Wall Street Warns of AI Bubble Risks as Semiconductor Stocks and Valuations Plunge

Wall Street heavyweights are warning of a potential AI bubble amid soaring infrastructure costs, surging debt, and slowing revenue growth. Semiconductor stocks and tech valuations face severe corrections as market participants question long-term profitability.

Wall Street Warns of AI Bubble Risks as Semiconductor Stocks and Valuations Plunge
Photo: Unsplash / tvick

Wall Street is increasingly questioning the financial viability of the artificial intelligence boom, as market heavyweights warn of structural imbalances, soaring debt, and a potential bubble. Citadel hedge fund manager Ken Griffin cautioned that the corporate sector is struggling to extract genuine value from language models, comparing the hardware to empty seats on an aircraft. Meanwhile, Ray Dalio of Bridgewater Associates pointed to classical bubble risks driven by rising interest rates and liquidity pressures, while Jefferies investment banker Chris Wood warned that intense competition from China is commoditizing language models and eroding U.S. margins.

This turbulence has triggered sharp corrections in the semiconductor sector. The Philadelphia Semiconductor Index (SOX) opened the year with a 9% gain in the first quarter and a 95% surge in the second, but subsequently plummeted by 11% in the third quarter and another 12% in the fourth. Amid this cooling market, Anthropic was forced to slash its targeted IPO valuation from $2 trillion to $1.5 trillion. This single adjustment equals the entire market capitalization of the Tel Aviv Stock Exchange, which stands between $570 billion and $580 billion.

Historical Parallels and Infrastructure Costs

Financial analysts draw parallels to the Panic of 1873, noting that modern AI infrastructure relies on massive, power-hungry data centers requiring approximately $10 trillion in investments. Training a cutting-edge AI model requires roughly 18 octillion computations—50 million times more than early iterations. Performing this computing load on a single Nvidia H100 GPU would take 64,000 years, requiring clusters of 100,000 processors to complete the task within eight months. Concurrently, investor Michael Burry has raised concerns over aggressive accounting practices, noting that tech firms amortize semiconductor depreciation over five years while the rapid technological cycle demands hardware upgrades every two to three years.

Mounting Pressures on Neocloud Providers

Financial strain is intensifying among neocloud providers such as CoreWeave. While CoreWeave projects revenue growth from $5 billion in 2025 to between $12.4 billion and $13.2 billion, the firm continues to operate at a substantial loss. With $7.5 billion in debt against $6.4 billion in cash and quarterly interest payments of $640 million, its bond yields have surged from 9% to 15% to compensate for rising risk. Leaked internal data from Anthropic reveals a similarly stark divergence, with $16.2 billion in revenue offset by an $80.7 billion loss, alongside a deceleration in quarter-over-quarter growth from 145% to 46%. With technology stocks currently accounting for 51% of the S&P 500 index, market participants are closely monitoring whether massive infrastructure investments will yield expected returns or trigger a broader market correction.

Related News