Anthropic's Revenue Skyrockets Past Salesforce

In a discussion on "The Brainstorm" podcast, ARK Invest analysts Frank Downing and Nick Grous delved into the explosive growth and market positioning of leading AI companies, particularly Anthropic and OpenAI, and the emerging role of xAI's Grok. The conversation highlighted the staggering revenue growth of Anthropic, a company that has reportedly added more run rate revenue in less than a year than Salesforce’s entire $45 billion business.

Anthropic's Unprecedented Revenue Growth

Downing revealed that Anthropic is rumored to have achieved an annual run rate revenue of $70 to $80 billion, a dramatic leap from its $9 billion run rate at the beginning of the year. This near-10x growth, achieved before the year is even three-quarters complete, places Anthropic among the fastest-growing software companies in history. Grous emphasized that Anthropic’s singular focus on the enterprise opportunity and delivering knowledge work productivity is a key driver, as this sector represents a larger near-term market than consumer applications, given the global expenditure on knowledge worker wages.

AI Companies Challenging Tech Giants

The scale of growth for Anthropic and its peer OpenAI is so significant that their combined revenue is now approaching that of Microsoft’s productivity and business processes division. Downing noted that OpenAI has internally reported crossing $40 billion in run rate revenue. He projected that if these growth rates continue, both companies could surpass Microsoft’s second-largest business segment in revenue by year-end. This places these AI-native companies on a similar scale to established software giants, fundamentally altering the competitive landscape.

The Rise of AI as a Productivity Default

Both analysts agreed that AI is rapidly becoming the new general-purpose productivity tool. Grous drew a parallel to the early days of the internet adoption, suggesting that AI is in a similarly nascent but rapidly expanding phase. He highlighted personal anecdotes where employees are willing to pay for AI subscriptions out-of-pocket due to their significant impact on job performance, signaling a strong demand signal. While acknowledging that traditional tools like Microsoft Office and Google Workspace are unlikely to disappear soon, Grous asserted that the incremental spending is shifting towards these new AI-powered productivity tools.

The "Pareto Frontier" in AI Models

The discussion then shifted to the concept of the "Pareto frontier" in AI model development. Downing explained that this refers to identifying the models that offer the best intelligence for a given cost, rather than simply having the smartest model overall. He pointed out that 14 out of the 16 models currently on the Pareto frontier are from private companies, with OpenAI leading in the number of models represented, followed by SpaceX (xAI), Meta, and Anthropic. Notably, Google's DeepMind was mentioned as being currently off the frontier, despite having significant resources.

xAI's Surprising Leap and Google's Challenge

Downing expressed admiration for SpaceX's xAI division, noting its rapid recovery and emergence onto the Pareto frontier with its Grok-1.5 model. He attributed this success to strategic hires, new data acquisition, and a focus on the three key ingredients for AI R&D: compute, talent, and data. The conversation also touched upon Google's current position, questioning why the company, with its vast resources, has not yet placed models on the Pareto frontier, despite having efficient models like their 3.7 flash model.

Grokbot and the Future of Virtual Workers

The introduction of Grokbot, Anthropic's co-working agent, was discussed as a significant development. Downing described Grokbot as a general-purpose tool that provides each AI agent with its own virtual machine, granting access to a wide array of tools to accomplish tasks. He likened this to a "Tesla Optimus" move for digital knowledge work, emphasizing the user-friendly interface and the potential for teams of agents to collaborate. The potential synergy with Tesla's supercharger locations for hosting these virtual machines was also highlighted.

IPO Rumors and Market Valuations

The conversation touched upon rumors of a potential multi-trillion dollar IPO for Anthropic later this year. Downing analyzed the valuation, suggesting that a $2 trillion IPO on a $100-$120 billion run rate revenue would represent a 20x multiple, which is considered lower than some comparable public software companies. He also discussed the market's scrutiny on profitability and token subsidization, contrasting it with rumors that Anthropic might already be profitable or on a path to it, which would significantly bolster its valuation narrative.

Forecasting Future Growth and Adoption Barriers

The analysts debated the ambitious revenue forecasts for these AI companies, with Grous citing a $250 billion revenue projection for Anthropic in 2027 as more realistic than a higher $400-$500 billion estimate. They acknowledged that continued 10x growth is possible but contingent on securing sufficient compute power and overcoming adoption barriers as customers adjust to higher spending on AI tools. The eventual realization of productivity gains will be crucial for sustained growth.

The Critical Role of Compute, Talent, and Data

A recurring theme was the foundational importance of compute, talent, and data for AI development. The success of companies like SpaceX and Meta in rapidly advancing their AI capabilities was directly linked to their ability to procure compute, attract top talent, and leverage vast datasets. This underscores the strategic imperative for AI companies to secure these essential resources to remain competitive.

Google's Path Back to the Frontier

Regarding Google's position, Downing expressed skepticism about its ability to regain a leading spot on the AI frontier, citing potential talent drain as a key concern. Despite Google's significant compute and data advantages, he estimated only a 40% chance of a comeback, contrasting it with Meta's successful resurgence. He emphasized that the current leaders, particularly OpenAI and Anthropic, benefit from a virtuous cycle where superior models drive research and further improvements.