Investors should avoid heavily indebted AI infrastructure plays like Oracle (ORCL), which is down 35% this year and burning $24 billion in negative free cash flow.

The ongoing AI price war—highlighted by low-cost Chinese models like DeepSeek priced at just 87 cents per million tokens—will severely compress profit margins for US-based closed-source labs.

High borrowing costs, recent credit rating downgrades, and demands for massive collateral like Wisconsin's $7 billion requirement significantly increase default risks for capital-intensive tech companies.

Prioritize companies with strong organic equity and positive cash flows rather than those relying on heavy debt financing to fund the data center boom.

Oracle (ORCL)

• Oracle's stock is down 35% for the year
• The company's credit rating has been downgraded to sit just one notch above junk status
• Oracle borrowed $43 billion over the past year to build data centers, against $67 billion in revenue
• The company is running negative free cash flow, burning roughly $24 billion over the same period
• The state of Wisconsin recently demanded that Oracle post $7 billion in collateral before building a data center in the state, which will increase borrowing and operational costs
• The discussion highlights risks associated with funding the AI infrastructure boom through heavy debt rather than equity or profits

Takeaways

• The heavy debt loads required to build out AI infrastructure are starting to create severe financial strain for participating companies, signaling potential bursting points in the broader AI investment bubble
• Investors should closely monitor corporate free cash flow rather than just top-line revenue when evaluating companies heavily exposed to the AI infrastructure build-out
• High borrowing costs and credit rating downgrades increase the risk of default for capital-intensive tech companies relying on debt financing

Artificial Intelligence (AI) Sector & Infrastructure Themes

• An AI price war has broken out, turning into a race to the bottom for model pricing as major players like Meta and Google's Gemini cut prices significantly
• Chinese open-weight AI models (such as DeepSeek and Moonshot AI's Kimi K3) are disrupting the market by offering performance comparable to top American models at a fraction of the cost (e.g., DeepSeek priced at 87 cents per million output tokens compared to OpenAI's $45 and Anthropic's $50)
• Chinese open-weight models have grown from less than a third of traffic in late 2025 to about two-thirds recently, capturing significant market share from American alternatives
• The broader AI industry is increasingly reliant on debt financing rather than equity or profits, raising systemic financial risks if investor funding slows down
• A divide exists in global sentiment regarding AI: while American consumers and markets exhibit growing anxiety and "doom trolling," international and Chinese populations show significantly higher levels of pragmatic excitement and adoption

Takeaways

• The rise of heavily subsidized, low-cost open-source and foreign AI models creates margin compression and pricing pressure for US-based closed-source labs like OpenAI and Anthropic, especially those reliant on venture capital
• Open-source AI models are proving that the "secret sauce" of big proprietary labs is diminishing, which may democratize AI capabilities globally but threaten the business models of pure-play frontier labs
• Investors should exercise caution regarding companies taking on high leverage to fund data centers and infrastructure amid shifting pricing power and margin compression in the AI sector