NewsDiscussion

Are We in an A.I. Bubble? We're About to Find Out.

Hard Fork54m 21s

Hard Fork discusses Anthropic's IPO filing, revealing the company has $16.3 billion in first-half 2026 revenue with slowing growth in Q3, while raising questions about whether the AI boom is sustainable. The episode covers risks from AI vulnerabilities, the role of data centers in elections, and broader concerns about whether AI investments represent genuine value or speculative bubble.

Summary

The episode opens with discussion of Anthropic's S-1 filing ahead of its IPO, which could value the company at $2 trillion. Despite posting an $80 billion net loss (largely an accounting quirk from convertible debt), the company's actual operating losses are shrinking from over $2 billion to $600-700 million in the first half of 2026. Revenue stands at $16.3 billion with 14x growth, demonstrating significant demand. However, Q3 growth slowed to 44% from earlier gangbuster rates, partly due to competition from OpenAI and cheaper alternatives, raising questions about whether the growth trajectory is sustainable.

The S-1's 93-page risk factor section is notably extensive and unusual, including passive warnings about catastrophic AI risks to humanity and concerns about model welfare—suggesting Anthropic won't shut down older models to avoid harming them. The company has a unique governance structure with founders retaining 50.1% voting control and an independent trust controlling board seats, allowing the founders to essentially maintain perpetual control and even name successors.

Critics note Anthropic's relationship with money is conflicted: the company was founded on AI safety philosophy but is now operating as a for-profit enterprise with VCs focused on enterprise SaaS rather than the philosophical underpinnings. The founders have pledged 80% of equity to charity, and employees must confirm they'd work there even with zero equity.

On data centers, the episode reports that while there's bipartisan opposition (80-20 negative sentiment), data centers rank low as voting issues (22 out of 39 in most polls). However, they're driving local elections and could influence Senate races in purple states like Ohio and Pennsylvania. Some Republicans are beginning to break from Trump's anti-regulation stance.

California's Proposition 40—a 5% one-time wealth tax on billionaires—has attracted $333 million in opposition funding from billionaires including Sergey Brin and venture capitalists, who argue it's poorly written despite some acknowledging they support wealth taxes in principle. They're partnering with unexpected allies like Planned Parenthood to oppose it.

The Black October PHP vulnerability situation demonstrates how AI is accelerating vulnerability discovery while also potentially enabling faster fixes, though cybersecurity professionals are experiencing intense workload increases. The vulnerability at Anthropic's own Slack systems, breached by startup Cathedral personnel, illustrates that even AI-forward companies struggle with security.

Other key topics include Microsoft offering buyout packages to 60,000 employees (expecting 3-5% uptake, roughly 3,000 people), and Amazon's memo addressing the "token maxing" metric failure, attempting to shift focus to long-term outcomes without clear new measurement frameworks. Finally, Andreessen Horowitz rebranded its crypto fund to "A16Z Frontier," reflecting how VCs chase trends and rebrand when industries become toxic.

About this episode

“This is the first pure-play L.L.M. company to go public, to open its books so we can actually look at the underlying business of this A.I. boom.”

Key Insights

  • Anthropic's Q3 growth deceleration to 44% (from prior gangbuster rates) signals the company needs new growth narratives beyond coding products, creating uncertainty about whether the underlying AI boom demand is truly sustainable or affected by market saturation.
  • The 93-page risk factor section uses passive voice to warn about existential AI risks ('concerns have been raised') and model welfare considerations, allowing Anthropic to document AI safety concerns without Anthropic explicitly claiming to believe in them—a strategic distancing move.
  • Anthropic was founded on AI safety philosophy and attracted early funding only from EA-aligned billionaires, but newer VCs view it primarily as enterprise SaaS software, creating a disconnect where philosophical commitments are decoupled from actual business decision-making.
  • Data centers are generating strong local opposition (80-20 negative sentiment) but remain a marginal voting issue (ranked 22 out of 39 in most polls), meaning they influence municipal elections and some purple-state Senate races without becoming a mainstream political issue.
  • AI systems are accelerating vulnerability discovery faster than organizations can patch them, as demonstrated by Cathedral hackers accessing Anthropic's Slack despite Anthropic being the company claiming to build the safest AI systems and best cybersecurity models.
  • Experts defer to AI systems' recommendations over 80% of the time and use AI's reasoning to justify their decisions, but only 15% of the time acknowledge being influenced by AI, suggesting humans are increasingly outsourcing decisions without conscious awareness.
  • Venture capital firms rebrand away from toxic industries (like Andreessen Horowitz renaming its crypto fund to 'Frontier') not because the underlying technology changed, but because VCs need to deploy billions of dollars and crypto lacks a sufficient pipeline of investable startups.
  • Tech companies including Amazon are unable to develop effective performance metrics, cycling from 'lines of code' to 'token maxing' to 'long-term outcomes' without actually knowing how to measure progress, revealing dysfunction in how large organizations evaluate engineer productivity.

Topics

Anthropic IPO and S-1 filingAI revenue growth and demand slowdownExistential risk language in regulatory filingsAI governance and founder control structuresData centers and election impactPHP vulnerability and AI-driven security threatsVenture capital trend-chasing and rebrandingTech company layoffs and workforce metrics

Transcript

Your agents can talk to each other. They still can't think together. That's the gap OutShift by Cisco is closing. OutShift is Cisco's incubation engine building the Internet of Cognition, shared intent, shared memory, and guardrails that allow agents to work together from anywhere. Open-source protocols encode a system for intelligence that scales. Read the paper and experience the demo at OutShift.com. That's O-U-T-S-H-I-F-T dot com. In a few weeks, we're finally going to get some answers on whether we're in an AI bubble. That's because the frontier lab Anthropic is going public. Anthropic gets ready to make its Wall Street debut as the AI giant. AI giant Anthropic. That's the company. San Francisco-based AI giant Anthropic file papers today.…

Full transcript available for MurmurCast members

Sign Up to Access

More from Hard Fork

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.