In his latest analysis, Christopher Wood, the global head of equity strategy at Jefferies, has cautioned investors that the rising efficiency and lower costs of Chinese artificial intelligence models could challenge the current market dominance of US technology firms. He suggested that the rapid advancement of Chinese large language models might lead to significant capital destruction in the US if these cheaper alternatives gain widespread global adoption.
Christopher Wood, Jefferies’ Global Head of Equity Strategy, warns that rapid growth in low-cost Chinese AI models could trigger significant capital destruction in US markets. The surge in Chinese AI usage, combined with cooling growth rates for major US AI players, has heightened investor scrutiny of high capital spending and future returns. He warns that hundreds of billions of dollars being poured into AI infrastructure could culminate in massive capital destruction.
According to Wood’s latest GREED & Fear report, Microsoft, Alphabet, Amazon, and Meta are projected to spend a total of $695 billion on capital expenditures in 2026, rising to $870 billion in 2027. Over the two-year period, their combined spending approaches $1.57 trillion. Alphabet raised its 2026 capital expenditure guidance by another $15 billion, to a range of $195 billion to $205 billion. Investors will now focus on guidance from Microsoft, Amazon and Meta as they report earnings.
Wood’s opinion is that market share will shift toward Chinese large language models, raising questions about whether US technology companies can generate adequate returns on their unprecedented capital expenditures. The level of spending has changed how businesses traditionally known for asset-light models operate. The four big players are projected to have capital expenditures totaling about 92% of their expected 2026 operating cash flow.
After watching Anthropic’s financial performance, the market welcomed the huge investment. Anthropic’s annualized revenue run rate jumped from $9 billion in December 2025 to $47 billion in May 2026. This was the catalyst that reinforced optimism about corporate adoption and the monetization of agentic AI. Now Wood says that investors are asking where the returns on this capital will come from.
China Emerges as a Major Threat
According to the Jefferies report, the top Chinese AI models processed 36.39 trillion tokens on OpenRouter during the week ended July 19, compared with 7.39 trillion for the leading US models. Chinese models, therefore, handled nearly five times as many tokens on the global aggregation platform. Competition increased after Moonshot AI launched its open-source Kimi K3 on July 17. The model was expected to achieve roughly 95% of the performance level of Anthropic’s Claude Fable 5.
The development builds on the “DeepSeek moment” of January 2025, which first drew global investors’ attention to the cost advantage of Chinese open-source models and the associated threat of commoditization. Wood worries that the continued decline in token prices might hinder large language models from achieving sustainable profitability. The Silicon Data LLM Token Expenditure Index, which measures the average cost per 1 million AI tokens, has decreased by 25% from its late-May high to $1.55. While lower prices could boost long-term demand for computing resources, they also jeopardize the profitability assumptions crucial to the current investment cycle.
However, the risk no longer remains limited to equity valuations. AI infrastructure investments are increasingly being financed through debt rather than cash from major players, adding a credit-market aspect to the boom. In 2026, the top players issued $194 billion in investment-grade debt, making them the largest source of issuance and significantly surpassing the US energy sector’s $55 billion.
Signs of Concern
Credit markets are signaling increased caution. The spreads on 10-year bonds from Amazon, Alphabet, and Meta have risen to 78, 70, and 104 basis points over US Treasuries, compared to 61, 57, and 87 basis points on July 3. Oracle, with a more leveraged position in the AI infrastructure sector, saw its $120 billion debt downgraded to BBB- on July 9, one notch above junk status. Since the downgrade, its 10-year bond spread has widened from 176 to 219 basis points.
According to figures cited by Wood, by the end of the first quarter of 2026, Microsoft, Alphabet, Amazon, and Oracle collectively held approximately $2.1 trillion in remaining performance obligations. These obligations, which represent contractual commitments for future revenue, have increased by 184% from $740 billion a year earlier. Nearly half of this backlog is owed by OpenAI and Anthropic. Microsoft’s share of the backlog is about 49% linked to these two AI companies, with Oracle’s exposure at 54%, Google’s at 43%, and Amazon’s at 51%.
No Company is currently profitable
It is widely believed that neither OpenAI nor Anthropic is currently profitable. However, Wood remains hopeful, noting that Anthropic appears to be more comfortably positioned. The concentration means big players have effectively extended large, unsecured commitments to cash-burning customers while building data-center capacity on the assumption that future computing demand will materialize. The risks are even greater for specialized cloud providers that have borrowed to finance chips and infrastructure. CoreWeave has borrowed about $30 billion, while its five-year credit-default-swap spread has climbed from 452 basis points in early June to 701 basis points.
Wood has also raised questions about balance sheet risks. He cited an estimate that the five leading US players had accumulated $662 billion in future data-center lease commitments that had not yet commenced, up from $152 billion at the end of 2023. Another study put their off-balance-sheet or hidden debt at $1.65 trillion in the second quarter, exceeding the roughly $1.35 trillion of debt reported on their balance sheets.
The investment wave’s impact on accounting has yet to fully materialize in spending. In the first quarter, Microsoft, Amazon, Alphabet, and Meta collectively allocated $130 billion to capital expenditures, while their depreciation and amortization expenses totaled $41.6 billion. Despite this, depreciation rose by 33% compared to the previous year. Over the past year, their annualized earnings grew by $106.6 billion, reaching $447 billion in the four quarters ending March. Non-operating income also increased by $71.5 billion to $83 billion, representing roughly two-thirds of the overall earnings growth.
Markets Took a Note of It
Global technology markets have been sensitive to these concerns. The Kospi index in South Korea recently fell nearly 11% in a single session, reflecting broader regional anxiety about the sustainability of the AI rally. Major technology firms, including Samsung Electronics and SK Hynix, faced heavy selling pressure during this period. Market observers note that after an extended period of rapid stock price appreciation, it is not uncommon for investors to re-evaluate their positions and take profits, especially amid uncertainties about future growth and spending efficiency.
Wood agreed that AI is a durable advancement rather than a passing fad. He notes that falling computing costs may boost its adoption. However, he warns that markets may have overestimated short-term gains while overlooking the substantial capital and credit risks crucial to long-term success. But he also foresees a period of disillusionment with AI after the initial enthusiasm.
