AI Power Duel: Hedge Funds Piled Into AEIS While Baron Bailed, Then the Stock Fell 21%

AI Power Duel: Hedge Funds Piled Into AEIS While Baron Bailed, Then the Stock Fell 21%

Both camps agree on the sequence. Advanced Energy Industries (AEIS) sells precision power conversion — the electronics that feed plasma into chipmaking tools, and the AC-DC and DC-DC converters that power the server racks running AI. On Aug. 3 it reported a quarter that beat and raised: revenue up 30% year over year to $574 million, non-GAAP earnings up 83% to $2.74 a share, full-year growth guidance lifted to the low-to-mid 30s. The stock ran to a 52-week high near $397, then dropped about 21% over five sessions to roughly $275 — down another 4% in Tuesday's early trade, still up about 31% for the year.

Now the camps split on what that means. The bulls see a beat-and-raise growth machine on sale, cheaper than its data-center-power peers, with a genuine 2028 act (the industry move to 800-volt racks) worth owning early. The bears see a stock at about 50 times trailing GAAP earnings whose headline profit runs far ahead of its cash, and whose prized next act produces no meaningful revenue until 2028.

And then there is the ownership snapshot everyone is arguing about, which is stranger than either side wants to admit.

The 13F memo everybody is citing

Holdings for the quarter ended June 30 — filed in mid-August — show the number of hedge-fund portfolios holding AEIS jumping from 45 to 70. D.E. Shaw built a stake of roughly 586,000 shares, AQR grew its position 135%, Millennium 26%. The same filing window shows Baron Capital's Discovery Fund, a long-time holder, selling about 60% of its shares at roughly $375 apiece. Baron's stated logic: the stake had grown to nearly 4% of the fund, the price had approached its long-term target, and it would consider adding again at lower prices — while noting that most of the 800-volt opportunity is still future revenue.

Notice what this snapshot does not tell you. The hedge-fund "flood" is dated June 30, before the Q2 report and before the run to $397. Those buyers paid roughly $250-$350; at $275 they are underwater. Baron sold into the spike at a price today's market no longer offers, and kept a seat. That is a fact about entry prices and about who is insuring whom, not a verdict on the company.

Shared facts

• As of Tuesday's trade: stock ~$275, market cap ~$11 billion, about 40 million shares, 52-week range roughly $144-$397.
• Q2 (period ended June 30, reported Aug. 3, 2026): revenue $574.1 million, +30% year over year and above the high end of guidance; non-GAAP EPS $2.74 (+83%); GAAP EPS $1.29; record 21.9% non-GAAP operating margin; 41.9% gross margin.
• Segments: semiconductor $278 million (record, +33% year over year); data center computing $192 million (+35% year over year but down 1% sequentially); industrial & medical $80 million; telecom & networking $24 million.
• Guides: Q3 revenue about $640 million with non-GAAP EPS of $3.00; full-year 2026 revenue growth raised to the low-to-mid 30s; data center computing expected to grow at least 50% for the year.
• Trailing tally (calculations from reported figures): about 32 times trailing non-GAAP EPS of $8.51, about 50 times trailing GAAP, about 125 times trailing free cash flow of roughly $87 million (~$2.20 a share) — with inventory at 145 days and about $1.4 billion of cash on hand.

Round 1 — Is the growth a machine or a cycle in a bull costume?

Bull's best punch: two engines. Semiconductor revenue printed a record $278 million, and AI chip capacity is pulling wafer-fab spending with it. Data center computing more than doubled in Q1 and is guided to grow at least 50% for the full year, with management describing "second wave" hyperscalers ramping to volume next year and a more bullish outlook for the second half of 2026 and 2027 as customers resolve supply-chain issues. If one engine hiccups, the other carries the quarter. This is an acceleration, not a turnaround.

Bear's best answer: semiconductors are a cycle with a startlingly short memory. AEIS was fighting through a down cycle in 2024-25; this is the upswing, and upswings end. The data just showed the tell: data center computing revenue fell 1% sequentially in Q2 even as everything else surged, so the "at least 50%" full-year number quietly requires a sharp second-half reacceleration — a promise, not a print. And a handful of hyperscalers moves that segment.

The measured evidence runs the bull's way this quarter. Winner: bull on Round 1, with the bear's duration doubt noted rather than dissolved.

Round 2 — What you actually own: $2.74, $1.29, or $2.20?

Bull: margins are real. Gross margin rose 3.8 points, operating income hit a record, and the quarter produced a record $86 million of operating cash. The balance sheet carries about $1.4 billion of cash and a stated appetite for acquisitions. This is a profitable, self-funding company whose numbers improve every quarter.

Bear: in that same quarter, non-GAAP EPS was $2.74 and GAAP EPS was $1.29 — GAAP came in at less than half the headline figure, with acquisition-related amortization and adjustments eating the difference on a portfolio built through deals. The cash confirmation is thin: trailing free cash flow is under $2.20 a share against a $275 stock. Inventory is parked at 145 days — a bet the boom continues; if a customer defers, that bet turns into discounting and write-downs, which is how "record EPS" becomes a "record charge."

Bull's answer: free cash flow lags because the company is stockpiling components to ship booked orders; the cash arrives as the ramp ships, and Q2's record operating cash flow is the early proof.

Bear's answer: that is the thesis, not the evidence. The bill comes due in cash, and the cash is late.

The gap between the $2.74 headline and the $1.29 GAAP number, and the thin trailing cash flow, are the two most important figures in this file. Winner: bear on Round 2.

Round 3 — The 800-volt option, and whether you're paying 2028 prices in 2026

Bull: NVIDIA is moving data centers from today's 48-volt distribution to 800 volts so that racks can draw a megawatt or more, and Advanced Energy introduced its ADH-series 800V DC-DC converters in June. Content per rack rises, and this is a second multi-year curve beginning around 2027. You generally do not buy optionality that concrete at the price of a plain fast grower — particularly when Monolithic Power trades near 19 times sales and Vicor near 18, against AEIS's roughly 5.4. On annualized second-quarter EBITDA, this stock sits closer to 20 times. The $397 price paid for the 2028 story up front; $275 pays for less of it.

Bear: management itself dates the prize — initial 800-volt revenue late 2027, meaningful volume in 2028. So today's 30-to-50x earnings multiple must be justified by current hyperscaler and wafer-fab momentum alone, and the option is being prepaid years before its evidence arrives. AEIS is also not alone on the rack: Monolithic Power, Vicor, and Delta are chasing the same sockets at healthier margins. The bull is buying a promise delivered on a competitor's schedule.

Winner: bull on the option being genuine; the question of whether to pay for it now belongs to the price round.

Valuation round — what the price demands

Market-data feed, Aug. 25 session:

Metric	AEIS	Monolithic Power	Vicor
Price / sales	5.4x	19.3x	18.5x
EV / EBITDA	36x	62x	76x
Trailing P/E (GAAP)	50x	79x	60x

On headline multiples AEIS is the cheapest of the three data-center-power names. That is not the same as cheap: the discount is a price for thinner GAAP margins (about 11% operating on a trailing basis against the 21.9% non-GAAP record), cyclical semiconductor exposure, and weaker cash conversion.

Now reverse-engineer the stock. At roughly $275 and an $11 billion market cap, the market is paying about 50 times trailing GAAP earnings, about 32 times trailing non-GAAP, and about 125 times trailing free cash flow. For the bull's arithmetic to work, revenue keeps growing 25-30% into 2027, margins hold near the record, and the 800-volt ramp lands on schedule; if non-GAAP EPS reaches the low teens, the multiple alone falls toward the low 20s and today's price looks reasonable in hindsight. For the bear's arithmetic, only one ordinary thing has to happen: growth decelerates toward the high teens (a mild miss against a hot guide) or the multiple normalizes to the mid-20s as the cycle matures. That is a much lighter burden. And the market has just shown how fast this re-rating can run: five sessions, minus 21%.

Ruling

The business case goes to the bull; the stock call at this price goes to the bear. The company is executing superbly, the balance sheet can fund its path, and 800-volt is a real option. But at $275 the market pays roughly 50 times trailing GAAP earnings and about 125 times trailing free cash flow, and it prepays a 2028 story — which means the bull must be flawless for the arithmetic to work, while the bear needs only ordinary moderation. The burden of proof sits with the bulls. A 30% round-trip from the high leaves the stock less frothy than at $397; "less frothy" is not "cheap."

The ruling flips by the Q3 report due in early November — guided at roughly $640 million of revenue and $3.00 of non-GAAP EPS — if data center computing reaccelerates above the 50% pace, GAAP EPS starts closing the gap with the headline number, and free cash flow at least doubles. That combination would be the shakeout floor Baron's "add it lower" note implied, and the verdict changes to bull.

The bear's failing indicator, to monitor as the earliest tell: any softening of the "at least 50%" data center commitment, a miss of the Q3 guide, or a second straight sequential decline in data center revenue. Any one of those deflates the multiple on its own, no coaching from the 2028 story required.

The next evidence event is the November print. That is when "at least 50%" becomes a number instead of a promise.