Aehr Test Systems makes semiconductor burn-in equipment -- the machines that stress-test chips before they ship. The bull thesis: AI data center buildout drives demand for AEHR's FOX-XP wafer-level burn-in systems. On January 8, management guided $60-80M for the second half of fiscal 2026 and referenced a "lead AI processor customer production ramp" in early FY27. The stock is up 832% in a year. Today it filed a 10-Q. The numbers tell a different story than the stock price.

What the filing says

Revenue: $10.3M for Q3 FY26, down 44% year-over-year. Nine-month revenue $31.2M, down 31%. The prior $60-80M H2 guidance required Q3+Q4 combined to hit that range. With Q3 at $10.3M, Q4 would need $50-70M -- a 5x to 7x step-up in a single quarter. The 10-Q does not mention the guidance. It contains no forward revenue outlook.

The AI customer isn't buying. US revenue fell from $10.6M to $2.7M, down 74%. The MD&A attributes this to "lower sales of wafer-level burn-in systems and contactors to a customer providing AI applications." The Q2 call's language about a "lead AI processor customer production ramp early FY27" is absent from the filing. No updated timeline, no explanation.

The EV business is gone. Contactor revenue -- the recurring razor-blade that measures installed system utilization -- fell $21.2M year-to-date, down 66%. Customer E, which was 45% of nine-month FY25 revenue, has fallen below 10%. The installed SiC burn-in base is not being used.

Gross margins: 30.9% vs 43.8% prior year, a 12.9pp decline. Lower overhead absorption on reduced volume, mix shift to lower-margin Sonoma products, and newly disclosed tariff risk on imported components -- "not yet determined the impact."

Operating loss: $(12.9M) for nine months versus $(2.5M). Cash burn funded by a $19.6M ATM equity offering now fully exhausted. A $21.9M deferred tax asset is at risk of valuation allowance if losses continue.

China IP lost at first instance. AEHR sued Suzhou Semight for replicating its FOX technology. The Chinese court dismissed the claim in December 2025. Appeal pending; Semight counter-filed to invalidate the patents. China is 54% of Q3 revenue.

Sonoma package-level burn-in for AI is genuinely growing, contributing an incremental $6.1M year-to-date. Customer A, now 42% of Q3 revenue, is likely the Sonoma buyer. This is real AI-driven demand. But Sonoma is lower-margin, smaller-scale, and a different product than the wafer-level FOX systems that drove the stock from $7 to $63.

What the market thinks

Stock: $63.16, up 25.7% today, up 832% over the past year. Market cap: $1.99B on $40M annualized revenue (48.7x). Forward P/E: 379x. Short interest: 15.3% of float. RSI: 74.3. ATM implied vol: 120%.

Analyst mean target: $36.33 -- 43% below current. Jan 2027 options max pain: $18.

Every insider sold in the past year. CEO Posedel: $420K. Sporck: $83K. Oliphant: $143K. Danesh: $223K. Slayen: $154K. Zero open-market purchases.

Why the gap exists

The most important finding is not in the AEHR filing -- it's in what peers reported. The broader AI chip ecosystem is accelerating: AVGO AI revenue +106% YoY with a $73B backlog, MRVL data center +46% with FY27 guidance raised to $11B, AMD data center +39% with MI350 ramp pulled forward. COHU won its first AI device burn-in handler order. ONTO reported record AI packaging orders and a $240M+ volume purchase agreement.

AI chip production is ramping. It is going to COHU and ONTO for test and inspection, not to AEHR for wafer-level burn-in. AEHR's customer going dark is not a leading indicator of an AI slowdown. It appears company-specific or technology-specific -- which is a harder problem than a market cycle, because there is no rising tide to wait for.

The SiC/EV collapse, by contrast, is industry-wide and well-known: ON Semi reported 700bps underutilization charges and cut 12% of fab capacity. Wolfspeed emerged from Chapter 11 with $20M in underutilization costs and shut its Durham fab. Axcelis called SiC implant demand "muted." Onto expects power semi down 10% in 2026. This is consensus; AEHR's contactor data confirms it but doesn't add novel information.

The stock's 832% move has the characteristics of a short squeeze compounding on an AI narrative: 15.3% short interest, wildly unstable factor loadings (semiconductor beta swung from 1.0 to 2.5 across recent 60-day windows), and trailing returns that dwarf even the hottest semiconductor names (+528% excess over SMH in one year). The price is not currently a function of the business.

Risks

1. Squeeze persistence. 15.3% SI, +832% momentum, heavy call positioning. Price has already demonstrated it ignores fundamentals. The gap between filing and price could widen before it closes.
2. AI wafer customer returns. If the absence is a design respin rather than technology displacement, orders could resume in Q4 FY26 or FY27. No evidence supports this, but the filing doesn't explain why the customer stopped buying. The gap is open in both directions.
3. Sonoma scales faster than expected. Package-level AI burn-in is real and growing. If it accelerates, AEHR has a viable business -- smaller and lower-margin than the wafer-level thesis implies, but not zero.
4. Semiconductor beta. AEHR currently moves 1.8-2.5x with SMH. A broad semiconductor rally lifts the stock regardless of company-specific fundamentals. A selloff amplifies on the way down.

Catalysts

• Q4 FY26 earnings (~July 2026): H2 revenue totals will be measurable against the $60-80M guidance. If Q4 is below $20M, H2 comes in around $30M -- roughly half the low end.
• Q1 FY27 earnings (~October 2026): The "early FY27 ramp" was supposed to begin by this quarter. If there are still no wafer-level orders, the timeline claim from the Q2 call is definitively falsified.
• China IP appeal ruling (H2 2026): Binary outcome. 54% of current revenue geography depends on maintaining IP protection against a domestic competitor.
• Jan 2027 options expiry: Max pain $18. Largest single-expiry open interest.

What would change our mind

• An 8-K announcing a binding purchase order of $10M+ for FOX-XP wafer-level systems from a US AI customer. Not a "forecast" or "pipeline" -- a committed order in an SEC filing.
• Any insider open-market purchase. One buy breaks the pattern of unanimous selling.
• Q4 FY26 revenue above $30M, which would suggest a ramp began without prior announcement.
• COHU or ONTO reporting loss of AI test orders, which would challenge the read that wafer-level is losing to package-level.

Evidence