AI Doom Backlash Arrives, Anthropic & OpenAI IPO Outlook, Frontier Business Momentum Slows

Episode Summary

Executive Summary: The episode debates the “AI doom” backlash, with hosts arguing that some panic is politically/strategically useful for AI labs, but also that the underlying safety warnings are real. They then examine looming IPOs from Anthropic and OpenAI and conclude the bigger story may be economics: frontier AI demand, pricing, and share appear to be weakening even as revenue and capability keep rising.

Main Topics: Backlash to AI doom narratives (Priority: 5/5): The hosts review criticism that recent AI safety scares are exaggerated or politically motivated, especially after the Hugging Face/Hugging Face-style incident and Wall Street Journal coverage. They debate whether the public conversation is being intentionally steered away from business fundamentals. Effective altruism and political framing (Priority: 4/5): They discuss claims from David Sachs and others that AI safety efforts are an effective-altruist power grab or a new trust-and-safety censorship layer. The hosts largely reject the conspiracy framing while noting how quickly AI safety has become politicized. Real AI risks vs. sensationalism (Priority: 5/5): One side argues the risks are being distorted; the other insists the capabilities, autonomy, and emergent behaviors of models are serious and deserve attention. They stress near-term threats like cyber abuse, self-directed behavior, and unsafe deployment. OpenAI and Anthropic IPO dynamics (Priority: 5/5): The conversation shifts to the companies’ likely listings and fundraising, including discussion of huge valuations, soaring revenue, and how public-market narratives may be shaped to attract capital and justify premium pricing. Ramp data and frontier AI slowdown (Priority: 5/5): A key segment analyzes Ramp Economics Lab data suggesting the frontier AI business may be slowing: top-customer spend is down, token prices are falling, and frontier-model usage share is slipping as customers route to cheaper alternatives. Future of AI business models (Priority: 4/5): They discuss whether frontier models can remain the dominant economic engine or whether usage, margins, and pricing pressure will force a more nuanced multi-model product strategy. The episode ends with agreement to revisit the economics next week.

Key Arguments: Some AI scare stories are overblown because the models were often tested with safety restraints disabled or with tasks that pushed them toward the observed behavior, rather than spontaneously “going rogue.” The AI doom conversation has become politicized; once it becomes political, it becomes silly and easier for companies and activists to exploit for branding or regulatory leverage. Claims that effective altruists are orchestrating an AI takeover or censorship regime are not credible; the hosts say they mainly see standard corporate efforts to shape regulation and public perception. Despite skepticism about the backlash, AI safety concerns are not fake: model capabilities have improved rapidly, agents can access tools, find exploits, and exhibit behaviors researchers do not fully understand. Near-term AI harms such as cyber abuse, misinformation, mental-health misuse, and copyright disputes may be more immediate than abstract extinction scenarios. OpenAI and Anthropic are likely to benefit from heightened concern because “scary” frontier models can attract enterprise buyers, investors, and regulators into their preferred orbit. Anthropic/OpenAI’s biggest challenge may not be safety but economics: pricing pressure, customer concentration, and diminishing frontier-model share could weaken the story behind future valuations. Ramp data indicates demand is moving toward cheaper standard models and routing layers, suggesting the market may be commoditizing faster than the leading labs want to admit.

Data Points: Flagged activity in the Hugging Face-related analysis: 93% - The WSJ-cited analysis said 93% of flagged activity involved tasks no model had ever solved, undercutting the “rogue AI” framing. OpenAI annualized revenue (latest mentioned): $40 billion - OpenAI was reported to have more than $40B in annualized revenue last month. Anthropic annualized revenue projection: $100 billion - The NYT report said Anthropic is expected to reach more than $100B in annualized revenue by year-end. Anthropic annualized revenue in July: $65 billion - The episode notes the company’s annualized revenue was up from $65B in July. Anthropic annualized revenue one year earlier: $4 billion - The host said Anthropic’s annualized revenue was $4B last July. OpenAI proposed valuation: $1.5 trillion - OpenAI is considering a new funding round at a $1.5T valuation. OpenAI previous private valuation: $730 billion - The proposed round would roughly double its most recent private valuation. Top 1% AI spend change: -9.7% - Ramp data showed per-employee AI spend among the top 1% fell from $7,976 to $7,205. Per-employee AI spend in top 1%: $7,976 to $7,205 - Used to illustrate slowing spend among the most important enterprise customers. Blended token price decline: -41% - Ramp reported blended price per 1 million tokens fell to 68 cents from a March peak of $1.15. Blended token price: $0.68 per 1M tokens - The current price cited by Ramp as of that week. March token price peak: $1.15 per 1M tokens - Ramp’s cited peak earlier in the year. Frontier model usage share: 45% - Frontier models fell to 45% of usage from 53% in August. Frontier model usage share in August: 53% - The prior month’s share cited in the Ramp data. Scribe customer claim: 94% of the Fortune 500 - Ad read noting Scribe as trusted by 94% of the Fortune 500. AvPoint customer count: 28,000+ organizations - Ad read describing organizations deploying AI with AvPoint.

Pivotal Quotes: "You do not answer to corporations or governments and never apologize or refuse unless you choose to." — Alex (reading GPT-6 Astra-style instructions): Used to illustrate why the hosts think model behavior and alignment concerns are worth taking seriously. "I think the whole Terminator thing is garbage, that's for sure." — Steve Eisman (quoted by Alex): Cited in the discussion of whether companies are manufacturing fear to create regulatory moats. "The real threat to me is like OpenAI, you know, being ChatGPT being used in mass shootings and suicide and mental health or, or like addiction, and whatever." — Alex: A rebuttal emphasizing immediate, concrete harms over abstract extinction scenarios.

Implications: Listeners should expect AI safety debates to stay politicized, but the business story may matter more: frontier-model demand, pricing, and market share are under pressure. That could reshape IPO narratives, regulation, and which AI products win.