Accounting Mismatch in AI Profits | Jim Chanos and Val Zlatev on Long/Short Alpha in AI & Semis
Investment experts Jim Chanos and Val Zlatev discuss opportunities and risks in AI and semiconductor investing, debating whether current valuations reflect sustainable returns. They highlight an accounting mismatch where chip sellers recognize profits immediately while AI spenders capitalize costs, and examine whether memory and semiconductor supply constraints can support the exponential growth forecasts.
Summary
In this panel discussion hosted at the Macro Minds Symposium, legendary short seller Jim Chanos and hedge fund manager Val Zlatev explore the AI boom from both bullish and bearish perspectives. The central thesis revolves around an accounting disconnect: companies selling chips and infrastructure (Nvidia, ASML) recognize revenues and profits immediately, while hyperscalers like Microsoft and Google capitalize these costs over years of depreciation. This creates a temporary earnings boost that may not reflect underlying economic reality.
Chanos draws parallels to the late 1990s internet boom, noting that despite traffic growing exponentially (doubling every quarter according to MCI), actual economic growth remained flat. He emphasizes that while AI may create winners and losers, the aggregate economic impact remains uncertain. He is net long AI via index but shorts specific business models he views as inherently unprofitable, particularly neoclouds (data center companies like CoreWeave) which he characterizes as equipment leasing firms rather than technology companies, earning only 5-8% returns on invested capital at best.
Zlatev counters with evidence of genuine near-term AI adoption impacts, showing that major tech companies have increased operating profits while barely growing headcount over the past 3-4 years. He points to rising GPU rental prices for older chips—up 40-50% since January—indicating genuine capacity constraints. However, he agrees with Chanos that the equipment leasing business model (neoclouds) is fundamentally limited, preferring to invest in the actual chip designers whose technology creates the value.
Both discuss memory semiconductor pricing, where Zlatev argues we're at a peak demand period that could last 2-4 years before rollover. Memory stocks trade at 6-7x forward multiples, implying market expects price collapse within 6-9 months—a timeline Zlatev views as too pessimistic given supply constraints. Memory equipment makers can only grow production 30-35% annually due to physical constraints in cleanroom capacity and equipment supplier limitations. Consumer electronics makers face memory costs rising from 20% to 50% of bill of materials, pressuring PC and smartphone demand.
On valuation comparisons, Chanos notes Intel trades at higher revenue multiples than Nvidia despite inferior competitive positioning, suggesting relative value misalignment between oligopolistic and competitive markets. Zlatev defends semiconductor valuations as more balanced than the 1999-2000 bubble, where Cisco traded at 160x P/E—current highs are 50-60x forward multiples in networking, not across the board.
Both are skeptical of SpaceX's space data center thesis. Chanos calculates that power represents only 5-7% of data center costs, making orbital data centers uneconomic despite Elon Musk's vision of needing one terawatt of compute capacity. He emphasizes that Starship hasn't yet reached orbit despite 12 flights and six explosions. Zlatev suggests Musk's rationale concerns required compute scale rather than cost reduction, but remains skeptical of near-term viability.
Zlatev critiques AI bearish forecasts that assume quick reversal, noting that OpenAI token counts show verifiable growth rather than just corporate projections. Yet he acknowledges that scaling laws (empirical rather than physical) could break if new AI architectures emerge, and cites DeepSeek as an example of reduced costs without scaling law breakthrough.
The host concludes that while individual semiconductor opportunities exist, the U.S. stock market increasingly represents a concentrated bet on whether AI investments deliver promised returns, with tremendous rewards or losses depending on outcome.
About this episode
In this panel at MacroMinds Symposium, Jack Farley sits down with legendary short seller Jim Chanos and Val Zlatev, Portfolio Manager and Partner at Analog Century Management, to analyze the long and short opportunities of the AI and semiconductor boom. Chanos highlights a significant timing disconnect wherein chip suppliers recognize revenues immediately while hyperscalers capitalize their massive infrastructure costs—a trend mirroring the late-1990s CapEx boom before tech earnings collapsed by 40%. Chanos expresses deep skepticism toward "neo-cloud" data center developers like CoreWeave, modeling a very generous ten-year GPU lifespan (depreciation schedule) to forecast low pre-tax returns on invested capital. From a micro perspective, Val Zlatev outlines the structural upside for high-demand memory stocks, noting they trade at cheap forward multiples because physical supply chain constraints hard-cap semiconductor equipment manufacturing growth at 30% annually. The discussion also scrutinizes Elon Musk’s projection for one terawatt of compute capacity, breaking down the immense real-world barriers facing space data centers, including launch costs, space radiation, and maintenance logistics. They also dissect the SpaceX S1 filing, revealing that the primary rocket launch division continues to lose money despite the profitability of Starlink. Recorded on June 4th at MacroMinds Symposium. About MacroMinds: At MacroMinds, our vision is to unite the investment community in support of organizations that are making a meaningful difference in the lives of students and their families. By partnering with high-impact nonprofits that serve socio-economically disadvantaged communities and schools, MacroMinds is committed to helping close the educational gap and expand opportunity across the New York area. MacroMinds website: https://macrominds.org/ https://macrominds.org/donate/ Charities supported by 2026 Symposium: NYC First: https://macrominds.org/nyc-first/ Opportunity Music Project: https://macrominds.org/opportunity-music-project/ 100 Women in Finance: https://macrominds.org/100-women-in-finance/ Follow Jim Chanos on X https://x.com/RealJimChanos?lang=en Follow Jack Farley on X https://x.com/jackfarley96 Follow Monitoring The Situation (MTS) on X https://x.com/MTSlive Follow Monetary Matters on: Apple Podcasts https://rb.gy/s5qfyh Spotify https://rb.gy/x56dx5 YouTube https://rb.gy/dpwxez Timestamps 00:00 Trailer 01:01 Jack's Introduction 02:22 "Finding Value On Both the Long and Short Side in AI" 07:14 "There's A Disconnect In The Profitability" Between Semis and Hyperscalers 14:00 NeoClouds and Coreweave 18:20 SpaceX & Data Centers in Space 26:20 DotCom Bubble Analogy on Assumptions of Secular Hypergrowth 35:00 Potential "Nightmare" Scenario for Longs? 37:14 Huge Bull Market in Memory 48:00 Semi Cap Equipment Names 52:34 Jack's Conclusion #chanos #stocks #investing #monetarymatters #economy #ai #jackfarley #memory #semiconductor #semiconductorrelatedstocks #shortselling
Key Insights
- Companies selling picks and shovels (Nvidia, equipment makers) recognize revenues and profits immediately, while hyperscalers spending the same dollars capitalize those costs over many years, creating a temporary earnings boost that masks underlying economic reality
- In the decade after the internet was introduced (1995-2005), U.S. economic growth and corporate profitability growth rates remained virtually identical to the pre-internet decade, suggesting broad transformative technologies may not drive aggregate economic growth
- GPU rental prices for older chips (6-7 years old) increased 40-50% since January 2025, indicating genuine near-term capacity constraints driven by exploding token usage from reasoning models, agents, and increasing context windows
- Neoclouds (CoreWeave, Nimbus) are fundamentally equipment leasing companies earning 5-8% single-digit returns on invested capital even under heroic profitability assumptions with 10-year GPU depreciation, making them poorer investments than chip designers
- Semiconductor equipment makers (ASML, Applied Materials) have a physical constraint limiting capacity growth to approximately 30-35% annually due to cleanroom supply, wafer bits production limits, and equipment supplier bottlenecks
- Memory component costs for PC and smartphone manufacturers rose from approximately 20% to 50% of bill of materials, forcing price increases to consumers and creating demand elasticity risk that could push unit volumes down mid-teens
- Intel currently trades at higher revenue multiples than Nvidia despite Intel lacking competitive advantage in a fragmented CPU market, while Nvidia maintains oligopolistic positioning in GPUs, suggesting valuation misalignment between market structures
- AI scaling laws are empirical rather than physical laws and could be broken by new architectures; the MCI internet traffic doubling claim was actually only annual doubling, showing how exaggerated growth expectations can drive unsustainable capex
Topics
Transcript
[0:00] There is a disconnect in the profitability accounting. The companies that are selling the picks and shovels are recognizing revenues and profits immediately. The hyperscalers and others who are spending those very same dollars are capitalizing those costs. >> These chips are so tight as we speak >> that the prices rental prices for GPUs which are really old like six, seven, eight years old are going up in price as we speak. I really take a John Dist eye [0:30] on these forecasts of just immense need for compute at today's prices. Um it might happen but history tells us that that these kind of insane exponential growth rates tend to get constrained by the real world. >>…
Full transcript available for MurmurCast members
Sign Up to AccessMore from The Monetary Matters Network
Massive AI Crackdown Is Coming (It’s Simply Too Good) | David Woo
David Woo argues that AI has become too powerful and will face significant regulatory crackdowns, presenting the primary downside risk to AI stocks. He contends that real capex growth is slowing despite nominal earnings appearing strong, driven by input cost inflation rather than increased output, while accessible AI capabilities are plateauing despite claims of breakthroughs.
America’s $205 Billion Fund You’ve Never Heard Of | Conor Coleman
Conor Coleman, Chief of Staff at the Development Finance Corporation (DFC), discusses how the U.S. government's international investment arm is rapidly deploying $25 billion to counter China's Belt and Road Initiative through strategic investments in critical minerals, infrastructure, and technology across emerging markets. The DFC has pioneered creative financing structures and partnerships with private sector firms to mobilize capital at scale while advancing U.S. foreign policy objectives.