Summaries > AI > Anthropic > Anthropicâ€™s Growth is Slowing: Is AI Capex Next? | The Weekly Wrap...

Anthropicâ€™S Growth Is Slowing: Is Ai Capex Next? | The Weekly Wrap

https://www.youtube.com/watch?v=z-pClpi6eTY

TLDR AI’s explosive growth is running into questions about profitability, massive capital commitments, and power constraints, while disappointing tech results and rising borrowing costs are fueling doubts about whether the boom can last. Anthropic’s filing shows strong margins but slowing growth and huge future obligations; results from Google, Amazon, Intel, and ServiceNow raised further concerns, while CoreWeave faces expensive financing and data centers struggle to secure electricity. The episode also covers France’s limited threat to global debt markets, mixed company earnings, and the potential disruption of U.S. telecom by Starlink.

Key Insights

1. Separate AI revenue growth from durable profitability

When evaluating AI companies, look beyond headline revenue growth to operating margins, cash flow, and the cost of computing. Anthropic reported a 19% adjusted operating margin in Q2 2026, but its large future compute commitments and falling revenue relative to compute expense raise questions about whether those margins can last. Growth can also be temporarily boosted when customers use unusually large amounts of AI at subsidized prices, a practice described as “token maxing.” Track whether usage and revenue remain strong after pricing changes, such as Anthropic’s October token-price cuts of up to 50%.

2. Measure AI spending against returns and financing costs

Capital expenditure is not automatically a sign of healthy AI growth; compare the scale of spending with revenue, cash flow, and expected customer demand. Microsoft reported Azure growth of 48%, but its $45 billion in quarterly capital spending and finance leases came in below consensus, unsettling investors in data-center suppliers. Amazon’s decision to keep its 2026 capital-expenditure budget at $220 billion also prompted questions about whether spending growth is peaking. For infrastructure companies that borrow heavily, check the effective cost of debt: CoreWeave’s bonds implied yields near 15%, creating a potential challenge to financing new data centers.

3. Treat electricity access as a core data-center constraint

Before assuming a data-center project can scale, assess whether it can obtain dependable power and grid connections on schedule. Transmission lines can take more than two years to build, while substations may require one to two years, leaving some projects at risk of delay or becoming stranded. These timelines can limit AI capacity even when companies have GPUs, funding, and customer demand. Evaluate operators by how well they planned site locations and secured electrical infrastructure, not just by their announced construction plans.

4. Distinguish company results from stock-market expectations

A company can report growth or beat estimates and still see its share price fall if investors expected more or worry about future performance. Boeing narrowed its loss but missed expectations, while Intel beat recent results yet issued a weaker outlook. T-Mobile added customers and increased earnings, but its stock remained flat amid concerns about Starlink entering the mobile market. To interpret earnings reactions, compare actual results and guidance with market expectations, and identify whether the concern is demand, margins, competition, or future investment.

5. Assess whether AI is a threat or an opportunity for each sector

AI’s impact varies by business model, so avoid treating every technology or infrastructure company as a direct beneficiary. Investors worried that AI could weaken software and search businesses, contributing to steep declines in ServiceNow and Google after concerns about guidance, free cash flow, and search growth. Meanwhile, GE Vernova and Quanta reported strong results linked to data-center and infrastructure demand. Ask whether AI is strengthening a company’s products, disrupting its existing revenue, or increasing its costs before drawing conclusions from broad AI-market trends.

6. Put national debt concerns in global market context

To judge whether a country’s debt poses a systemic risk, compare its scale with the size and liquidity of global markets. Steve Eisman argues that France’s roughly $4 trillion in sovereign debt is relatively small beside U.S. debt of about $40 trillion and BlackRock’s approximately $15 trillion under management. That comparison does not mean French debt is risk-free or that repayment obligations can be ignored. It does suggest that claims of an immediate global fixed-income crisis should be weighed against the broader market’s capacity to absorb the debt.

7. Separate market narratives from evidence and treat forecasts cautiously

Stories about an AI boom, a “SaaS apocalypse,” or a “tele-apocalypse” can move stocks before the underlying economics are settled. For example, satellite-to-phone service could face high costs and capacity limits compared with conventional cell towers, even if the possibility of a new competitor pressures telecom shares. Prediction-market odds and political polls are also estimates rather than certainties; the speaker cautioned that polls may undercount some voters. Use these narratives as prompts to investigate business models, costs, and evidence—not as substitutes for them.

Questions & Answers

What does Anthropic’s S-1 reveal about its recent performance and financial risks?

Anthropic reported a 19% adjusted operating margin in Q2 2026 and manageable customer concentration, but disclosed about $800 billion in future compute and equipment commitments. Its revenue growth slowed sharply after surging earlier in the year, and revenue relative to compute expense declined, raising doubts about whether its growth can translate into sustained profits.

Why is Steve Eisman concerned that Anthropic’s growth may have been inflated?

He points to “token maxing”—encouraging heavy AI use without regard to cost. Revenue growth slowed from 135% and 147% sequentially in Q1 and Q2 to 45% in Q3, while revenue divided by compute expense fell from 2.08 to 1.70. Anthropic’s October token-price cuts of up to 50% add further uncertainty about demand and margins.

What is Eisman’s outlook for Anthropic’s and OpenAI’s IPOs?

He sees too many variables to predict whether Anthropic will ultimately deliver profits, but expects its IPO—and likely OpenAI’s—to revive enthusiasm for AI.

What concerns did this week’s technology earnings raise about AI spending and growth?

Disappointing results and stock reactions prompted questions about whether AI capital spending is slowing and whether AI-backed growth is sustainable. Amazon held its 2026 capital-expenditure budget at $220 billion, while Microsoft’s quarterly capital spending and finance leases came in below consensus, weighing on data-center and AI infrastructure stocks.

Why did investors react negatively to Google’s results?

Google shares fell 12% after earnings per share came in below the prior year, free cash flow was negative $9.1 billion, and search revenue growth slowed to 12% from 19% in Q1 2026. Investors worried that AI search could displace Google’s highly profitable, capital-light search business.

Why did ServiceNow shares plunge?

Shares fell 27% after third-quarter earnings slightly missed expectations and fourth-quarter guidance for current remaining performance obligations implied slower backlog growth, renewing fears that AI could threaten software companies.

What did Microsoft report, and why did its results affect Neo Cloud stocks?

Microsoft beat earnings and revenue expectations, and Azure revenue grew 48% to $33 billion. However, its quarterly capital spending and finance leases totaled $45 billion, below the $51 billion consensus, prompting declines in CoreWeave, Nebius, and Iren.

Why does the speaker consider CoreWeave’s borrowing costs a serious risk?

Although CoreWeave bonds carry a 9% coupon, their lower market prices imply an effective yield of about 15%, potentially making future data-center financing uneconomical. CoreWeave has $36 billion in debt and more than $50 billion in total obligations, and its borrowing-dependent model leaves it exposed to fickle credit markets while competing with stronger-balance-sheet hyperscalers.

What is the main constraint on AI data-center expansion, according to the electrical contractor?

Power infrastructure is becoming the key constraint—not GPUs, money, or memory. Transmission lines can take more than 24 months to build and substations one to two years, leaving some projects at risk of being stranded without electrical connections. The contractor believes Microsoft and Google are better positioned, while Amazon, Meta, CoreWeave, and Oracle may face greater pressure.

Why might Amazon and Meta face pressure from power constraints?

The electrical contractor says Amazon and Meta may be less prepared to secure the power required for data-center expansion than Microsoft and Google. Delays in transmission lines and substations could slow projects or leave them without electrical connections.

Why did T-Mobile’s strong results fail to lift its stock?

T-Mobile’s earnings per share rose 34% and it added 274,000 accounts, but its stock was flat because Elon Musk’s plan to make Starlink a fourth U.S. mobile provider raised concerns about a disruptive new competitor—the “tele-pocalypse.”

What is the concern about Starlink becoming a mobile carrier?

The host says Musk’s plan could pressure AT&T, Verizon, and T-Mobile. A subscriber argues that satellite-to-device service would require many short-lived, low-orbit satellites and could be costly compared with cell towers, likely serving relatively few remote-area customers. The host agrees the economics are challenging but says Musk may pursue the plan anyway.

What did Boeing’s results indicate about its turnaround?

Boeing narrowed its loss to 34 cents per share but missed expectations. Revenue grew only 4%, commercial aircraft revenue fell 3%, and disappointing cash-flow guidance helped drive the stock down 10% on Tuesday and more than 20% year to date.

What did Visa’s results indicate about consumer spending?

Visa earnings per share rose 17% to $3.48 and payment volume increased 9%, suggesting consumer spending remains supported by higher-income consumers. Mastercard’s dollar volume rose 7%.

Which companies reported strong AI-infrastructure-related results?

GE Vernova reported earnings per share up 193% and orders up 96% to $29 billion. Quanta reported earnings per share up 82%, organic revenue growth of 37%, and a record $64 billion backlog, and raised guidance.

Why did Intel and Amazon shares fall after their results?

Intel fell 18% after hours despite beating earnings and revenue because its outlook was weak: it expected the PC slump to continue, cited data-center supply-chain problems, and forecast fourth-quarter results below consensus. Amazon fell about 6% after missing earnings and revenue estimates; its unchanged $220 billion 2026 capital-expenditure budget raised questions about whether AI spending growth is peaking.

How does Eisman assess the risk posed by France’s sovereign debt to global markets?

He says France’s $4 trillion in debt is too small to pose a major threat to global fixed-income markets compared with U.S. debt of $40 trillion and BlackRock’s $15 trillion under management. He argues France depends more on BlackRock’s liquidity than BlackRock depends on French debt, and rejects claims that France need not repay its debt.

What did Kalshi traders predict about the U.S. midterm elections, and how did the speaker assess that prediction?

Traders on the federally regulated prediction market gave Democrats a 65% probability of winning both houses of Congress. The speaker considered that estimate somewhat high because polls may undercount Republican voters.

What other topics and interviews did the episode mention?

The episode mentioned a recent interview with Doomberg on energy markets, a forthcoming interview with Herb Greenberg, and planned discussion of CoreWeave, power constraints for AI data centers, and the “tele apocalypse.” It also highlighted an October 26 interview with Ben Steil of the Council on Foreign Relations about inflation, the Federal Reserve, the deficit, France, and Steil’s book on Henry Wallace.

Summary of Timestamps

Anthropic’s S-1 shows strong recent performance, including a 19% adjusted operating margin in Q2 2026, but leaves major questions about whether AI growth can translate into lasting profits. The filing also disclosed roughly $800 billion in future compute and equipment commitments.

Steve Eisman questions whether Anthropic’s earlier growth was boosted by “token maxing”—encouraging heavy AI use without regard to cost. Revenue growth slowed from 135% and 147% in Q1 and Q2 to 45% in Q3, while revenue relative to compute expense declined. October price cuts of up to 50% add further uncertainty about demand and margins.

Despite the uncertain economics, Eisman expects Anthropic’s IPO—and likely OpenAI’s—to renew enthusiasm for AI. He also points to Meta’s more than $800 billion in AI-related commitments and its planned $95 billion equity raise, which Nvidia supported with a $3 billion investment.

Disappointing technology results from companies such as Google, Intel, and Amazon have raised questions about whether AI-backed growth and capital spending can continue at the current pace. These concerns have also revived talk of a “SaaS apocalypse,” as investors reassess AI’s potential impact on software companies.

The earnings roundup showed sharply mixed results. Visa benefited from resilient spending among higher-income consumers, while S&P Global shares fell amid AI-related concerns. GE Vernova reported strong growth in AI data-center-related orders, though its stock declined after a substantial run-up.

Microsoft beat earnings and revenue expectations, and Azure grew 48% to $33 billion. However, capital spending and finance leases totaled $45 billion—below the $51 billion consensus—prompting declines in several Neo Cloud stocks, including CoreWeave, Nebius, and Iren.

Google shares dropped after earnings per share fell below the prior year, free cash flow turned negative at $9.1 billion, and search revenue growth slowed. Investors worry that AI search could weaken Google’s highly profitable, capital-light search business. ServiceNow also plunged after a slight earnings miss and guidance that suggested slower backlog growth.

Other company results were uneven: T-Mobile grew earnings per share 34% and added 274,000 accounts, but its stock was flat amid concerns about Starlink becoming a fourth U.S. mobile provider. Quanta and Eli Lilly reported strong growth, while Caterpillar shares fell as investors questioned the durability of AI-infrastructure spending.

Intel fell 18% after hours despite beating earnings and revenue estimates, as its outlook pointed to continued weakness in PCs, data-center supply-chain problems, and results below consensus. Amazon also declined after missing estimates and keeping its 2026 capital-expenditure budget at $220 billion, fueling speculation that AI spending growth may be peaking.

Eisman sees CoreWeave’s financing as a significant risk. Its bonds’ 9% coupon translates to an effective yield of about 15% at current market prices, potentially making future data-center construction uneconomical. With $36 billion in debt and more than $50 billion in total obligations, CoreWeave is also competing with hyperscalers that have stronger balance sheets.

A subscriber working on data-center projects says electricity—not GPUs, money, or memory—is becoming the central constraint on AI expansion. Transmission lines can take more than 24 months to build and substations one to two years, leaving some projects at risk of being stranded without power.

The subscriber believes Microsoft and Google are better positioned because they planned their sites more effectively, while Amazon and Meta may face greater pressure. CoreWeave and Oracle are described as further behind in securing power, suggesting that electrical infrastructure could slow data-center growth.

The host discusses Elon Musk’s plan to make Starlink a fourth U.S. mobile carrier, calling the threat a “tele-apocalypse” for traditional carriers. A subscriber argues satellite-to-device service would be costly and suited to relatively few remote customers; the host agrees the economics are challenging but says Musk may pursue it regardless.

The episode also highlights an October 26 interview with Ben Steil of the Council on Foreign Relations, covering inflation, the Federal Reserve, the deficit, France, and Steil’s book on Henry Wallace. The host mentions upcoming conversations and topics, including energy markets, CoreWeave, AI power constraints, and the “tele-apocalypse.”

Eisman argues that France’s $4 trillion in sovereign debt is too small to threaten global fixed-income markets compared with U.S. debt or BlackRock’s $15 trillion under management. He says France depends more on BlackRock’s liquidity than BlackRock depends on French debt, making claims that France need not repay its debt unrealistic.

The host closes by asking listeners to support Real Vision Playbook through Substack and YouTube and reminding them that the podcast is informational, not investment advice. The broader theme is uncertainty: AI offers significant growth potential, but financing, profitability, power supply, and competition remain major risks.