The boom in AI infrastructure now carries a familiar telecom warning sign. Demand still looks strong, yet investors are questioning the funding model behind it.
In late October 2026, major cloud players reported fast growth. Google cloud revenue more than doubled year over year. Amazon grew 39 percent, while Microsoft rose 48 percent. Even Intel posted almost 80 percent data center sales growth.
However, markets still punished several stocks. Investors appeared less worried about AI usage. They focused instead on massive spending for chips, power, and data centers.
This pattern looks familiar to telecom veterans. In the late 1990s, carriers built huge fiber networks. Much of that capacity sat unused after the crash. Yet the cheap fiber later powered streaming, cloud services, and today’s traffic-heavy internet.
AI may follow a similar route. Early builders could lose money, while later owners buy assets cheaply. That could drive down computing costs and expand everyday AI use.
Analyst Vish Nandlall compares the cycle to earlier technology waves. Economist Carlota Perez has studied these patterns for decades. She told the Financial Times: “I have not seen a golden age happening without a crash.”
The core issue is not weak demand. It is expensive ownership. Companies are spending faster than many services generate cash. Meta, Alphabet, and Amazon reported negative free cash flow. CoreWeave, a major AI chip rental player, carries heavy debt.
If these structures break, assets will not vanish. Data centers, grid connections, and cooling systems will still matter. Chips age quickly, often within three to five years. Buildings and power connections can remain valuable for decades.
That split matters for future owners. Utilities and infrastructure funds may buy distressed sites. They usually borrow at lower rates than riskier tech players. They also understand electricity economics better than most software companies.
Power will become central to AI pricing. Once hardware costs fall, electricity becomes a larger share of each task. Reliable, low-cost power could separate winners from weaker providers.
There is also a utilization challenge. AI chips cost money whether idle or busy. Chatbots create daytime peaks, much like old electricity demand peaked at night. Background agents could help fill quiet hours by running tasks later.
That model could make AI resemble a utility. Users would delegate shopping, research, booking, or admin work. Providers would earn small fees across many daily tasks.
Still, a crash is not guaranteed. Builders may refinance debt. Demand may rise fast enough to fill capacity. Scarce powered sites could keep prices high.
For telecom and cloud observers, the lesson is clear. Infrastructure bubbles can destroy investors while strengthening the future network. Fiber proved that point. AI infrastructure may be the next test.
