In this episode of The Register’s Kettle podcast, host Brandon Vigliarolo, along with systems editor Tobias Mann and senior reporter Tom Claburn, discuss Anthropic’s recently filed IPO prospectus, which spans over 600 pages. They highlight that while the document is filled with ambitious promises about the future of AI, it leaves many tough questions unanswered. Tobias points out that the financials show significant operating losses—around $700 million—though some accounting adjustments inflate the total losses to over $80 billion. Despite these losses, Anthropic’s revenues have, for the first time, outpaced their compute expenses by a 2:1 ratio, which has given some investors cautious optimism.
However, the panelists express concern over Anthropic’s massive compute commitments, which total upwards of $800 billion over the next two decades. These commitments pose a significant risk because profitability depends on the company generating enough revenue to cover these escalating costs. The discussion touches on the uncertainty of whether inference costs will continue to decrease, which is critical for Anthropic’s business model. They also note that many of Anthropic’s biggest customers are also competitors and cloud providers, which could create conflicts of interest and operational challenges similar to those faced by mobile virtual network operators.
The conversation also explores the broader AI industry landscape, noting that many companies are moving towards more cost-effective and specialized AI models rather than relying solely on large, expensive frontier models. This trend puts additional pressure on companies like Anthropic, which depend heavily on large-scale compute resources. The panelists mention examples like Cadence Systems using open-weight models to reduce costs and highlight the increasing restrictions on open-weight model licenses as companies try to protect their competitive advantages. They also discuss the potential shakeout in the AI sector due to tightening financing and economic pressures.
When comparing Anthropic to OpenAI, the hosts suggest that OpenAI might face similar challenges when it eventually goes public, but its different business model—such as exploring AI advertising revenue—could provide some insulation. OpenAI has delayed its IPO to better assess these risks, whereas Anthropic has moved forward, possibly to capitalize on being first to market. The panelists speculate that Wall Street scrutiny will intensify for AI companies going public, especially as the hype around AI matures and investors demand clearer paths to profitability.
In conclusion, the hosts agree that while AI represents a transformative economic opportunity, the current valuations and business models of companies like Anthropic carry significant risks. They caution potential investors against buying into the IPO hype without considering the long-term challenges, comparing the situation to the dot-com boom where initial valuations took years to be realized. Despite the uncertainties, the panelists anticipate that AI will remain a dominant topic in technology discussions for years to come, with ongoing developments and market shifts to watch closely.
Useful Links
- Mistral AI Large Language Models and Licensing — Explains licensing strategies impacting AI model usage and competition, relevant to the video discussion.