What Wall Street Really Thinks About the AI Trade Now | Odd Lots

Summary

Wall Street remains broadly bullish on AI, but the discussion highlighted a growing divide between enthusiasm for the technology and the financing and revenue realities of building it. Alex Altmann said investors still generally believe the AI buildout will continue, while signs of funding stress among “neocloud” companies suggest capital is becoming more selective. He cautioned that interest rates and financing constraints still matter, especially for highly leveraged businesses.

Key market themes

• AI financing risk is appearing first at the edges of the ecosystem. Neoclouds—companies building compute capacity to sell to customers—often have little or no operating cash flow and depend on external funding. Altmann described them as riskier than cash-generative hyperscalers such as Google and Amazon.
• Signs of tighter financing: Cipher’s convertible financing reportedly hurt its share price, while CoreWeave and other AI-related names came under pressure. Data-center bond spreads had already been widening for months. Altmann framed these developments as early warning signs, not proof that the AI investment thesis has failed.
• Capital is not unlimited. The discussion linked higher interest rates to constraints on issuance and financing. A Barclays survey found about 60% of investors expected financing to be a problem, a response Altmann viewed as surprisingly low given recent market developments.
• AI is not immune to rates. The hosts and guest pushed back on the idea that exceptional growth makes borrowing costs irrelevant. Rate sensitivity may be most consequential for leveraged firms and less-established companies at the edge of the AI buildout.
• AI revenue remains the central test. Altmann questioned whether enterprise spending alone can deliver returns on the timeline investors expect. He argued that broader consumer adoption and useful applications could be important to the next phase of growth.
• AI exposure is driving market concentration and rotations. Altmann said roughly two-thirds of next year’s S&P earnings growth was expected to come from semiconductors, rising to about 75% when the broader AI ecosystem is included. The hosts described the index as increasingly concentrated in AI-linked earnings.
• Volatility can look subdued at the index level even while individual stocks and factors swing sharply. Altmann attributed this partly to the growth of multi-manager funds, crowded positioning, and rapid rotations between AI-related exposures. He said the shifting nature of the “AI factor” makes it difficult for funds to map and hedge.

AI economics and technology

• Large total-addressable-market claims need context. The discussion referenced a roughly $30 trillion AI TAM estimate and compared it with historical technology revenues. The cumulative, inflation-adjusted revenue of Microsoft and Apple since 1990 was cited as less than $10 trillion.
• The S&P 500 was described as generating about $17 trillion in revenue, with $11–12 trillion in expenses and roughly $5–5.5 trillion in net profit. These figures were used to question how much revenue AI investment ultimately needs to generate.
• Open-weight models are gaining share and reducing costs. Altmann cited estimates that open-weight models’ share had risen from around 20% to as much as 80% in some measures. He said their token costs could be around one-hundredth of comparable alternatives, while noting that GPU rental rates had risen.
• Local AI could alter demand for data-center compute. The speakers discussed increasingly capable multi-billion-parameter models running on laptops or phones, potentially reducing the need to send every task to large data centers. Altmann suggested that existing technology applied in new ways may produce important consumer uses.
• Consumer applications remain an open question. The hosts discussed using ChatGPT for practical scheduling tasks. Altmann said a widely adopted consumer application could be important for expanding AI revenues, but did not identify a specific “killer app.”

Market data, company results, and numbers cited

• Net issuance: Barclays’ historical analysis found a relationship between interest rates and corporate-market issuance that appeared to break down in 2026, with unusually high issuance relative to benchmark rates. The analysis covered equities, Treasuries, investment-grade and high-yield debt, using data back to 1995.
• Investment-grade and high-yield net issuance was forecast at $1.1 trillion for the year discussed.
• AI capital expenditure: Forward-looking CapEx figures discussed were estimated at roughly 12% of U.S. GDP.
• Oracle: The stock was said to be up 12% on the day after new guidance. The hosts also said Oracle planned to limit additional fundraising to around $20 billion; the precise timing and conditions were unclear in the subtitles.
• Alphabet: Shares fell 7.9% after revenue missed estimates.
• Microsoft: Shares rose 6.4% after its Q1 results. Azure revenue grew 48%, ahead of a Bloomberg-compiled estimate of 44.8%.
• Google Cloud: Revenue was cited at about $30 billion, with growth of 103%.
• Alphabet was also said to process 32 billion tokens per minute, up 40%.
• Semiconductors and memory: Some memory stocks, including Hynix, were described as trading at about 3× earnings. The discussion cautioned that low multiples may reflect expectations that current earnings are near a cyclical peak—not necessarily a market expectation of sustained rapid growth.
• Software: Altmann noted that software had moved close to 12-month highs, complicating earlier positioning that treated software as an AI-disruption short.

Frameworks and cautions

• Assess the AI ecosystem by financing position: Distinguish cash-generative hyperscalers from peripheral compute providers that rely on debt or new equity.
• Track early warning signs: Watch financing access, bond-spread widening, convertible issuance, and hyperscaler CapEx guidance.
• Test the investment case against revenue: Compare projected AI spending with the addressable revenues and profits available, rather than relying only on large TAM estimates.
• Consider model efficiency and deployment location: Open-weight models, falling token costs, and locally run AI could change demand for cloud compute and GPU capacity.
• Do not infer low risk from calm index measures: Index-level volatility can mask sharp stock- and factor-level moves.
• Avoid assuming high growth makes rates irrelevant: Financing costs may still constrain highly leveraged AI businesses.
• No direct buy/sell recommendation or explicit “not financial advice” disclaimer was stated.

Assets, companies, and instruments mentioned

• Indices and market measures: S&P 500, VIX.
• Companies and AI ecosystem: Alphabet/Google, Microsoft, Amazon, Oracle, IBM, Meta, Nvidia, Anthropic, Apple, CoreWeave, Cipher, Hynix, OpenAI/ChatGPT.
• Sectors and themes: AI, semiconductors, memory, software, data centers, cloud computing, biotech, industrials, quant trading, Bitcoin mining.
• Instruments and markets: Investment-grade and high-yield bonds, Treasuries, convertible securities, equities, GPUs, and GPU rentals. The discussion also referenced REITs as a comparison for neocloud business models.

Presenters and sources

Presenters: Tracy Alloway and Joe Weisenthal (Odd Lots); guest Alex Altmann, Barclays’ global head of equities tactical strategies.

Sources referenced included Barclays’ investor survey and research note “Placing Your AI Chips More Carefully”; Bloomberg-compiled estimates and market data; Anthropic’s S-1 filing; and a cumulative-revenue statistic credited to Cameron Crease.