A tech analyst has raised the alarm, suggesting the artificial intelligence investment frenzy is showing clear signs of an expanding bubble, with increasingly creative financing structures prompting questions about how much capital can be poured in without clearer returns.

The analyst, Sara Awad, indicates that the bubble risk is more pronounced in AI capital expenditure than in the broader semiconductor sector. She points to a growing disconnect between the funds allocated to AI infrastructure and the returns generated from those investments.

"I believe that on the issue of a bubble, it does exist and it is expanding," Awad states. She argues that the greater risk lies in AI capital expenditure and how companies are financing these construction projects.

Awad notes that capital flows within the AI ecosystem are increasingly displaying circular characteristics. She cites examples where Nvidia (NVDA.US) considered providing around $250 billion in funding support for OpenAI and engaged in negotiations for up to $350 billion to equip data centers with chips. Additionally, Nvidia has invested in so-called "new cloud" companies, including Nebius and CoreWeave.

The concern is that chipmakers may increasingly be financing the customers who purchase their products. "It appears that Nvidia is increasingly funding its own clients," Awad remarks, suggesting that this structure raises the question of how much additional capital can be funneled into AI infrastructure without a substantial return on investment.

Awad also points to financial pressure on major tech companies. Free cash flow at Google and Amazon has turned negative, and discussions by Meta about potentially leasing excess computing capacity have sparked external doubts about whether the company is spending too aggressively on AI infrastructure.

Another development Awad highlights is Nvidia's collaboration with six major Wall Street banks to establish a standalone computing financing platform, with plans to gradually mobilize over $500 billion in third-party capital for AI infrastructure construction. This initiative aims to provide funding to Nvidia's customers at favorable interest rates, while Nvidia CEO Jensen Huang has described chips as an investable asset class.

Awad suggests that the sheer scale of planned AI infrastructure spending makes financing issues increasingly critical. She notes that AI capital expenditure has already exceeded $740 billion, even as the cost of using AI models is declining. She also mentions growing competition from Chinese AI models as another source of pressure.

A report from Jefferies indicates that the average price of AI inference services has fallen from $2.04 to $1.45 per million tokens between late May and mid-July, to between $1.16 and $1.18 in August. This price decline suggests that the AI industry is becoming more cost-sensitive, even as infrastructure spending remains high.

Therefore, in Awad's view, a potential bubble burst is less likely to be driven by a single event and more by the sustainability of financing behind AI infrastructure construction. "I believe the key to the bubble bursting lies in how we begin to finance AI infrastructure construction, moving from free cash flow to credit... and that leads us into a higher-risk territory," she says.

In other words, the real risk is not about the promise of AI technology itself, but about where the money to support this high-stakes bet comes from and whether it can be repaid. AI is undoubtedly a revolutionary technological direction, but when capital games replace technology itself as the main narrative, the market's pricing logic needs to be re-examined. And that day may not be far off.