Enjoying this bite?
Sign in (free) to track this channel, unlock new bites the moment they drop, and search every summary we've ever made.
Source: YouTube. Summary is AI-generated from the video's captions and may contain errors. It does not represent the views of TubeBite, the creator, or YouTube. Watch the original before relying on anything important.
SUMMARY
Steve Eisman analyzes Anthropic’s newly released S-1, highlighting a significant slowdown in revenue growth and rising concerns over future compute costs. The episode also reviews major earnings reports, the AI capex cycle, and challenges facing both AI infrastructure and telecommunications sectors.
MAIN POINTS
- Anthropic’s S-1 discloses major future compute and equipment commitments totaling $800 billion, raising questions about sustainability.
- Anthropic’s revenue growth slowed sharply in Q3 as token maxing ended, and the company cut token pricing to spur demand.
- Microsoft, Google, and other tech giants reported earnings that triggered market concerns about a potential slowdown in AI capex and the resurgence of the 'SaaS apocalypse' narrative.
- Intel and Amazon posted disappointing forecasts and kept capex budgets flat, intensifying doubts about continued AI infrastructure investment.
- CoreWeave faces rising debt costs and uncertain viability as a specialty finance-style AI data center operator amid tightening fixed income markets.
- Power supply constraints, not capital or hardware, are emerging as the main bottleneck for new AI data center construction according to industry insiders.
- The 'Telepocalypse' narrative grows as Starlink’s ambitions to become a fourth U.S. mobile provider challenge incumbent telecoms, though technical and economic hurdles remain.
DETAILED ANALYSIS
Anthropic’s recently released S-1 filing provides a detailed look into the company’s financial trajectory and operational commitments, revealing both strengths and vulnerabilities. For the first half of 2026, Anthropic reported that 25% of its revenue came from its top five customers, indicating a relatively diversified client base. The company carries $6.2 billion in debt, a manageable figure in the context of its sector.
Notably, its adjusted operating margin in Q2 2026 was a robust 19%, but this period coincided with the phenomenon of 'token maxing,' where enterprises encouraged widespread AI usage without regard to cost. The S-1 is notably silent on the sustainability of these margins post-token maxing.
A critical concern arises from Anthropic’s future commitments for compute and equipment, which total approximately $800 billion across major technology partners such as Microsoft, Google, Amazon, SpaceX, AMD, and others. These obligations suggest that, beginning in 2027, the company’s compute costs could rise dramatically, leaving little room for operational error. Analysis of quarterly revenue trends shows that while Anthropic experienced explosive sequential revenue growth of 135% and 147% in Q1 and Q2 2026 respectively, this growth decelerated sharply to 45% in Q3.
The efficiency ratio, defined as revenue divided by compute expense, also declined from 2.08 in Q2 to 1.70 in Q3, reflecting rising costs and slowing growth. In response, Anthropic reduced token pricing by up to 50% in October to stimulate demand, but it remains uncertain if increased usage will compensate for lower prices. The company’s strategy may involve embedding its products deeply within enterprises before raising prices, but the competitive landscape and variable data center capacity complicate this approach.
The broader technology sector mirrored Anthropic’s challenges during a turbulent earnings week. Meta announced over $800 billion in AI-related commitments and subsequently raised $95 billion in equity, with Nvidia purchasing $3 billion as a show of support. Microsoft’s earnings were strong, with Azure revenue growing 48%, but its capex of $45 billion fell short of expectations, leading to a sell-off in related AI infrastructure stocks.
Google’s results were mixed: while total revenue increased 25%, earnings per share declined and free cash flow was negative. Most concerning was the slowdown in Google’s search revenue growth to 12%, fueling fears that capital-intensive AI search models may be eroding the company’s traditional, high-margin business. ServiceNow’s muted guidance revived concerns about a potential 'SaaS apocalypse,' as software sector growth projections decelerated.
Other sectors reflected similar uncertainty. Boeing continued its protracted turnaround, with defense unit growth offset by weakness in commercial aircraft. Visa and Mastercard reported strong payment volumes, indicating robust consumer spending at the higher end of the market.
S&P and Moody’s, despite their duopoly status, saw stock declines due to renewed AI-related anxieties in the software sector. GE Vernova and Quanta, both involved in power generation and infrastructure, posted strong results, benefiting from the ongoing demand for AI data centers. However, Caterpillar’s modest earnings beat was insufficient to reassure investors amid concerns about the sustainability of AI infrastructure construction, resulting in a sharp stock decline.
Intel and Amazon’s earnings further dampened market sentiment. Intel, despite a 126% year-over-year increase in earnings per share, issued a weak forecast, projecting a persistent PC slump and ongoing supply chain issues in its data center business. Amazon’s earnings and revenue both missed estimates, and for the first time, the company kept its capex budget flat at $220 billion.
This departure from the trend of continual capex increases among hyperscalers raised questions about whether the AI capex cycle has peaked, with broad implications for the entire hardware and data center ecosystem.
The episode also addressed the precarious position of CoreWeave, an AI data center operator whose debt costs have surged from 9% to an effective yield of 15% due to declining bond prices and increased default risk. With total obligations exceeding $50 billion, CoreWeave’s business model resembles that of specialty finance companies, which are vulnerable to shifts in fixed income market sentiment. The company faces stiff competition from hyperscalers with far greater financial resources, and the viability of building new data centers at such high borrowing costs is questionable.
A key operational constraint for AI infrastructure emerged from industry feedback: power supply, not capital or hardware, is now the primary bottleneck in data center construction. Electrical contractors report that delays in transmission lines and substations are pushing project timelines out by years, with labor shortages and site planning challenges compounding the issue. Microsoft and Google are seen as better positioned due to superior planning, while companies like Amazon, Meta, CoreWeave, and Oracle may face greater delays in securing power for new facilities.
Finally, the 'Telepocalypse' narrative is gaining traction as Starlink, led by Elon Musk, seeks to become the fourth major U.S. mobile provider. While the technical and economic challenges of deploying a large constellation of low-earth orbit satellites are significant, the mere prospect has put pressure on incumbent telecom stocks. Industry experts suggest that terrestrial cell towers remain more cost-effective for most consumers, and the primary benefit of satellite-based service would be for remote areas.
Nonetheless, Musk’s history of pursuing ambitious, capital-intensive projects means the threat to established telecoms cannot be dismissed outright.
LINKS
- The Real Eisman Playbook Premium subscription page.
- Steve Eisman’s Financial Literacy Masterclass video.
- Steve Eisman’s Masterclass on the 2008 Financial Crisis.
- YouTube subscription link for The Real Eisman Playbook.
- Steve Eisman’s Linktree for social media and updates.