Earnings Call Highlights: FormFactor (FORM) | FY2026 Q2 — The Toll Gate for HBM4 and CPO Testing: $1B+ Annualized Revenue, Gross Margin Above 50% for the First Time

2 days ago

7 min read

This quarter, FormFactor fully lived up to its role as the "invisible toll collector" of AI infrastructure. Revenue hit a record $258.2M, formally crossing a $1B annualized run rate, and non-GAAP gross margin topped 50% for the first time (53.3%). The real signal isn't the strong quarterly numbers but three things happening at once: high-speed stack testing for HBM4 has won it strong share at two major memory makers; full-year CPO test revenue was raised to "significantly above $20M"; and its probe card for NVIDIA's GPU (Rubin) has passed qualification, with shipments and revenue recognition starting in Q3. Test intensity is rising alongside HPC and advanced packaging, and FORM sits right at the gate of that intersection.

This article is for technology and industry trend analysis only and does not constitute investment advice.

1. Three Signals to Remember First

First, this is the third consecutive quarter of record revenue, and it crossed two milestones: annualized revenue above $1B and non-GAAP gross margin above 50%. The 2030 target model management laid out at its May investor day (revenue doubling to $1.6B, 55% gross margin, EPS doubling to $5) was shown this quarter, with actual results, to be more than a slide deck.

Second, growth is broad-based, not driven by a single product. The probe card and systems segments both grew sequentially, with DRAM, Foundry & Logic and CPO all firing together. The "diversification" strategy FORM has talked about for years finally shows up in the cash flow statement — by not betting on a single customer or application, it has a seat on every AI growth vector.

Third, DDR has briefly cut in front of HBM. Q3 DRAM revenue is expected to be flat with Q2, but the mix will shift noticeably from HBM to DDR — because memory customers are moving wafer starts to DDR, where prices are rising, to capture margin. Probe cards are custom consumables, "one card per chip design," so when customers change their wafer-start mix, FORM's product mix changes with it. This isn't weakening demand; it's customers optimizing profit.

2. Revenue and Financials

The hard numbers at a glance (non-GAAP basis, with GAAP EPS noted):

- Revenue $258.2M: +14% QoQ, +31.9% YoY, $18.2M above the guidance midpoint, an all-time record. - Non-GAAP gross margin 53.3%: +430 bps QoQ, +14.8 points YoY, and 230 bps above the top of guidance. GAAP gross margin was 50.7%. - Non-GAAP operating income $72.0M, operating margin 27.9%; non-GAAP opex was $65.7M, or 25.4% of revenue (27.4% in the prior quarter), showing clear operating leverage. - Non-GAAP net income $65.0M, diluted EPS $0.82: up about 46% QoQ and more than 3x YoY. GAAP diluted EPS was $0.71. - Segments: probe cards $209.7M; systems $48.5M, a record, +74% QoQ (engineering probers recovering plus the CPO ramp). - Cash: Q2 free cash flow $52.6M, operating cash flow $61.8M; cash and investments of $349M at quarter end.

One detail you can't skip: the 53.3% includes one-time items. The CFO broke it down plainly — about a third of the sequential margin improvement came from sustainable structural cost reductions, a third from higher revenue, and the remaining third from non-recurring items (IEEPA tariff refunds and precious metals recovered from the Baldwin Park closure). Excluding one-time items and mix tailwinds, the "baseline" non-GAAP gross margin at current volumes is about 51%. When reading FORM's gross margin, anchor on that 51%, not the headline 53.3%.

3. Technology and Business Highlights: HBM4, CPO and GPU Moving Together

HBM: SmartMatrix turns "known good stack" into FORM's moat. HBM made up about two-thirds of DRAM revenue this quarter, driven by two customers adopting FORM's SmartMatrix full-wafer contactor — which can test hundreds of complete HBM stacks in parallel at HBM4's 10Gbps-plus I/O speeds. The key is that this is the final test insertion "after stacking, before packaging," ensuring that what goes to TSMC CoWoS or Intel EMIB to be packaged with expensive GPUs/ASICs is a known good stack. The competitive barrier for this high-speed insertion is far higher than for ordinary DRAM wafer sort, which is why FORM's share is especially strong at two major memory makers; its relatively weaker share at the third customer is something to watch. We broke down the advanced packaging theme in full in The Great Shift in Optical Packaging (Part 3): TSMC vs. ASE vs. Intel.

CPO: full-year revenue raised, with test insertion one as the strategic foothold. At the start of the year FORM guided CPO revenue at $10M–$20M; this quarter it raised that outright to "reaching $20M by the end of Q3 and significantly above $20M for the full year." FORM's position is the first test insertion before the PIC (photonic integrated circuit) wafer is combined with the EIC into an optical module, ensuring the PIC is a known good die. Management itself admits CPO is a new technology and its acceleration–digestion cadence is hard to model quarter by quarter, but a serviceable market of about $400M by 2030 is on the table, and it's still the early innings. Worth contrasting is the market debate over when CPO will ramp — we discussed it in CPO Volume Ramp Pushed to 2028, but That's Not Bad News. FORM's acceleration on the test side is a confirmation that "upstream test comes before volume production."

Foundry & Logic: agentic AI relights the CPU line. F&L rose sharply quarter over quarter, mainly from rising probe card demand for data center CPUs — a quarter ago, agentic AI driving up CPU compute density was still a story; now the demand is showing up in cards. FORM's CPU exposure has three legs: long-standing high share at a leading data center CPU vendor; its relationship with the world's HPC leader (the major GPU vendor), expanding from networking into CPU/GPU product lines; and multiple design wins at a large fabless XPU customer after passing qualification. The three legs are at different stages, but together they mean that however CPU market share shuffles, FORM has a seat.

4. Management Outlook: Guidance Numbers and Tone, Broken Down

Q3 FY2026 guidance (non-GAAP):

- Revenue $270M ± $10M (midpoint another record). - Non-GAAP gross margin 54% ± 1.5% — but it needs to be unpacked. The jump from the 51% baseline to 54% comes mainly from a non-recurring IEEPA tariff refund of about $7.9M (about 300 bps) — tariffs levied from 2025 that were later ruled unlawful and refunded. The genuine volume tailwind will be partly offset by a weaker DRAM mix (HBM→DDR). In other words, 54% isn't the new normal — 51% is. - Non-GAAP opex $70M ± $2M (including about $7M of pre-production expenses for Farmers Branch). - Non-GAAP diluted EPS $0.86 ± $0.09; tax rate 15%–19%.

In terms of wording, there are two levels of certainty to keep apart. The HBM→DDR shift is "confirmed" — management said plainly it's customers' rational response to rising DDR prices, and FORM is simply following the wafer-start mix. The Farmers Branch margin gain, by contrast, is a "target": the new fab comes online in Q4 and ramps through 2027, and management explicitly placed the full "meaningful gross margin uplift" in 2028, with 2027 diluted by ramp inefficiencies and only aiming to offset that through operational improvements. That sentence matters — don't treat 2027 as the year of a big margin jump. Broadcom's narrative of extending AI visibility straight into 2027–2028 makes useful comparative reading: Earnings Call Highlights: Broadcom (AVGO) | Q2 FY26.

5. Supply Chain and Customer Clues

TSMC is again a 10% customer, and Taiwan is home turf. Taiwan revenue grew more than 30% sequentially this quarter, and the world's largest foundry was once again a 10% customer. Yesterday FORM also announced an expanded multi-year partnership with Keystone Microtech in Zhubei, Taiwan, to strengthen local assembly and service capability — to keep up with the "extremely steep ramps" of complex devices such as GPUs and custom ASICs. Most HPC and advanced packaging manufacturing happens in Taiwan, so FORM pushing its service radius toward Taiwan is following the money.

GPU (Rubin) qualification passed, with revenue recognition starting in Q3. This is the key second-half event: this quarter's 10% customer revenue had "no GPU component at all," and FORM has completed Rubin probe card qualification, with shipments and revenue recognition starting in Q3. GPUs completed the shift from legacy cards to advanced MEMS cards two or three years ago; now custom ASICs (next-generation Google TPU class) are taking the same path — once power and speed cross a threshold, advanced MEMS probes become mandatory. This wave of ASIC MEMS conversion will contribute little in 2026; 2027 is the ramp point. GPU and CPO revenue will be recognized at different points in the fabless/foundry ecosystem depending on product line — we unpacked the context of NVIDIA buying optics into its own footprint in NVIDIA Buys Into Optics: When Optical Communications Goes from Growth Stock to Index Heavyweight.

Hyperscaler custom ASICs: the real discontinuity is still ahead. FORM is engaged with all hyperscalers; some customers have already released multiple chips, while others are earlier. Management noted that next-generation TPU-class ASICs will require advanced MEMS probes — a replay of what happened with GPUs two or three years ago — and FORM expects to be a key supplier as these chips ramp in 2027.

6. Summary

This quarter FormFactor cemented its position as the "toll gate of the test layer": the more expensive AI chips get, the more critical yield becomes and the more complex packaging gets, the more valuable wafer test and known good die/stack become — and FORM sits right at the gate of HBM4, CPO, GPU and custom ASIC. Starting next quarter, watch three indicators:

1. Whether the gross margin "baseline" holds at 51% — don't be fooled by Q3's 54% including the tariff refund; look at the true level after one-time items. 2. When the DRAM HBM/DDR mix swings back — DDR cut in line this quarter, but high-speed HBM4/4E testing is FORM's long-term profit engine, and a mix shift back to HBM is the acceleration signal. 3. The actual revenue recognition curves for GPU (Rubin) and CPO — Q3 is the first real revenue for both lines, and the slope will determine the upside for 2027; further out, watch whether the Farmers Branch margin gain materializes in 2028.

In one sentence: this isn't just one pretty earnings report; it's the quarter in which a test equipment company proved it's standing in the right spot in the AI supply chain — but the real leverage (the Rubin ramp, ASIC MEMS conversion, Farmers Branch gains) is still in 2027–2028, and what you're buying now is the credibility of that slope.