Executives building the artificial intelligence boom are unwavering in their conviction. Demand, they say, is effectively bottomless — even as the stocks that ride on that confidence continue to wobble. Pat Gelsinger, the former Intel CEO who now leads investment firm Playground Global, put it plainly: he thinks of AI demand as almost unlimited, with energy availability being “the only real limiter.” His statement echoes a growing chorus across the technology sector, from chipmakers to data-center operators, all of whom insist that the hunger for AI compute will only intensify.
The order books support the narrative. Lumentum, a supplier of optical components essential for high-speed data-center connectivity, reports that its products are sold out for the next five years. That kind of forward booking is rare even in high-demand industries, and it signals that the infrastructure build-out is far from complete. Similarly, Nvidia has been unable to keep up with GPU demand for months, and its upcoming Blackwell architecture is already oversubscribed. These are not fabricated signals; they reflect real purchasing commitments from hyperscalers, enterprises, and governments racing to deploy AI.
Yet despite these bullish indicators, the stock market has become jittery. The PHLX chip index, a benchmark for semiconductor stocks, has gained roughly 60% year-to-date. At that level, the price already assumes years of flawless execution. When expectations run this hot, good news stops being good enough. Samsung, for example, forecast a massive profit increase — yet its shares fell after a 12-month rally of more than 360%. The pattern repeated at Cerebras, a startup specializing in wafer-scale AI chips, which doubled its revenue only to see the stock drop. Investors, it seems, are pricing in perfection, and any hint of a miss — even a beat that is not massive enough — triggers sell-offs.
Why the market is no longer taking executives at their word
The disconnect between corporate confidence and market behavior is not new, but it has become particularly acute in the AI sector. Meta added to the anxiety by announcing it would sell off its excess AI computing capacity. Investors were left to parse the message: was this a smart monetization strategy, or an admission that the company bought more compute than it needs? The ambiguity fueled uncertainty. Similarly, after a wave of massive earnings beats from Nvidia, AMD, and Taiwan Semiconductor, analysts are now asking whether the returns on hundreds of billions of dollars in capital expenditure are sustainable. The market is no longer willing to take executives at their word; it wants proof that the demand is translating into profitable, long-term growth, not just a spending spree.
This skepticism is rooted partly in historical parallels. During the dot-com boom of the late 1990s, companies also reported surging demand for internet infrastructure. The result was a bubble that eventually burst, leaving investors with enormous losses. Today, the differences are notable: the companies driving the AI rally are extraordinarily profitable, and the demand signals from suppliers like Lumentum are grounded in real procurement contracts, not speculative hype. But the bears note that market concentration now exceeds 2000 levels, with a handful of megacap tech stocks accounting for an outsized share of total market returns. That concentration leaves the entire sector vulnerable to any slowdown in AI spending.
The bull case: real demand backed by profitable companies
Bulls point to the hard numbers. Unlike the dot-com era, AI infrastructure spending is happening alongside massive profits from the same companies that are buying the hardware. Microsoft, Amazon, Google, and Meta are generating tens of billions in free cash flow each year, providing a solid foundation for capital outlays. SoftBank’s Masayoshi Son has gone as far as to say that calling AI a bubble is an insult. From his perspective, this is a generational infrastructure project — akin to building out the electric grid or the interstate highway system — not a mania. The comparison is apt: early-stage infrastructure investments often appear overdone until the network effects kick in.
Moreover, the use cases are proliferating. AI is being deployed in healthcare, drug discovery, autonomous vehicles, finance, cybersecurity, and manufacturing. Each of these verticals requires specialized compute, and the total addressable market is in the trillions of dollars. Executives like Gelsinger see not just a technology shift but a fundamental restructuring of the global economy. Energy, he argues, will eventually become the binding constraint — an insight that suggests the current demand will only grow more intense as energy solutions improve.
The bear case: overpriced markets and unproven returns
The bears do not really dispute the existence of demand. They dispute the price. At current valuations, the AI sector is pricing in decades of uninterrupted growth. Any hiccup — a trade war, a regulatory clampdown, a sudden shift in AI architecture — could trigger a sharp correction. The returns on the massive capex are still unproven. While companies like Microsoft and Google are embedding AI into their products, it remains unclear how much incremental revenue those features will generate. Some analysts estimate that AI could add just a few percent to revenue growth over the next five years, far below the expectations baked into stock prices.
Market concentration is another concern. The PHLX chip index’s 60% gain this year means that a handful of stocks — Nvidia, AMD, Broadcom, TSMC — are carrying the entire index. If one of these companies falters, the ripple effects could be severe. The dot-com collapse was similarly concentrated in a few high-flying sectors. The structural differences are real, but so is the risk of overvaluation.
The constraint nobody can buy their way out of: energy
Gelsinger’s caveat about energy is perhaps the most important one to unpack. If energy becomes the binding limit on AI compute, then chips are no longer the bottleneck. The entire sector’s valuation rests on infrastructure it does not control: power grids, renewable energy farms, nuclear reactors, and transmission lines. These assets require years of planning, regulatory approvals, and construction. They cannot be ramped up with a quarterly earnings call or a new investment round.
That reality has started to attract capital. Nvidia has backed startups like Crusoe Energy that aim to power data centers with natural gas or nuclear energy. Other firms are exploring small modular reactors (SMRs) and advanced geothermal systems. Amazon and Google have signed power purchase agreements for renewable energy, but the scale needed is enormous. A single large AI data center can consume as much electricity as a small town, and the growth trajectory suggests that by 2030, data centers could account for 10% of global electricity demand — up from about 2% today.
The infrastructure challenge is not just technical but also geopolitical. Energy security is a national priority, and countries may restrict the availability of power for AI compute if it threatens other sectors. Meanwhile, planning permission for new power plants and transmission lines is often delayed by local opposition or bureaucratic inertia. These are constraints that cannot be solved by throwing money at them quickly; they require decades of systemic investment and policy alignment.
Capital is chasing that gap. Venture funding for AI infrastructure startups hit record highs in 2024, with companies like Crusoe, CoreWeave, and Lambda Labs raising billions to build out compute capacity. But even with aggressive investment, the timeline to bring new power online is measured in years, not quarters. That disconnect between financial markets’ short-term horizon and the physical reality of energy infrastructure is likely to remain a source of volatility.
The awkward truth under the current volatility is that the industry has convinced itself demand is infinite. It may well be right. But the electricity is finite, the share prices are not, and the market is beginning to price in the gap between promise and physical reality. Whether that gap leads to a correction or a renewed rally depends on how quickly the energy constraints can be resolved — a question that no amount of executive confidence can answer alone.