Daily Newsletter Digest

Anthropic's $430B compute bet reveals massive AI infrastructure forecasting risk

October 29th 2026

Today's News

Google Cloud revenue surges 103% YoY to $30.7B; Gemini 4 Carbon tops benchmarks at 72% lower cost than Claude, but Alphabet stock drops 9% as investors question AI capex ROI

Techmeme

Google's cloud business is accelerating sharply with token throughput now at 32B tokens/min (up 40% QoQ), yet the market remains skeptical of Alphabet's $195B–$205B annual capex forecast. The company's new Gemini 4 Carbon model ranks #1 on Artificial Analysis Intelligence Index and achieves 72% lower cost than Claude Opus 5.5, signaling competitive progress in frontier model capabilities and cost efficiency that could reshape pricing dynamics across the AI stack. Despite strong cloud revenue growth and demonstrated model competitiveness, investors are concerned about whether the company's massive infrastructure investment will generate sufficient returns, particularly as YouTube ad revenue growth slows to 9% YoY.

Anthropic commits $430B+ through 2036 in multiyear compute and equipment deals, betting on models not yet invented

Liberty’s Highlights

Anthropic's S-1 filing reveals $269B in cloud compute commitments and $161B in equipment leases with major providers through 2036, while 80%+ of Q3 2026 revenue came from models released in the prior six months. The massive forward commitment signals confidence in sustained demand growth and model velocity, though it represents a significant financial risk if adoption or performance trajectories diverge from internal projections.

Google monetizing AI competitors through hardware and platform lock-in: OpenAI to buy TPUs outright, Oracle to offer them in datacenters from 2027

Liberty’s Highlights

Google is selling custom TPU chips to both OpenAI and Oracle, with OpenAI planning to purchase chips outright while Oracle will offer them in its datacenters starting 2027. Anthropic's S-1 filing reveals $114.5B+ in Google Cloud commitments through 2033, demonstrating Google's ability to monetize competitors' model choices through both hardware sales and platform lock-in—a strategic lever that could generate sustained revenue even as the company competes directly in frontier models.

Microsoft offers 27% of workforce voluntary exits while posting record margins, signaling confidence in AI-driven productivity gains ahead

Azeem Azhar, Exponential View

Microsoft offered voluntary exits to approximately 60,000 employees (27% of its 223,000-person workforce) across the US, UK, Canada, and India—a move that appears forward-looking rather than reactive to past bloat. Revenue per employee has grown two-thirds since 2022 to $1.5M, and revenue growth has outpaced headcount by 20% in the latest year, the widest gap on record. Rather than addressing Covid-era hiring (which was absorbed years ago), Microsoft appears to be pre-emptively rightsizing ahead of efficiency gains it expects from AI and agentic systems, betting it can operate significantly leaner while maintaining or accelerating revenue growth.

OpenAI deploys thousands of GPT-Rosalind agents to analyze whole-brain CRISPR atlas in 32 minutes, replacing months of human literature review

Noahpinion

OpenAI deployed thousands of GPT-Rosalind agents to examine a whole-brain CRISPR atlas and identify new leads for neurodegenerative diseases like Alzheimer's and Parkinson's, completing work that would have taken human researchers months to a year. The deployment demonstrates AI's ability to instantly synthesize scientific literature and make novel conceptual connections across fields, signaling early progress in using AI agents to eliminate theory and literature bottlenecks that have historically constrained scientific productivity.

Anthropic reports 70% shorter CAR-T design cycles and 35,000+ antibody designs from Claude-powered drug discovery

Noahpinion

Early results from six pharmaceutical collaborations show Claude enabling dramatic acceleration in drug development timelines, including 10x tighter antibody binding and 89% less batch-record review time. These preliminary results from Claude-powered drug discovery signal that AI is beginning to eliminate the theory and literature bottlenecks that have constrained scientific productivity, with potential to reshape timelines across the pharmaceutical industry.

Worth Reading

Anthropic's $430B Compute Bet, Seinfeld's Accidental $7B Startup, The Endless Y2K, Google's TPU Offensive

Liberty's Highlights

Liberty's Highlights published an essay analyzing Anthropic's S-1 filing and broader AI infrastructure economics. Main takeaways:

  • Over 80% of Anthropic's Q3 2026 revenue came from Claude models released in prior six months, while company committed $430B+ in multiyear compute deals extending to 2036, creating massive forecasting risk if demand softens.
  • Anthropic spent $8.95B on compute/infrastructure in H1 2026 but only $2.38B in cost of revenue, implying 73% of compute spending capitalized as R&D, enabling 85% gross margins despite enormous infrastructure spend.
  • Google is unbundling Nvidia's integrated stack by partnering with Broadcom and Marvell on specialized chips (inference, networking, storage, memory) while maintaining TPU architecture control, reducing dependence on Nvidia while selling TPUs to competitors like OpenAI.
  • OpenAI's Jalapeño custom chip reportedly delivers 1.5-1.9x better throughput-per-watt than Nvidia GB200/GB300 with lower latency, suggesting first-generation ASIC performance better than expected and meaningful leverage in Nvidia negotiations.
Read more

2030 Revenue Hurdle: Differ from Consensus

Investment Brief AI / Exponential View

Exponential View in Investment Brief AI / Exponential View published a report titled "2030 Revenue Hurdle: Differ from Consensus". Main takeaways:

  • Consensus estimates require $2.1-3.0 trillion in annual AI infrastructure revenue by 2030 (107% CAGR), far exceeding historical tech boom growth rates of 20-29% annually
  • Exponential View's model estimates $0.77-1.3 trillion required revenue (52% CAGR), based on $480bn capital in service, 5.8-year compute depreciation, 15% pre-tax ROIC, and $0.90bn annual OpEx per gigawatt
  • Van Nieuwerburgh assumes $1.25 trillion US CapEx in 2025-26 alone (vs. $500-700bn global estimates from Moody's/Morgan Stanley), inflating his $3.5 trillion 2032 revenue hurdle
  • Current AI revenue covers 188% of in-service capital costs; infrastructure revenue alone covers 135% against $66bn depreciation, $13bn OpEx, $72bn ROIC on $480bn deployed capital
  • CoreWeave's unsecured notes now yield ~10 points over Treasuries (up from 5.0 points in July 2025), stock down 67% from $187 peak, signaling funding stress despite profitable customer contracts
Read more

Articles Read