The rapid investment in artificial intelligence infrastructure raises questions about the sustainability of the current market dynamics, prompting concerns of a potential AI bubble.
Wall Street is known for its numbers, but this year, it has settled on a curious consensus regarding the S&P 500. Analysts predict the index will close 2026 between 7,000 and 8,250, with a median forecast of 7,620, only slightly above its current position. The Dow Jones Industrial Average is expected to hover around 52,500. Ed Yardeni, one of the most optimistic voices on Wall Street, has raised his target to 8,250, citing corporate earnings that are increasingly supported by solid data rather than mere market enthusiasm. In contrast, Bank of America and Stifel predict a more conservative range of 7,000 to 7,100. This narrow spread between the most bullish and bearish predictions is the tightest seen in nearly a decade, a period historically associated with market surprises.
The underlying reason for this consensus, as well as the unease that accompanies it, is the same: artificial intelligence (AI).
Consider the scale of investment currently underway. Major tech companies such as Microsoft, Alphabet, Amazon, and Meta, along with Oracle, are projected to spend between $600 billion and $700 billion on capital expenditures in 2026—almost double their spending from the previous year. Notably, around three-quarters of this amount, exceeding $450 billion, is earmarked for AI infrastructure, including data centers, chips, and power resources. Microsoft’s AI division alone has surpassed a $37 billion annual run rate, reflecting a staggering 123 percent year-over-year increase. Meanwhile, Google Cloud’s backlog has nearly doubled to $467 billion. These figures are substantial and, importantly, they are not mere projections.
However, when these investments are juxtaposed with the revenue generated by companies utilizing this computational power, the situation becomes concerning. OpenAI concluded 2025 with approximately $20 billion in annual recurring revenue, while Anthropic’s run rate reached $9 billion in January. Together, these figures represent only about three percent of what the hyperscalers plan to invest in building capacity for them this year. This discrepancy raises critical questions about the sustainability of the current AI investment boom.
What distinguishes this cycle from the dot-com crash and warrants serious consideration is the financing structure that underpins it. Nvidia has committed $100 billion to OpenAI, which in turn uses a significant portion of that capital to purchase Nvidia chips. Additionally, Microsoft and Nvidia have jointly pledged $10 billion to Anthropic, which has committed $30 billion to Azure cloud services. Michael Burry, known for his early warnings about the subprime mortgage crisis, is now shorting Nvidia and has raised a provocative question on social media: “OpenAI is the linchpin here. Can anyone name their auditor?” He further asserts that true end demand is “ridiculously small” and that many customers are effectively funded by their suppliers.
This situation does not constitute accounting fraud; rather, it reflects vendor financing, a practice with historical precedent. For example, Uber subsidized its driver and rider base in its early years to capture market share. However, there is a crucial distinction between subsidizing adoption in a proven industry and subsidizing demand for a technology whose return on investment remains a topic of debate in boardrooms. Even Sundar Pichai, typically measured in his assessments of Google’s strategic bets, acknowledged to the BBC that there are “elements of irrationality” in the AI market. Sam Altman has echoed similar sentiments regarding investor sentiment in general. When executives begin to recognize the exuberance, it becomes permissible for others to do so as well.
Compounding these concerns is a shift in financing that should alarm anyone who experienced the 2008 financial crisis. The capital expenditures of hyperscalers, after accounting for dividends and buybacks, now exceed their free cash flow. They are increasingly relying on debt to fund AI expansion, having issued over $100 billion in bonds in 2026 alone. Investors are demanding record levels of credit-default-swap protection against the risk that some of this debt may default. A boom financed by a company’s own balance sheet tends to be more resilient, while one increasingly reliant on external funding is inherently more vulnerable.
Meanwhile, the federal government is not regulating this trend; rather, it is facilitating it. In March, former President Trump convened seven hyperscalers at the White House to sign the Ratepayer Protection Pledge, committing them to fund their own power generation to prevent American households from bearing the costs associated with AI data centers. The administration has also offered chip tariff exemptions in exchange for cooperation, intervened in the PJM power market to expedite plant construction, and rolled back Biden-era AI safety reporting requirements in favor of a “Federal Preemption” standard. The message to the industry is clear: accelerate development, and we will remove obstacles. This represents industrial policy aimed at promoting a narrative of American AI supremacy, regardless of whether the underlying economics have caught up.
It is crucial for investors—both retail and institutional—to consider that ten companies, with Nvidia and Broadcom at the forefront, now account for more than a third of the Nasdaq-100’s total weight and over half of the broader Nasdaq Composite’s performance. The remaining ninety-plus companies span various sectors, including software, healthcare, retail, and industrials, with memory-chip manufacturers like Western Digital, Seagate, and Micron performing well amid a DRAM shortage. However, when an index becomes top-heavy and heavily reliant on a single, unresolved question—whether AI revenue reflects genuine demand or merely financed demand—a revaluation can have widespread repercussions.
None of this implies that the AI buildout is a mirage or that the technology will not eventually validate the expenditures. The compute shortage is real, the backlogs are substantial, and the productivity gains, where they have materialized, are genuine. However, distinguishing between “real” and “fully priced” is essential, and currently, Wall Street’s tightest consensus ever appears to be betting on both simultaneously. This is typically the moment to raise the question that no one at the table wants to voice.
According to The American Bazaar.