AI Bubble Fears Recede as Claude Code Drives Revenue Surge at Anthropic

Key facts

• Anthropic's annual run rate jumped from $14 billion to $30 billion in two months.
• Claude Code enables autonomous AI agents to complete programming tasks in minutes that previously took days or weeks.
• Developers using the latest AI tools completed tasks almost 20% faster than those without, per a re-run experiment.
• Over half of American businesses now have a paid subscription to at least one AI tool, up from a quarter in early 2025.
• Goldman Sachs found many companies overspending on AI tools by 'orders of magnitude,' with some at 10% of engineering labor costs.
• Anthropic expects profitability in 2028; OpenAI in 2030.
• Meta announced layoffs of 10% of its workforce, with Mark Zuckerberg citing AI for enabling smaller teams.

From Bubble Fears to Breakneck Growth

Six months ago, the artificial intelligence sector was widely viewed as a speculative bubble, with companies borrowing hundreds of billions of dollars to build data centers while lacking a clear path to profitability. Experts drew parallels to the railroad bubble of the 1800s and the dot-com bubble of the 1990s. Even OpenAI CEO Sam Altman publicly acknowledged that investors might be 'overexcited' about AI. Today, that narrative has been upended. Anthropic, the company behind the AI model Claude, has become what may be the fastest-growing business in the history of capitalism. Its revenue is increasing faster than Zoom's during the pandemic, Google's in the early 2000s, and even Standard Oil's during the Gilded Age. If current growth continues, Anthropic could be taking in more money than any other company in the world by early next year.

Claude Code: The Catalyst

The turnaround can be traced to a single product update: Claude Code, released by Anthropic in November. The tool allows a team of autonomous AI agents to take over a computer and complete programming tasks in minutes or hours that would have taken humans days or weeks, often requiring few or no human changes. Ethan Mollick, co-director of the Generative AI Lab at the University of Pennsylvania, called it 'a step change,' marking the transition from chatbots that 'mostly just say things' to agents that 'can actually do things.' Other companies have since released similar coding tools, including OpenAI's Codex and Anysphere's Cursor, which are considered nearly as impressive. The implications are enormous for any industry reliant on software.

Real-World Productivity Gains

Anecdotal evidence of productivity gains is mounting. Jordan Nanos, a member of the technical staff at semiconductor-research firm SemiAnalysis, reported that his small team now produces four times as much software as it did last year with the same number of employees. Tim Fist, director of emerging-technology policy at the Institute for Progress, said it 'feels sort of ridiculous' to pursue a computer-science Ph.D. because 'Claude can basically do 90 percent of it.' Meta recently announced layoffs of 10 percent of its workforce; CEO Mark Zuckerberg told investors that AI means 'projects that used to require big teams' can 'now be accomplished by a single very talented person.' Academic research supports these claims. Last year, the think tank Model Evaluation & Threat Research found that developers using AI completed tasks 20 percent slower, partly due to time spent correcting AI output. But when the same researchers re-ran the experiment with the latest tools, developers completed tasks almost 20 percent faster. That figure is likely an underestimate, as some power users refused to participate in the second experiment because they had become so reliant on AI.

Spending Surge and Infrastructure Strain

With clear productivity benefits, companies are spending heavily on AI. The percentage of American businesses with a paid subscription to at least one AI tool has risen from about a quarter at the start of 2025 to over half today. Researchers at Goldman Sachs, who interviewed 40 software companies in mid-April, found that many were 'overrunning their initial budgets' for AI tools 'by orders of magnitude,' with some spending as much as 10 percent of total engineering labor costs. 'The speed at which we're seeing companies adapting these tools is actually quite surprising,' said Gabriela Borges, a software analyst at Goldman Sachs. Demand is now outstripping supply. Anthropic has been forced to limit customers' use of its coding tools during peak hours, and OpenAI scrapped its video-generation app to free up computing power. Semiconductors are in such high demand that even Nvidia's fourth-best AI chip, released in 2022, costs more today than three years ago. CoreWeave, a 'neo-cloud' company renting chips and data-center space, saw annual revenue grow 168 percent last year; chipmaker Micron's revenue nearly tripled.

Accelerating Model Capabilities

The AI models driving this revenue growth continue to improve. In early April, Anthropic announced Mythos, a new model so powerful that the company did not release it to the public. Mythos has set new benchmarks across complex coding tasks and graduate-level problems, and it discovered cybersecurity vulnerabilities that had gone undetected by humans for decades. OpenAI's GPT-5.5 is not far behind. 'On basically every indicator we have, we were already seeing a big acceleration in the pace of AI progress,' said Jean-Stanislas Denain, a senior researcher at Epoch AI. 'And that was before Mythos.'

Skeptics Warn of a Narrow Boom

Despite the optimism, some analysts argue that the AI sector may still be in a speculative frenzy. Flagship companies like OpenAI and Anthropic are not yet profitable; they spend all revenue and more on developing their next models. Anthropic expects profitability in 2028, OpenAI in 2030. The pessimistic case holds that software development is uniquely suited for AI automation, with abundant training data, limited possible outcomes, and objective evaluation. Other white-collar fields, such as legal or marketing, lack these characteristics, which could limit demand. 'Even if white-collar workers use these AI tools for some things, it won't look like anything close to what we're seeing right now for coders,' said Paul Kedrosky, a managing partner at SK Ventures and research fellow at MIT, who is a prominent proponent of the bubble thesis. AI companies are investing heavily in chips and infrastructure in anticipation of continued demand. But if the boom is limited to coding, new data centers may lack enough customers, leading to huge losses. Kedrosky compared the situation to the real-estate market in 2006-2007: 'Market hype leads to more demand. More demand makes you think you need more supply. Before you know it, you've built more homes than anyone can actually afford. And eventually it all falls apart.'

The bottom line

• Anthropic's revenue growth, driven by Claude Code, has made it the fastest-growing company in history, but profitability remains years away.
• AI coding tools have delivered measurable productivity gains, with developers completing tasks 20% faster in recent experiments.
• Demand for AI infrastructure now outpaces supply, leading to rationing and rising chip prices.
• The sustainability of the boom hinges on whether AI's utility extends beyond software development to other white-collar sectors.
• Skeptics warn that overinvestment in data centers could lead to a painful bust if demand fails to materialize broadly.
• AI model capabilities continue to accelerate, with Anthropic's unreleased Mythos setting new benchmarks.