How to find AI stocks early, and check they are real

To find AI stocks early, look where AI revenue is real but still a minority of the business: companies that sell cooling, power, grid gear, chip-making tools and network parts to the people building data centers. Then check each one in its filings before the market's label catches up. You need a browser and about twenty minutes per company.

The catch is that "early" and "not really an AI stock" look the same on a chart. Both are companies with a small AI story inside a bigger business. The difference is whether a customer is paying for the AI part. Most of this guide is about telling those two apart. Supercycle's screener, leaderboard and stock pages are the worked example throughout.

What does "early" mean for an AI stock?

There are two kinds of early, and they need different checks.

The first is a company inside a layer the market tends to price after the chips. Spending on AI flows in a rough order: chips and servers first, then the buildings, power, cooling and wiring that let them run. The AI Buildout Tracker follows that money from hyperscaler capex to individual builds and power deals. The layers downstream of the chips, such as Cooling & Thermal (10 tracked stocks) and Utilities & Grid (42), hold companies that were valued as industrials long before data centers became their fastest-growing customer.

The second is a company where AI is real but not yet the main driver. Supercycle's AI Exposure Score puts every tracked company in a tier, and two of those tiers describe this stage.

Tier	Score	What it says about the business	Early candidates?
Pure Play	90 to 100	The business exists because of AI demand	Rarely. The market already prices these as AI stocks
Core	70 to 89	AI is the dominant growth engine	Sometimes, in smaller layers
Strategic	50 to 69	AI is a top-two driver beside a large existing business	Often. The AI part can be growing faster than the whole
Growing	30 to 49	Meaningful, growing exposure, not yet the main driver	Most often. This is the stage "early" describes
Peripheral	0 to 29	Real but minority exposure, or optionality	Watch list only. Most never move up

Step 1: Pick a layer downstream of the chips

1. Open all buildout layers and scan the layers that sell into data center construction: Power & Energy, Utilities & Grid, Cooling & Thermal, Chip Equipment & Foundries and Networking & Optics.
2. Open the Buildout Tracker and read the power deals and major builds. Each row names who is building and who is supplying. Suppliers that keep appearing are worth a closer look.

You should now have one or two layers to work through and a few company names from the tracker.

Step 2: Filter the screener to the middle tiers

1. Open the screener.
2. Under "AI Category", pick the layer from step 1, for example Cooling & Thermal.
3. Under "AI Exposure", select Growing and Strategic. Leave Pure Play and Core off for now.
4. Optional: set "Min Est. AI Revenue %" to a low floor to drop names where AI is still only a plan.

You should now see a short list. In Cooling & Thermal it includes industrial names such as Dover (DOV, Growing), Carrier (CARR, Growing) and Modine (MOD, Strategic).

Step 3: Read the rationale, not just the score

Every company on Supercycle carries a written rationale beside its score, on the leaderboard and on the stock page's AI Profile tab. The rationale is where "early" either holds up or falls apart.

Take Dover, scored 40. Its rationale names specific products: quick-disconnect couplings used in direct-to-chip liquid cooling loops, and heat exchangers. That is a part a data center buyer orders, inside a company most people know for pumps and industrial equipment. Compare a rationale that only says a company "plans to use AI" or "is exploring" a market. The first describes a customer. The second describes a press release.

You should now be able to sort your list into two piles: names whose rationale names a product and a buyer, and names whose rationale describes a plan.

Step 4: Compare the score with the revenue estimate

1. Open a stock page and go to the AI Profile tab.
2. Read the Estimated AI Revenue figure next to the score. It is Supercycle's estimate of the share of revenue tied to AI, built from segment disclosures and product mix.
3. Look for a gap. A company with a small revenue share but a fast-growing data center line can score in the Strategic tier while most of its sales still come from other customers. Trane (TT, Strategic) is one: data center chillers are a growth driver its rationale names, inside a much larger heating and cooling business. A gap the other way, a high revenue share with a middling score, usually means the score is weighing something else, such as an early-stage company or a business the market could serve from many suppliers. The rationale says which.

You should now know, for each name, whether the AI part is already a large slice of revenue or a small slice that is growing.

Step 5: Confirm it in the filings

The score is a judgment. The filing is the record. Two places in a 10-K settle most cases, and the SEC's guide to reading one explains the layout.

1. Search the company in EDGAR full-text search with a phrase like "data center", limited to 10-K and 10-Q forms.
2. Read Item 1, the business section. Does it name data centers or AI customers as an end market, or only mention AI as a trend?
3. Find the segment table in the financial statements. If the company reports a segment or end market that includes data centers, you can see its growth rate against the rest of the business.
4. Read the most recent 10-Q's discussion of results for orders or backlog tied to data centers.

You should now have a number from the company itself, or a clear sign that the company does not break it out. Both are useful. "The filing does not say" is a real answer.

How do you spot a fake AI stock?

The opposite of early is AI washing: AI in the story, not in the revenue. The glossary entry has the short test, which is to ask what the customer buys. Regulators use the term too. In March 2024 the SEC charged two investment advisers with making false and misleading statements about their use of AI.

Warning signs while you work through steps 3 to 5:

• AI appears in the press releases and the call transcript but not in the segment table or the business section.
• No named customer, product or order is tied to the AI claim.
• The company changed its name or description to include AI, while its revenue mix stayed the same.
• The rationale on Supercycle uses words like "plans", "optionality" or "exploring". Many Peripheral-tier names read this way, and that is why they are Peripheral.

Step 6: Keep the short list and watch what changes

1. Add the names that passed step 5 to your watchlist.
2. Recheck each quarter. The things that move a company from Growing to Strategic are new segment disclosures, a named hyperscaler or data center customer, and data center orders growing faster than the rest of the backlog.
3. Watch the Buildout Tracker for new power deals and builds, since that is where demand for these layers starts.

You should now have a short list of companies with checked AI revenue and a reason to look at each one again next quarter.

What does the score not tell you?

The AI Exposure Score measures how tied a business is to the buildout. It does not say whether the shares are cheap, and it says nothing about timing. A company can be early in its AI exposure and late in its valuation, because the market priced the story before the revenue arrived. The stock page's Financials tab shows the valuation ratios, and those are a separate question from exposure.

The score is also an editorial judgment, disclosed as one in the methodology. Two readers can look at the same rationale and score it differently. That is why step 5 exists. The filing is the part nobody has to take on trust.

The next thing to watch is the coming round of hyperscaler earnings. Each one restates how much the largest buyers plan to spend, and the layers in step 1 are where that money lands after the chips.