AI Investment Risks: US Market Exposure and China's Cheap Models

The stock market's current reliance on Artificial Intelligence (AI) investments has introduced significant risks, particularly due to recent developments that highlight the industry's vulnerabilities. Over the past three years, 80% of the US stock market's value gains have been attributed to AI. The top 10 companies in the S&P 500, largely driven by AI, now constitute over 40% of the entire index, meaning that owning an index fund is increasingly a bet on AI.

A recent event that underscores this risk is Coinbase's decision to nearly halve its AI expenditure by switching from expensive top-tier US AI models to cheaper, open-source alternatives that offer comparable performance. While this might seem beneficial for consumers, it reveals a deeper geopolitical struggle that threatens the stability of the US AI industry.

AI as an Arms Race: US vs. China

AI is better understood as an arms race rather than a traditional industry. Both the US and China recognize AI's "winner-take-all" dynamic. The US has implemented export controls on advanced chips crucial for AI training, while China is reportedly attempting to steal US AI technology and undermine the financial viability of the US AI market. China understands the US AI industry's vulnerability to investor confidence and its substantial debt obligations.

China's Strategy: Computational Efficiency and Distillation

Denied access to advanced chips, China has focused on achieving computational efficiency. Their strategy involves leveraging US models, potentially illegally, to undercut the revenue growth essential for the US AI industry. They achieve this through a technique called "distillation."

Distillation involves: 1. Building a simpler AI model. 2. Training it on the outputs of a massive, well-trained frontier model (like ChatGPT or Claude). 3. Creating numerous fake accounts to send millions of queries to the frontier model. 4. Capturing the answers and using these distilled patterns to teach a smaller, cheaper model to mimic the behavior of the large, expensive model.

This method allows for the creation of highly competent models at a fraction of the cost, as it bypasses the need for extensive physical data centers to process vast amounts of data.

While distillation is a legal technique used by major labs on their own models, the issue arises when a rival nation applies it to a competitor's model without authorization. In February, Anthropic accused three Chinese labs (Deepseek, Moonshot, and Minimax) of running over 16 million queries through approximately 24,000 fake accounts to distill their models. In April, the White House directly accused China of "deliberate industrial-scale campaigns to steal American AI models." On June 10th, Anthropic further alleged that Alibaba engaged in the largest such effort to date, effectively cloning Claude's intelligence. Alibaba denies these allegations, but China's history of intellectual property theft is well-documented.

This strategy is proving effective, as cheaper Chinese models are attracting major US companies. Companies like Uber, Meta, and Amazon have experienced soaring AI costs, leading them to seek more affordable alternatives. Coinbase, for instance, reduced its AI bill by nearly half by using Chinese open-source models, including one from Moonshot. Lindy, another company, also switched from Anthropic's Claude to a Chinese open-source model due to escalating AI expenses.

Chinese open-source models are approximately five times cheaper than top US models and perform comparably on AI benchmarks, making them a clear winner in terms of price-to-value ratio. This poses an extreme danger to the US AI market, which is heavily reliant on debt to build its infrastructure. Even a small diversion of revenue to Chinese models could severely impact the US AI industry and the broader economy.

The US government is aware of this issue, with the House opening an investigation into Chinese model makers. However, the intense cost pressure may necessitate a faster solution than diplomatic or legal avenues.

Market Reactions and Financial Strain

Signs of market concern are already appearing: * SpaceX's post-IPO gains largely evaporated within two weeks. * A broad tech selloff on June 23rd wiped out nearly $700 billion. * Oracle experienced its worst week since the dot-com crash of 2001 due to concerns about AI financing. * Major AI stocks are diverging, indicating investors are differentiating between companies with strong technology and revenue growth versus those relying heavily on debt.

The pressure on AI revenue is not merely a passing price war but a deliberate strategy in the US-China cold war for AI supremacy. China, with its top-down authoritarian control, can direct funds to support open-source model development, driving down prices and undermining US models' revenue potential without the same infrastructure buildout concerns.

The US AI Industry's Internal Challenges

Even without China's influence, the US AI industry faces significant challenges:

Insufficient Revenue Growth

A recent MIT study found that 95% of corporate generative AI projects yielded no measurable impact on profits. This is a major problem for an industry with possibly the most expensive infrastructure buildout in history, financed primarily through debt.

Debt and Hype

The AI market is characterized by euphoria, leading retail investors to make irrational decisions, often using leverage. This creates a dual debt-based danger zone: for companies building infrastructure and for investors using debt to invest. This debt makes both parties vulnerable to market volatility, especially in a market detached from fundamentals.

Unsustainable Spending

Normal tech companies reinvest 5-20% of revenue into development. In AI, this figure is drastically higher: Oracle spends 57% of its revenue on AI buildout, Microsoft 45%, and the biggest players allocate 94% of their core business cash flow to AI infrastructure, leaving no cushion. Sequoia estimates the AI industry needs to generate $600 billion in new annual revenue to justify current spending, a figure far from being met and continuously growing.

Historically, revolutionary technologies with massive infrastructure buildouts often bankrupt the first wave of investors (e.g., canals and railways in the UK and US, the early internet). While the technology eventually thrives, revenue often doesn't materialize fast enough to save early investors.

Accelerating Spending, Falling Prices

Unlike previous tech cycles where customer demand eventually caught up, AI is experiencing accelerating spending while prices for its usage are falling, partly due to the influx of cheap open-source models from China. This dynamic threatens the industry's viability in its current form.

Hidden Costs and Circular Payments

The true financial state of the AI industry may be worse than reported. A significant portion of revenue involves circular payments. For example, Nvidia's investment in OpenAI largely flows back to Nvidia through chip purchases. Nvidia also commits billions to buy unused capacity from CoreWeave, a cloud company it partly owns, making its own investment appear more valuable. Investment firm GMO compares this to the circular financing that made dot-com bubble companies fragile.

Furthermore, the four largest US cloud companies hold $2.1 trillion in future revenue commitments, with roughly half owed by OpenAI and Anthropic, both deeply in the red. Michael Bur warns that understating the lifecycle of data center chips (5-6 years instead of a more realistic 2-3 years) could hide over $175 billion in sustained losses. Even insiders are becoming cautious, with Microsoft stepping back from its commitment to supply all of OpenAI's computing power.

The impact of a single low-cost Chinese model is evident: Deepseek's entry into the US market in January 2025 led to a trillion-dollar loss in US AI market value in one day, causing Nvidia's largest single-day loss ever. This highlights the vulnerability to increased competition, especially as interest rates rise, making debt more expensive and shortening the runway for AI companies to generate sufficient revenue.

AI Companies' Response: Seeking Government Intervention

AI companies are increasingly seeking government protection and taxpayer support. In November, OpenAI floated the idea of federal government backstopping its financing, suggesting taxpayers guarantee its debt. While this was quickly walked back, it reflects a pattern. In March 2025, OpenAI requested tax credits, loans, and other government support for AI infrastructure development.

While government support for strategically important industries can be justified, there's a risk of creating regulatory monopolies and stifling competition. OpenAI's CEO, Sam Altman, has publicly denied wanting a bailout, stating that taxpayers shouldn't rescue companies making bad bets. The White House's AI czar, David Saxs, also stated there would be no federal bailout. However, Saxs also noted that AI investment accounts for half of America's economic growth, and a reversal could lead to a recession. Altman has also mused that for sufficiently large entities, "the government is the insurer of last resort." These statements suggest an industry positioning itself as "too important to fail."

Regulatory Capture and Safety Concerns

Companies like Anthropic are actively pushing for government regulation, framing it as essential for safety. Anthropic CEO Dario Amodei advocates for binding government regulation, mandatory safety testing, and government power to block or reverse the release of "dangerous" AI models. While AI's power warrants careful rollout, there's a significant risk of regulatory capture, where regulations are designed to benefit established players by raising barriers to entry for smaller competitors and open-source projects. Analysts note that such rules, while ostensibly for safety, conveniently make it harder for new companies to compete, especially against cheap open-source models from China.

Public sentiment towards AI is largely negative, with only about a quarter of Americans holding a positive view. Many blame AI data centers for rising power bills, and the AI buildout is driving up the cost of electronics. This widespread public aversion, combined with AI's systemic importance for national security and the economy, creates a complex situation where the industry faces immense financial strain and seeks government intervention under the guise of safety.

Investor Strategy: Understanding and Managing Risk

Given these dynamics, investors must understand the underlying game of risk and reward. AI risk is not always obvious and can be hidden within seemingly safe investments.

The 2008 Playbook: Spreading Risk

Banks, aware of the AI industry's shaky financial footing and massive debt, are employing a strategy reminiscent of the 2008 financial crisis. They are taking risky AI debt, packaging it to appear safe, and selling it off to private credit funds, insurance companies, and pension funds. This shifts risk away from banks and distributes it deeply throughout the financial system, making it so pervasive that if AI falters, the government might feel compelled to intervene with bailouts.

In 2025 alone, hyperscalers took on over $100 billion in new AI data center debt. Lenders are pooling these loans and selling pieces to pension funds and asset managers. The Federal Reserve reports that major life insurers hold nearly a trillion dollars in private debt, and pension funds are also investing in debt funds fueling AI-exposed companies like Meta and Oracle.

Protecting Yourself

Investors must: 1. Know what you own: AI risk will not be clearly labeled. It can be embedded in pensions, insurance, "safe" bond funds, and index funds that are increasingly dominated by AI stocks. 2. Diversify: Actively inquire about the sources of your investment yields and diversify beyond AI, even if you have strong conviction in its long-term potential. 3. Be humble and play defensively: History shows that even with the right thesis, getting the timing wrong or being over-leveraged can lead to significant losses. The current environment, with China driving down costs and increasing competition, adds further fragility.

While AI is transformative and will likely change the world, the current wave of investors faces a brutal historical warning: revolutionary technologies often bankrupt early investors before their full potential is realized. The combination of massive debt, insufficient revenue, geopolitical competition, and the potential for regulatory capture creates a highly volatile and risky environment.