When the Market Forgets the Business
There are moments in financial markets when prices drift far from the fundamentals of the companies they supposedly represent. A sharp selloff at the end of one week, a strong recovery the next day, a session that opens higher, plunges by midday, and then claws its way back before the close — these are the rhythms of a market caught between two impulses. One is the impulse of the trader, watching wiggly lines on a screen and reacting to momentum. The other is the impulse of the investor, attempting to attach themselves to durable, long-term business trends.
Benjamin Graham's old distinction is worth recalling here: in the short run, the market is a voting machine, a popularity contest driven by sentiment and crowd psychology; in the long run, it is a weighing machine, measuring the actual substance of a business. Right now, both forces are visibly at war. Much of the recent volatility in technology stocks appears to have little to do with any single headline. Instead, it reflects investors trimming crowded positions as a heavy calendar of initial public offerings approaches. When capital reshuffles itself for reasons unrelated to the health of individual companies, the result is dislocation — and dislocation, for the patient investor, is opportunity.
Stocks on Sale
The arithmetic of these giant IPOs is itself a source of opportunity. When enormous offerings come to market, money moves around in ways disconnected from the underlying value of existing companies. Positions that were attractive at a thousand dollars do not become less attractive at nine hundred; if anything, the logic runs the other way. Volatility puts quality names on sale, allowing investors to acquire the positions they want at prices better than they could find on calmer days.
This is not a blanket endorsement of buying every dip. The honest assessment is more nuanced: looking back at today's prices, some will prove to have been moments to sell, and others moments to buy. The discipline lies in distinguishing between the two — in knowing the businesses well enough to tell which decline is a bargain and which is a warning.
The One Thing Every Company Is Talking About
The most striking signal from across the technology landscape is not about pricing, and not even about market share. Speak with company after company — in Silicon Valley and far beyond — and the conversation returns relentlessly to one theme: supply, and the scarcity of the ability to deliver what customers want. The question on every executive's mind is not "What should we charge?" or "How do we win share?" It is "Can we build it? Can we build more? What can we do to build more?"
This scarcity explains a wave of deals that might otherwise look strange. Consider the pattern of investments rippling through the supply chain: a major cloud and retail giant putting money into a specialty-glass and optical-components maker; a dominant chip designer investing in optical-networking companies like Coherent and Lumentum; and a tangle of arrangements forming ahead of high-profile space and AI offerings. Most telling of all are the arrangements between direct competitors. One leading AI lab and a major search company — both building large language models that compete head-to-head with a rival's chatbot — have moved to rent computing capacity that might otherwise have gone to that very rival. When fierce competitors are willing to do business with one another simply to secure access to compute, it reveals just how acute the hunger for Nvidia's GPUs has become.
The Circular Handshake
These cross-investments and mutual rentals form what can fairly be called a series of circular handshakes — companies financing, supplying, and renting from one another in overlapping webs amid a simultaneous AI arms race and AI buildout. The skeptic has a ready interpretation: when a company that is also an investor in a venture turns around and pays that venture for services, perhaps it is merely propping up the deal rather than responding to genuine need.
But the on-the-ground evidence cuts against pure cynicism. The recurring message from direct channel checks is blunt: everyone needs compute right now. The demand is real even where the financial relationships are tangled.
It is also worth noting how cautious the parties remain. These rental contracts are riddled with escape clauses — one memorable description held that there are more easy outs in such a contract than in a child's tee-ball game. The companies clearly want compute, but they also want optionality, preserving the freedom to walk away as conditions change. The demand is genuine; the commitment is hedged.
A Cautionary Tale in the Middle of a Boom
Even amid this overwhelming demand, the AI buildout offers a sobering lesson about how hard it is to succeed. Some of the most impressive data centers ever constructed — vast facilities built in Mississippi and, before that, in Memphis, Tennessee — were not being used to their full capacity. They had been overbuilt relative to the actual needs of the model they were meant to serve, in part because that model underperformed on a range of comparable benchmark tests against its competitors.
The result is a paradox. A company can possess world-class infrastructure and still struggle to reach the metrics its rivals achieve, leaving it with surplus compute. That it can rent out the excess is itself proof of how ferocious overall demand is. But the underlying story is a warning: in a world where powerful new models are released almost constantly — including newly launched models that are performing exceptionally well on key metrics — owning the hardware is no guarantee of winning the race. Compute is necessary, but it is not sufficient. The quality of the model running on that compute is what ultimately decides who thrives.
The Signal to Watch: GPU Rental Pricing
If demand currently outstrips supply across the board, what would be the first sign that the dynamic is shifting? The single most informative variable is the pricing on GPU rentals. Data on what one major company is willing to pay another for compute access offers a real-time read on how tight the market remains. As long as rental prices hold firm and well-financed players keep competing for capacity, the scarcity story stands. A meaningful softening in those prices would be the earliest warning that the balance between supply and demand is beginning to turn.
Oracle and the Primacy of the Backlog
Few companies illustrate the transformation more vividly than Oracle. It is natural to think of it as a software company — and it is one — but it has also become one of the preeminent builders of data centers in America. The number that has moved its stock most powerfully over the past year and a half is not a traditional software metric at all. It is the figure for Remaining Performance Obligations, or RPO: the contracted, committed future revenue that has yet to be delivered. That backlog is, in essence, a measure of data-center and AI demand locked in by binding agreements.
Committed contracts and backlogs of this kind serve as a crucial differentiator. They convert the abstract enthusiasm of the moment into concrete, contractually obligated business. In direct meetings with the company's management, the conversation has centered almost entirely on data centers; software was barely a concern by comparison, because what is happening in the data-center business is simply more consequential right now.
The company is building at remarkable speed, and the deals are evolving in real time. It is striking special power-supply agreements that differ from one data center to the next, and it is going to unusual lengths to keep host communities satisfied — structuring arrangements in ways that look very different from facility to facility precisely because it is trying to make everyone along the way comfortable. This is, quite literally, a land rush: a scramble to acquire sites and energy and build capacity as fast as physically possible.
Why One Earnings Call Tells a Larger Story
This is why a single company's results can illuminate an entire industry. Its earnings reveal not only how fast its own data centers are being built and how much of that enormous RPO demand it is satisfying, but also where the buildout is headed next. And because so much of that demand originates with the developers of large language models, the report becomes a window onto the broader ecosystem — the health of competing neocloud providers, the trajectory of LLM development, and the prospects of the wave of massive IPOs clustered around AI labs, from one xAI-style venture to leading research labs to other major model developers.
In an environment where the loudest message from every corner of the industry is "we cannot build fast enough," the companies supplying the picks and shovels of the AI age — the chips, the optical components, the power agreements, and above all the data centers themselves — sit at the center of the story. The volatility of their share prices reflects the market's voting machine at work. Their backlogs, their build rates, and the price of compute reflect the weighing machine. For investors, the task is to keep watching the latter while everyone else is mesmerized by the former.