The AI revolution is real. That has never guaranteed the stocks were worth any price. Here is every AI, cloud, and quantum name we cover — with a fair value estimate, an honest method label, and the discipline to say when a number cannot be trusted.
Of the 75 AI, cloud, and quantum stocks we cover, 53 have real earnings our value model can price. 28 of those currently trade more than 15% above our fair value estimate, while 4 trade at a 10%+ margin of safety below it. The remaining 21 are pre-profit companies where no earnings-based intrinsic value exists — for those we show a clearly-labeled analyst consensus anchor instead of pretending our model has an answer.
That split is the entire thesis of this page: "AI" is not one asset class. A chip fabricator earning $40B a year and a pre-revenue quantum startup are different species that happen to share a headline. Value investing does not tell you to avoid AI — it tells you to price the two species differently, and to know which one you are holding.
1. The earners. NVIDIA, TSMC, Broadcom, Microsoft, Alphabet, Meta: these companies convert the AI buildout into audited earnings today. They can be valued the classic way — our Predicted EPS × Fair PE formula, built from SEC EDGAR filings, with a 15% value discount and a hard PE cap of 40. When one of these trades below fair value, that is a statement the numbers can defend.
2. The promises. CoreWeave, IonQ, D-Wave, C3.ai: real technology, negative earnings. An earnings multiple on a loss has no meaning, so for these we anchor to the analyst consensus target and stamp the method on the page — "Watch — Analyst Anchor," never "Strong Buy." Where analyst coverage is too thin to anchor, we print N/A. A value site that hands out buy zones on pre-profit stocks is not doing value investing.
Benjamin Graham's line from 1934 still cuts cleanly: an operation promising safety of principal on thorough analysis is an investment; everything else is speculation. Both can make money. Only one should be sized like a conviction position.
Memory is where the 2026 AI trade has run hottest. High-bandwidth memory (HBM) is effectively a three-player game between SK Hynix, Micron, and Samsung — and SK Hynix, the current HBM leader, became directly investable for US investors in July 2026 when its $26.5B Nasdaq ADR listing (SKHY) set the record for the largest US share sale by a foreign company. Two of the three HBM players are now on this page; Samsung remains Seoul-only and outside our coverage universe. Storage has caught the same updraft: training data has to live somewhere, and Seagate, Western Digital, and SanDisk have repriced violently on AI datacenter demand — which makes the margin-of-safety column below worth a long look before chasing.
Every GPU cluster is only as fast as the links between its racks, which is why optical transceivers and interconnect have been among the most violent movers of the entire AI trade. The US-listed core is here: Coherent and Lumentum on modules and lasers, Fabrinet assembling much of the industry's output, Ciena on the systems layer, Applied Optoelectronics on datacenter transceivers, and Credo and Astera Labs on the copper-and-retimer side of the connectivity stack. Worth knowing: the two highest-volume 800G module makers — Innolight and Eoptolink — trade only in Shanghai and Shenzhen, outside our SEC EDGAR coverage universe, so the US names above are where that demand shows up in filings we can audit.
The 2026 twist on "sell shovels" is that the scarcest shovel turned out to be electricity. This group spans the whole power answer: small-modular-reactor developers with hyperscaler deals (Oklo with Meta, NuScale with the TVA), fuel-cell power that is already profitable (Bloom Energy), the uranium fuel cycle (Cameco, Centrus), nuclear-heavy utilities signing long-term datacenter contracts (Vistra, Constellation), grid and turbine capacity (GE Vernova, Eaton, nVent), and the cooling that keeps racks alive (Vertiv, Modine). Note the split personality: the utilities and electrical names are profitable businesses our core model can price, while the SMR developers are pre-revenue stories that get an analyst anchor at best — the gap between those two method labels is itself the risk lesson.
Someone has to physically build the racks: Celestica, Jabil, and Flex assemble AI servers and networking gear at scale, TTM Technologies supplies the high-layer-count PCBs underneath the accelerators, and Symbotic applies the same AI wave to warehouse robotics. These are thin-margin, real-earnings industrial businesses — exactly the kind our EPS-based model prices without excuses, which makes their margin-of-safety readings unusually trustworthy for an AI-adjacent group.
Every generation gets one technology so obviously transformative that price discipline starts to feel like cowardice. The railroads had 1873. Radio had 1929. The Nifty Fifty — Coca-Cola, McDonald's, Disney, "one-decision stocks" you supposedly never needed to sell — peaked in 1972 at 40-80× earnings, then fell 60-80% over the next two years while the underlying businesses kept growing. Investors who bought the companies were right about the companies and still lost a decade.
The dot-com era repeated the pattern with better graphics. Cisco was the literal backbone of the internet, and the internet delivered on every promise — yet a share bought at the March 2000 peak needed more than twenty years to break even. Amazon, the ultimate survivor, drew down 94% before compounding its way into history. Being right about the technology is not the same skill as paying the right price for it.
The value investor's answer to the AI boom is therefore not abstinence. It is the same three questions as always: Does this company earn real money? Does a durable moat protect those earnings? And does today's price leave a margin of safety if the future arrives a few years late? When all three answers are yes — as they were for Meta at $90 in 2022 — a technology stock is a value stock. When the answers are "not yet," "unclear," and "no," you are welcome to speculate. Just call it that, and size it that way.
It depends on which kind. Profitable AI infrastructure companies (NVIDIA, TSMC, Broadcom, Microsoft) can be valued on earnings, and several trade within a defensible range of our Predicted EPS × Fair PE estimate. Pre-profit AI and quantum names (CoreWeave, IonQ, D-Wave) have no meaningful earnings to value — for those, any fair value number is an analyst-consensus anchor, not a value-model output, and we label it as such. History suggests the technology succeeding and the stock being a good buy at today's price are two different questions.
For profitable AI companies we use the same transparent formula as the rest of the site: Predicted EPS × Fair PE, where Fair PE is the stock's own trailing multiple with a 15% value discount, capped at 40. For loss-making AI and quantum companies the formula has no defined answer, so we anchor to the analyst consensus target where coverage is deep enough (3+ analysts) and clearly mark the method — or show N/A rather than invent a number.
Graham drew a hard line between investment and speculation: an investment promises safety of principal and an adequate return on thorough analysis; everything else is speculation. By that definition, buying a profitable chipmaker at a reasonable multiple of earnings can be an investment, while buying a pre-revenue quantum company is speculation regardless of how transformative the technology proves. Graham did not say speculation is immoral — only that you must know which one you are doing and size the position accordingly.
Twice, famously. The Nifty Fifty of 1972 — Coca-Cola, McDonald's, Disney — were genuinely great businesses whose stocks still fell 60-80% when multiples normalized, taking a decade to recover. In 2000, Cisco was the backbone of the internet and the internet kept growing; the stock still needed over 20 years to reclaim its dot-com peak. The lesson is not that AI will fail. It is that a great technology bought at the wrong price can be a poor investment for years.
Yes — when real earnings meet a depressed multiple. Meta in late 2022 traded near $90 at a single-digit forward PE while its core business kept generating tens of billions in free cash flow; that was a value investment in a technology company. The test is not the sector. It is whether audited earnings power justifies the price with a margin of safety.
How our Predicted EPS × Fair PE model works, and every stock's margin of safety.
Stocks trading above our fair value estimate — including several AI names.
Durable competitive advantages — the quality filter value investors apply before price.
Every formula documented, including when we refuse to publish a number.