Revenue Up 70%, Stock Down 3%: Teradyne Hits an AI Testing Plateau

The headline numbers:

On its own, it's hard to find fault with this set of numbers. The company beat its own guidance, and its outlook came in above Wall Street's expectations. But put it back on the quarter-by-quarter chart and the picture changes:

Revenue has now sat in a narrow $1.28–1.33 billion band for three quarters in a row. The year-over-year growth still looks impressive only because last year's starting point was so low. In Q4, the comparison base jumps to $1.083 billion, and the $1.3 billion guidance midpoint implies growth of only about 20%. By Q1 2027 the base becomes $1.28 billion, so for growth to stay positive, the next guidance has to keep revenue above that number.

Run the full-year math: revenue for the first three quarters totals about $3.915 billion. Add Q4 guidance of $1.25–1.35 billion and 2026 lands between $5.16 and $5.26 billion, up 62%–65% from 2025's $3.19 billion. That full-year figure is essentially locked in. The real question is 2027. To deliver 20% growth next year, Teradyne would need about $6.25 billion in revenue, or an average of $1.56 billion a quarter — about 20% above today's plateau. So there is really only one number to watch in the next report: whether the midpoint of Q1 2027 revenue guidance comes in above $1.28 billion. If it does, the plateau is slowly rising. If it doesn't, year-over-year growth turns negative, and the market will quickly start valuing Teradyne like a cyclical stock.

On the demand side, there's no sign of weakening. Three months ago management said "our Q3 guidance reflects robust AI-related demand" (company 8-K, 2026-07-29). This quarter, CEO Greg Smith said "Our wafer to data center strategy and the breadth of our portfolio sustained positive momentum in Q3," and described the company heading into 2027 with "multiple growth engines" across test and robotics (company 8-K, 2026-10-28). The company also reported record revenue in memory test and system-level test (company 8-K, 2026-10-28). Still, "sustained momentum" is a different tone from the "surging demand" language of six months ago. The foot has moved off the accelerator and onto cruise control.

Teradyne is an asset-light equipment maker. It doesn't borrow heavily or issue new shares to fund growth. So the right questions here are: how much of the profit actually turns into cash, and how much is flattered by "adjusted" accounting? The answer is cleaner than at most AI hardware companies:

The gap between GAAP and non-GAAP is only around 5%, mostly amortization and restructuring items. Stock-based pay is modest even by semiconductor equipment standards, and existing shareholders weren't diluted this year — thanks to buybacks, they actually own slightly more of the company. The short-term loan that worried the market last year turned out to be working capital for the busy season, not a funding gap for capacity expansion, and it's now paid off.

The one thing to nitpick is cash conversion. Q3 free cash flow of $227 million against non-GAAP net income of $360 million is a conversion rate of about 63%. That's not bad, but it means some profit is still tied up in receivables and inventory. Test equipment is shipped in batches and booked as revenue on customer acceptance, so when revenue levels off, working capital should gradually free up and cash conversion should improve. If revenue stays flat next quarter but conversion gets worse, that would suggest inventory is piling up in the channel. For now, earnings quality is good enough that the earnings behind the valuation are real. The only debate is whether those earnings will grow next year.

ATE (automated test equipment — machines that screen out defective chips one by one at the wafer and packaged stages) has an unusual place in the chip supply chain. It doesn't sell manufacturing capacity. It sells "how long each chip needs to be tested, and how many times." AI chips have supercharged this business because a large GPU or custom ASIC (a chip designed for one specific customer's workload) takes far longer and needs far more test steps than a smartphone processor. And HBM — high-bandwidth memory built from many stacked layers — has to be tested for yield repeatedly, both before and after stacking, which multiplies test hours.

Where are the upstream bottlenecks? Teradyne's machines rely on high-end FPGAs, high-speed digital channel boards, precision power supplies and interface boards. The company doesn't disclose its specific suppliers or how concentrated its purchasing is, and no verifiable list or share data is publicly available. The real choke points may actually sit around the test cell: probe cards, handlers and test sockets — consumables and add-on equipment often made by a separate set of specialist firms. Whenever AI chips move to a new generation, interfaces and sockets have to be redesigned, and those suppliers may not keep pace with the testers themselves. TrendForce said on 2026-10-13 that integrated optical-electrical test platforms are becoming critical for volume production of CPO (co-packaged optics, a technology that places optical components right next to the switch chip) and will reshape the test equipment supply chain. For Teradyne, that's a new market to expand into — but also a new battlefield where it must compete with optical test specialists.

Three types of customers pay the bills: vertically integrated giants building custom AI chips (what the company calls "VIP compute" customers — cloud providers designing their own ASICs, plus their design partners), HBM and DRAM memory makers, and the OSATs (outsourced assembly and test firms) that package and test chips on their behalf. Of Q2's $1.12 billion in Semiconductor Test revenue, SoC accounted for $843 million and memory for $212 million (Futurum summary of the earnings call, 2026-07). More than three-quarters of test revenue depends on compute chips, and compute-chip purchasing is concentrated among a handful of customers. The company's annual report explicitly flags customer concentration as a risk, but it doesn't publish the revenue share of any single customer, and we could not find that figure.

On contracts: test equipment is delivered against orders and recognized on acceptance. There are no long-term take-or-pay contracts, so backlog can be pushed out or even cancelled. That makes the customer end the most fragile link. If a single custom ASIC customer delays its next-generation chip by one quarter, Teradyne's revenue line dips immediately, and there's nothing in its contracts to protect against that. The slip in Semiconductor Test revenue from $1.12 billion in Q2 to $1.072 billion in Q3 is a mild example of exactly this kind of customer timing.

Put Teradyne next to other semiconductor equipment and materials companies for the third quarter and its position is clear:

TER's gross margin is higher than the lithography and etch giants', and its growth ranks near the top. That says the test business has at least as much pricing power in this AI cycle as the front-end equipment that actually makes chips. But there's a divergence worth noticing. In the same window, UCTT's initial guidance on October 27 was driven by volume growth and surging demand, LRCX's October 21 guidance for its next fiscal quarter also pointed to surging demand, and ASML, CAMT and INTT all cited surging demand behind their outlooks too (company guidance announcements, 2026-10). Front-end and component peers are still guiding higher, while TER's Q4 guidance is merely flat. In other words, chip factories are still expanding capital spending, but demand at the test stage has entered a plateau. These two facts don't contradict each other: front-end tools get installed first, wafers ship later, and test demand should in theory catch up with a lag. That time gap is exactly what long-term bulls should be counting on.

How has the market priced similar report cards? We don't yet have comparable earnings-day stock moves or options data from test equipment peers this quarter, so we won't force a comparison. We also couldn't find verifiable public figures from main competitor Advantest's latest quarter, so it's left out of the table.

The trailing twelve-month P/E and PEG ratios are close to useless here, for a simple reason: they're based on an earnings curve whose slope is flattening sharply. A PEG of 0.18 uses a historical earnings growth rate of 278.7%, which can't repeat next year. The better yardsticks are "plateau" annualized earnings and enterprise value (EV) relative to operating profit.

What does a 37x plateau P/E really mean? It means that if Teradyne doesn't grow at all next year, an investor would need 37 years of profits to earn back the purchase price. Mature front-end equipment leaders typically trade at somewhere in the 20s mid-cycle. In other words, this multiple already pays in advance for two things: first, that the plateau lasts at least another two to three years; second, that a second step up arrives from the second half of 2027 — whether from the next HBM generation, volume CPO testing or the ramp of the next custom ASICs, at least one of them has to deliver. Of the 22 Wall Street analysts covering the stock, 4 rate it Strong Buy, 12 Buy, 6 Hold and 0 Sell — 73% bullish. Consensus is clearly betting the second step will come.

What's missing before this becomes a "double win" — rising earnings plus a rising multiple — is one condition: earnings have to start improving again. Right now profits are flat at a high level, and the pullback from the stock's peak has only squeezed the multiple from around 50x back to around 40x. Room for the multiple to expand only opens once growth picks up again. A low PEG based on past growth can't carry that story on its own.

Semiconductor test equipment has its own cycle, broadly following the roughly four-year rhythm of chip inventories and capital spending. This time, AI accelerators and HBM have pulled it into an unusually steep upswing. Three sets of readings help locate where we are:

Taken together, my read is this: the AI testing cycle is in the middle-to-late stage of its upswing. Growth has peaked and volume is hovering near the top, but it has not yet turned down. Think of a car with the speedometer steady at 120: still moving fast, but no longer accelerating. The question now is whether the road ahead has another climb.

The two leading signals most worth watching: first, the quarter-over-quarter trend in the Semiconductor Test segment. It has already slipped from $1.12 billion to $1.072 billion; if Q4 falls below $1 billion, that would mean top customers are slowing purchases, and the odds of a peak rise sharply. Second, gross margin. When demand weakens, test equipment makers often cut prices first to protect market share, so gross margin falling below 58% and declining for two straight quarters usually flags a cycle turn earlier than revenue does. On the other hand, if the Q1 2027 guidance midpoint climbs back above $1.35 billion while memory test revenue keeps setting records, that would be evidence the second step is starting to form.

This section covers week-to-week signals — timing for the next few weeks only. Readers planning to hold for a year or more can skip it.

The money-flow picture is remarkably consistent: the metrology and inspection niche is being sold down in bulk.

In trader terms: institutions are swapping inspection-related semiconductor equipment holdings for defensive and value names. TER's post-earnings drop landed in the middle of a selling wave that had already been running for two weeks. The stock didn't lead the money — the money left first. Historically, after similar sub-industries first fall into the weakest tier, the median return over the next 20 trading days has been 0.0%, with a 55% chance of a gain (3,077 cases, covering 2021-04-12 to 2026-09-29). In other words, weakness alone doesn't mean further declines, but it doesn't give dip buyers a statistical edge either. The real change signal is a break in the streak of net outflows and the money-flow score moving back toward zero. Until then, any bounce is more likely a window for sellers to get out.

Long term (quarters to a year): worth following. The fundamental case is solid: gross margin holding around 59%, profits converting almost entirely into cash, zero debt with ongoing buybacks, and market share in AI compute and HBM testing validated by two quarters of actual results. Chip-factory capital spending is still expanding, and test demand naturally lags equipment installation, so there's a realistic source for a second step up. The costs are clear too: on plateau annualized earnings the stock still trades at about 37x, and next year's year-over-year growth will most likely drop sharply from above 60% to single digits or even turn negative. During that transition the valuation has no fuel to expand, so holders need to tolerate a stretch of "results are fine, the stock doesn't move." Position decisions should be tied to checkpoints: Q1 2027 revenue guidance holding above $1.28 billion is the minimum for staying in; a move back above $1.35 billion alongside new records in memory test would justify a bigger position; if guidance falls below $1.25 billion and Semiconductor Test keeps declining quarter over quarter, it's time to admit the plateau is starting to crumble and reassess the long-term thesis.

Short term (days to weeks): don't chase; wait for the outflows to stop. The niche has seen 11 straight days of net outflows, volume momentum sits in the weakest percentile of the market, and the stock has fallen below both its 50-day and 200-day moving averages. No support level has been confirmed, and no one on the money-flow side is stepping in to buy. In setups like this, rebounds tend to be sharp and short-lived — better suited to people getting out than people getting in. In the short term, we're waiting for two things to happen together: the sub-industry's outflow streak breaking, and the stock reclaiming its medium-term moving average.

After these two conclusions, the remaining question is how to put them into practice: which two conditions to watch and at what price levels to make decisions. That's covered in the second half.