Cohu's Parabolic Rally Meets a Wall — What Just Happened to the AI Test Equipment Play

After surging 175% in three months, COHU cratered 14% on Thursday. The business story is intact — $750M HPC pipeline, Eclipse platform wins, GaN power testing — but the stock ran far ahead of fundamentals, and systematic selling is now in control.

Cohu (NASDAQ: COHU) just delivered a textbook example of what happens when a small-cap stock runs too far, too fast — even when the story behind it is real. The stock hit an all-time high of $74.60 last Monday, capping a 175% rally from $33 in early April. By Thursday's close, it had given back 23% from that peak, including a 14.2% single-day collapse on July 2 — its worst session since 2020. There was no company-specific news. The selloff was a collision of profit-taking, a market-wide algorithm-driven distribution event, and a stock price that had simply outrun the math.

The business story: genuinely good

Cohu makes semiconductor test and inspection equipment — the machines that check whether advanced chips actually work before they go into servers, cars, phones, and AI accelerators. As chips get hotter, denser, and more complex, testing becomes harder and more valuable. Cohu has positioned itself at the center of that trend with three key platforms.

Eclipse thermal handlers. When an AI accelerator draws 1,000+ watts, you can't test it at room temperature — you need active thermal control that simulates real operating conditions. Eclipse, paired with Cohu's T CORE technology, does closed-loop junction temperature control. In Q1, it won platform-of-record status at a major US fabless customer developing AI server and inference devices — an engagement the company estimates as a $100 million incremental revenue opportunity over three years.

Diamondx for GaN power testing. Every AI data center needs precise power delivery, and the industry is shifting to gallium nitride (GaN) technology for efficiency. DiamondX was qualified for GaN power device testing, and Cohu secured multiple $5 million orders in Q1 for next-gen power architectures — including for NVIDIA's Vera Rubin platform.

Neon HBM inspection. High-bandwidth memory (HBM) stacks are the silicon sandwiches that sit next to AI accelerators. Cohu's Neon platform inspects HBM3, HBM4, and is already being designed for HBM5. The company forecasts 80% year-over-year HBM revenue growth to approximately $20 million in fiscal 2026, with the overall inspection business targeting $70 million.

On the Q1 call, CEO Luis Müller laid out a $750 million computing segment pipeline: roughly $650 million in test handlers, $100 million in HBM inspection, spread across 12 customers — five in qualification, seven in early engagement. For fiscal 2026, the company raised its HPC revenue target to $80–$100 million, up from an initial $25–$30 million.

Müller also highlighted an under-the-radar software story: a single $20 million system order in Q1 came attached with $330,000 per year in software subscriptions, which over the life of the equipment could yield about $5 million in high-margin recurring revenue.

The numbers people are betting on

Q1 FY2026 (reported April 30) showed revenue of $125.1 million — up 29% year over year. Orders surged 57%. Gross margin hit 46.5%, above guidance, helped by a favorable mix of recurring revenue that made up 60% of sales. But on a GAAP basis, the company still lost $12.1 million, or $0.26 per share. Non-GAAP net income was barely positive at $0.6 million, or $0.01 per share.

Q2 guidance calls for $144 million in revenue (±$7 million), a 15% sequential jump and 34% year over year. Gross margin is guided to approximately 44%, down slightly as systems revenue outpaces recurring. Full-year revenue growth is projected at 20–25%.

The balance sheet is in decent shape: $489 million in cash and investments against $305 million in total debt, mostly from a Q4 2025 convertible offering. Free cash flow was $8.3 million in Q1.

The valuation problem

Here's where the rubber meets the road. At $59.27, Cohu trades at:

• 40x forward earnings — on barely-positive non-GAAP EPS
• 5.8x sales — triple where it was in April, when the stock was at $33 and P/S was roughly 1.8x
• No trailing P/E — the company is still losing money on a GAAP basis

At $74.60, those multiples were even more stretched. The market was pricing in not just the $80–$100 million in near-term HPC revenue, but the full $750 million pipeline converted at high margins — years before it would materialize. When a stock goes up 175% in three months on a story that will take years to play out, the air gets thin.

The seven analysts covering the stock are all bullish — consensus mean price target is $60.29, almost exactly where the stock closed Thursday. The range is wide: $50 at the low end, $80 at the high end. After the 175% run, the Street's average target was already in the rearview mirror.

Insiders sold at $46 — stock ran to $74

Perhaps the loudest signal in the data: not a single insider has bought shares in the open market over the last 180 days, even as the stock was at $33, $43, or $59. Every transaction was either an option exercise (compensation) or a sale.

On May 22, CEO Luis Müller sold 45,154 shares at $46.62 — a $2.1 million transaction. CFO Jeffrey Jones sold 13,000 shares at $46.84 — $609,000. Three directors also sold blocks in mid-to-late May, all in the $44–$50 range. The stock then ran to $74.60 — nearly 60% above where management cashed out. Even at Thursday's close of $59.27, the stock sits 27% above where the CEO and CFO sold.

Insiders don't sell for just one reason, and volume-based selling plans are common. But the complete absence of open-market purchases — during multiple pullbacks — is the kind of signal the market notices.

What broke on Thursday

There was no single news catalyst for the July 2 crash. The move was a convergence of three forces.

First, the stock had gone vertical. It gained 50% in June alone. At $74.60, the RSI was deep in overbought territory — a 175% rally compressed into a single quarter invites gravity.

Second, the broader market entered a confirmed algorithm-driven distribution regime on Thursday. Across all tracked equities, 54 names were flagged, the basket returned -1.83%, and breadth was 55% negative. COHU's basket correlation — a measure of how much it moves with the broader market's distribution signal — jumped from 0.24 on Monday to 0.58 on Thursday. Nearly 60% of COHU's move was systematic: the stock got dragged down with the liquidation wave.

Third, the options market amplified it. COHU's at-the-money implied volatility was 91% — extreme for a $2.8 billion company. A $70 call wall had been acting as a magnet all month. Once the stock lost $70, momentum buyers stepped away and there was no structural support until the $55 put wall.

The algo's verdict: "watch"

Bargo's distribution scoring engine gave COHU a 64.2 on July 2 (out of 100), up from a pristine 2.0 on Monday. The signed delta — measuring whether actual buy or sell orders dominated the tape — flipped to -0.205, meaning 20.5% of the volume delta was net selling. Fifty-eight percent of Thursday's volume printed below the session's volume-weighted average price. The OBV (on-balance volume) slope cratered. The system flagged COHU with a "watch" designation — not yet an alert, but on the radar.

The correlation trap

Over the last 21 days, COHU has been sitting in the same correlation cluster as NVIDIA, Google, and 471 other names — the "everything AI" bucket. Over 63 days, it migrates to a cluster with Corning, Coherent, and Lumentum. Over 126 days — the longest-term lens — it's a near-singleton, paired only with Amkor Technology (AMKR).

The short-term message is simple: COHU will move with the broad AI trade. If the mega-cap AI names stabilize on Monday, COHU has a shot at bouncing. If they keep selling, COHU goes with them.

What to watch

The $55 / $70 range. The put wall at $55 is structural support; the call wall at $70 is resistance. Until the stock reclaims $70 or loses $55, it's in no-man's land, and either direction is a signal.

July 30 earnings. Q2 FY2026 results are the most important catalyst on the calendar. Consensus is $0.04 in non-GAAP EPS — but the real numbers to watch are HPC revenue (how much of the $80–$100 million is already booked?), Eclipse customer count (any new qualifications?), gross margin (the 44% guide is below Q1's 46.5%), and the full-year outlook.

Monday's open after the long weekend. When a stock crashes 14% into a three-day holiday weekend, Monday's open is a sentiment test. A quick reclaim of $63–$65 would suggest Thursday was overdone. Another leg lower toward $55 would put the put wall in play.

The broader AI equipment trade. COHU's 63-day correlation to Applied Materials, Lam Research, and KLA means it won't decouple from the semiconductor equipment sector. Watch those names on Monday for a read on COHU's path.

Cohu's AI testing thesis is real. The Eclipse handler, DiamondX GaN tester, and Neon HBM inspection platform all address genuine bottlenecks as chips get hotter and denser. The $750 million pipeline and accelerating HPC revenue are not fictional. But the stock ran 175% in three months while insiders sold in the $40s, and the July 2 crash was a reminder that narrative alone doesn't hold a stock up — especially when the algorithms start selling everything at once.

Sources: Cohu Q1 FY2026 earnings call transcript (The Motley Fool), Bargo algo distribution engine (Jul 2, 2026 session), SEC Form 4 insider filings, analyst consensus via Yahoo Finance.

More research at bargo.ai/research.