era—tracking how AI is rewriting business, work, technology, and culture. Subscribe to get expert insights and curated updates delivered straight to your inbox.

Has Anthropic’s “Fail Whale” moment arrived?

The company announced this week on social media that, due to growing demand for Claude, users will run through their five-hour session limits much faster during peak hours. Notably, it comes days after Anthropic touted a “limited-time promotion” that doubled the usage limit for those who used Claude outside the 5 am to 11 am PT window on weekdays.

To be clear, I’m not suggesting that Anthropic is in decline, just that it’s experiencing extraordinary strain at a critical point in its history.

Starting now, anyone using Claude—free, Pro, or Max—will be subject to this new policy. However, weekly limits are said to be left unchanged. Thariq Shihipar, a member of Anthropics’ technical staff, posted on X that approximately seven percent of users will “hit session limits they wouldn’t have before, particularly for pro tiers.” He recommends shifting token-intensive background jobs to off-peak hours.

Several distinct forces seem to be driving the surge in demand for Claude right now. One of which is Anthropic’s deepening feud with the Department of Defense—it saw a surge in new sign-ups after OpenAI made a deal with the Pentagon. The company’s series of Super Bowl commercials, which apparently irked Sam Altman, has contributed to the growth, as have Anthropic’s release of new products such as Computer Use, Claude Code, and Claude Cowork.

Growing demand can be a good sign that your product is valuable, but it can also expose whether a company is truly ready to scale. Although Anthropic’s fortune appears to be on an upward trajectory—securing a legal, albeit temporary, victory against the DoD and testing a powerful new model called Mythos—throttling the user experience is a shock.

Scroll through LinkedIn, X, or other social media platforms, and you’ll find no shortage of business and tech leaders crediting Claude Code with transforming how their teams work. Personally, I have found Claude invaluable in assisting me with research and developing my OpenClaw intelligence site, ClawBeat.co.

Organizations are even distributing Claude Code to their entire workforce. Andrii Yakovenko, the senior analytics engineer for AI transformation at Intercom, posted on LinkedIn this week about what happened when the company gave everyone access to the technology. He writes that, by doing so, it dissolved the boundary between who could and could not build software and created a self-improving loop in which the entire organization helped the AI get smarter about Intercom. And Yakovenko is not the only one to tout the advantages of AI.

Meta, Apple, Google, Salesforce, Shopify, Klarna, and Spotify are some of the companies that have highlighted how AI has transformed their businesses. And last week, at GeekWire’s Agents of Transformation event in Seattle, Microsoft’s Executive Vice President for Agents and Business Apps, Charles Lamanna, remarked that there are humans on his teams who “do not write a single line of code directly anymore—zero lines of code written by a person directly into the code base.”

And coding won’t be the only thing impacted. Agentic workflows could also be affected, whether they’re bots for personal use or for enterprise use, potentially compromising the quality of customer service. Negatively impacting the enterprise would be bad, especially since Anthropic reportedly holds a 73.3 percent market share among first-time customers. For months, the AI model maker has been aggressively pursuing the business market, deploying new programs meant to entice those in finance, marketing, and design. Though some may consider this change in usage limits to be an inconvenience, it could foreshadow potential scaling issues for the company.

Hopefully, Anthropic has thought through the consequences of its decision. It’s reasonable to believe the company has examined Claude usage through peak hours and run simulations to assess how this action will impact customers. And to be fair, Anthropic says only the consumption rate will be faster, not that it’s reducing the total token limits. Still, it could disrupt how organizations use Claude—will teams be willing to reschedule deployments simply because “it’s peak hours,” or start prioritizing development based on how many tokens may be required? Will token-rationing become a thing?

Shihipar acknowledged the change is “frustrating” but said Anthropic has worked to improve efficiency to soften the blow—and promised more to come. As always, time will tell to see how customers really respond. Will they adapt or flee? Rate limiting and usage caps are standard practice, so it shouldn’t surprise anyone. However, suggesting when to use a tool like Claude may be uncommon. It seems similar to a utility company telling residents to conserve power to avoid brownouts or blackouts in their neighborhoods.

All of this is unfolding at a particularly consequential time for Anthropic. Reports surfaced this week that the company is eyeing an IPO as early as October, with preliminary talks already underway with Goldman Sachs, JPMorgan, and Morgan Stanley. It stands to be one of the largest tech listings in recent history. That said, when Anthropic eventually files its S-1, it won’t just be selling a vision of safe, powerful AI, but also proof that it can scale reliably. How it handles the gap between surging demand and strained infrastructure between now and its filing may matter as much to prospective shareholders as any benchmark result.

Featured Image: An AI-generated image depicting an Anthropic data center struggling to deal with increasing demand. Credit: ChatGPT