Compute Is the Scarcest Input
The market has hit all-time highs without the Magnificent Seven, earning that group the nickname "Lag Seven." The reason traces to a single shift: compute has become the scarcest input in the economy. The market now rewards companies that control more of their own AI economics, either by turning CapEx into lower cost per token or by capturing higher margins. When a company looks like a vertically integrated profit engine, its CapEx gets rewarded. When it cannot communicate that story, CapEx gets punished as an unproven cost line item.
Before 2025, and arguably into 2026, the whole Mag Seven moved together. That changed when inference began booming and inflecting. At that point the market started separating winners from laggards, stopped worrying about opportunity cost, and rushed toward chips. That is why semiconductors performed so well. There will be a circle-back moment toward the Mag Seven names, because their business models have a durability that benefits them most once AI gets cheaper and cheaper, and those signs are already showing.
Name-Specific, Not Sector-Wide
Are investors too impatient, or underestimating the size of the AI opportunity? The move away is name-specific. Google is the clearest example of what happens when the market rewards you for owning the full AI stack: custom chips, frontier models, consumer distribution, and enterprise monetization. Google has the lowest internal cost per token, a major advantage it monetizes through Gemini, which trains end-to-end on TPUs, while Google Cloud sells TPU capacity to customers and the ecosystem compounds.
The rate of change for many of these companies over the past couple of months has been exponential, so any sign of a top gets scrutinized heavily. Samsung is the latest small data point pushing the market to question the length of this AI build-out. Yet every three months the market gets reminded how big the AI pie is, and in a couple of weeks Big Tech earnings will remind it again that momentum continues and the AI trade is still humming. Supply matching demand for AI remains years and years away.
Check Writers and Check Takers
The check writers must sustain their CapEx for the check takers to keep benefiting. Suppose a headline hit that Microsoft would not keep up its recent spending. That could actually help Microsoft, since people are worried about its margins. The market would reward the companies that own the ecosystems and distribution channels: Microsoft, Meta, Google, and Nvidia.
Nvidia has become a victim of how fickle the market is, rewarding the rate of change of a business rather than its true quality. Memory, networking, and other suppliers have held up because those names saw sharper upward revisions from lower bases, while the higher-quality businesses everyone believes in have been left for dead. If CapEx growth slows, expect a revisit toward the original winners, while the new hot layers of the ecosystem likely start revising lower. That would be healthy, because today's valuations across the stack already price in a lot of multi-year demand.
A clear dislocation exists. Micron and Nvidia trade below market multiples while Intel and many optical names trade significantly higher, which makes no sense. One camp assumes the AI cycle runs through the rest of the decade; the other camp assumes peak earnings. The "bottleneck bros," a phrase coined by Gavin Baker, probably need to cool off while the real winners benefit again.
Ranking the Mag Seven: Meta on Top
Ranked by AI upside, valuation, and execution risk, Meta comes out on top. Its core ad business has already proven it will benefit from AI, but the market is not yet convinced Meta's massive AI infrastructure spend can produce a clean new revenue stream beyond better ads. Zuckerberg appears to see this: Meta is building more of its own AI compute stack and may lease out compute, given how strong the market for it is. That compute business will not be large, but it creates a new revenue layer and helps investors grasp how massive Meta's infrastructure build-out could become if it monetizes beyond ads.
Meta also has the strongest monetization engine in the group through advertising. Its new video app uses all internal models rather than external ones for its latest AI product, more proof that AI infrastructure spend is creating revenue beyond ads. Meta trades at 18 times earnings while the S&P sits around 20 or 21. There is no world where Meta should trade below the market multiple given its opportunity and the distribution eyeballs it owns.
Tapping Debt and Equity Markets
Amazon and others are tapping debt markets, and some equity markets, to raise extra cash. Is that a concern? Google opened the floodgates a couple of weeks ago as the first Mag 7 name to go that route, and the market did not punish the stock. Expect more of these massive companies to leverage financial markets to optimize cash for data centers. That is fine, because right now they can mark up their CapEx at a return higher than 5%. If they left the cash idle, it would only earn 4 to 5% on the sidelines, and the market for their AI products grows with every data center they build.
The worry sits with the weaker players. CoreWeave is taking on 8% interest loans to fund its data center build-out. Amazon, Google, and Microsoft have the cash to self-fund but choose not to, because carrying all the risk yourself is poor financial engineering. CoreWeave lacks that luxury and must absorb the 8% rate. That single case is not alarming, but if more bad agents pile in, and the numbers reach hundreds of billions of dollars, it could turn systemic. We are nowhere close to that right now.
