SolarEdge's 2029 Target Is a Bet on an AI Customer It Hasn't Signed

SolarEdge walked into its September 10 investor day with the two things a speculative solar stock wants: a big number and a famous name. Management set a target of $2.4 billion in revenue by 2029, and it leaned on Nvidia — the company published a joint paper with the chipmaker on how to protect the high-voltage DC power that AI data centers are moving toward. The stock had already been on a wild ride: over the past year it has swung between about $28 and $81, it is up roughly 22% this year, and it is down about a third over the last four months to sit near $35, at a market value of about $2.2 billion.

The seam in the story showed up in that same presentation.

The target and the caveat, side by side

In the deck where SolarEdge sold the $2.4 billion future, management attached the one sentence that defines the whole case: it is in talks with hyperscalers and neoclouds, but it has no concrete customer order or agreement for the AI data-center business. Management described it as "early stage."

That is a polite way of saying the AI part of the plan has no backlog. The Nvidia item is a white paper — a technical framework offered for the industry to evaluate, not a design win, not a contract, not a dollar of revenue. The related work with Infineon on solid-state transformers is a development collaboration, also unsigned. None of it, as of the day, is a customer committed to pay.

What the AI bet actually is

For the reader who does not follow data-center power: the thesis is that as AI racks get denser, they will stop running AC power all the way to each server and switch to a high-voltage DC "powertrain" — 800 volts carried from the medium-voltage grid down to the rack. SolarEdge's version hinges on a solid-state transformer, a device that uses power electronics to do, faster and smaller, what a traditional iron transformer does. It is a real and plausible technology, and one SolarEdge is credibly positioned for. The question the market keeps voting on is whether "plausible and positioned" is the same as "being bought." By the company's own words, no one has committed to buy it yet.

The turnaround is real, with two asterisks

This is not a vaporware story, and that matters. The core solar business is genuinely turning around. In the second quarter, revenue rose 19.6% year over year to $346.2 million, and gross margin expanded for a sixth straight quarter to 28.6% on a non-GAAP basis, from 13.1% a year earlier. The company got back to non-GAAP operating profitability for the first time since mid-2023, and it closed the quarter with about $265 million of cash and investments net of debt — a clean balance sheet, not a stretched one.

Two asterisks. First, that Q2 margin included a one-time $13.3 million benefit tied to U.S. tariff (IEEPA) matters, worth roughly 3.8 points of revenue. Strip it out and the GAAP margin was still around 23.7%, against 11.1% a year ago — so the recovery is real and not just the tariff, but it is flattered. Second, the next quarter is guided down: SolarEdge sees Q3 revenue of $310 million to $340 million, a midpoint about 6% below Q2, with non-GAAP gross margin falling back to 22% to 26%, on persistent weakness in U.S. residential solar. The trend that carried the stock to its high is, right now, pointing down.

Priced for 2029, before the AI revenue exists

Put a price on it. SolarEdge trades at about 1.6 times trailing sales — roughly $2.2 billion of market value against about $1.3 billion of revenue over the last four quarters — and it is still loss-making on a GAAP basis, so there is no clean forward price-to-earnings to anchor to. That is not an accident. The forward earnings are the disputed variable.

Look at the bull's own 2029 numbers: $2.4 billion of revenue and $360 million of EBIT. Getting from today's roughly $1.3 billion to $2.4 billion in three years needs about $1.1 billion of new revenue, and a solar business that is guiding down next quarter and leaning on a soft U.S. residential market cannot supply it alone. That means the 2029 target is, in large part, a forecast that the AI data-center power business scales from zero to a very large share of the company in about three years. Value that $360 million of 2029 EBIT at a plain 6x and you are at about $2.2 billion — which is where the stock already trades, three years out. The market is paying today for a big slice of the 2029 outcome before any of the AI revenue exists. And $2.4 billion is the bull's number, not even consensus; some models land closer to $1.7 billion, which still needs the AI story to show up.

The one line that changes the math

So what does this mean for you? It is not that the AI idea is wrong. It is that the evidence is not there yet, and the investor day's real use is that it told you, in the company's own words, that the gap is still open. What you are looking at is a solar-recovery business with a genuine turnaround, layered with an AI power option that is priced into the multiple but not earned in the backlog.

The thing that closes the gap is specific and observable: a named hyperscaler or neocloud signing an actual order or design win for the 800-volt powertrain. Until that name shows up, the $2.4 billion figure is a target to track, not a number to price in — and the stock trades as though someone already believes it.