Hua Hong's AI Census: More Than Half the Company, Depending on the Definition
Record revenue, a freshly consolidated fab, and a 10–15% price hike promised for 2027 — powered partly by a memory squeeze and partly by an AI label management concedes is 'a bit nebulous.'
1347.HK · Earnings Call · 2026-11-05
The merger math, and the guidance that rode in with it
Hua Hong Grace's third quarter was the first to consolidate Huali Microelectronics, and the arithmetic of that deal is doing a lot of heavy lifting. Revenue came in at $987.1 million, up 19.9% year-on-year on a restated comparable basis and 9.9% sequentially. Gross margin was 24.6%, up 7.6 percentage points. Net profit attributable to shareholders was $123.4 million, up 147.1%. Strip Huali out and the core business still delivered $803.4 million of revenue and a 19.9% gross margin — both above the top end of guidance of $770–780 million and 16–18%.
The CFO was unusually direct about what the acquisition is worth: “the acquisition of Huali Microelectronics is an incredible transaction for the company” — Yu-Cheng Wang, Executive Vice President and Chief Financial Officer · 2026-11-05, adding roughly 40,000 wafers a month, about $700 million of revenue at a ~40% gross margin and more than $200 million of net profit — worth "about 3 to 4 points" on group gross margin. One small oddity worth flagging: on the same call, Dr. Bai put ex-Huali gross margin at 19.9%, while the CFO described the business without Huali as being "around like 17%." Both are above guidance either way, but two executives framing the same clean number 3 points apart is the kind of slippage that usually shows up when a company is simultaneously selling a pro-forma story and a reported one.
Q4 guidance of $1.08–1.09 billion at 25–26% gross margin is a big step from a $987 million base. Management was explicit that the quarter-over-quarter lift is not the acquisition: Huali's capacity is flat sequentially. It is Fab 9B equipment coming online, plus residual price hikes already negotiated but not yet fully realized in the average selling price.
The AI census problem
Here is the most interesting shift on this call. Dr. Bai volunteered that
more than 50% of our products in the AI-related sales. Otherwise, I wouldn't see this much growth this year.
He then immediately undercut the precision of that number, saying the old computing/communications/consumer taxonomy is outdated and the AI definition itself is "a bit of nebulous right now," with the company planning to re-cut its end-market disclosure.
The direction of travel is real — AI related demand is genuinely broad across MCU, flash, and power management, and the recurring keyword here dates back several quarters. But the magnitude claim is new, and it collides with the company's own arithmetic. On the November 2025 call, the CFO was asked exactly this question and sized AI servers at “about 10% to 12%” — Bai Peng, Chairman and President · 2025-11-06 of revenue via power management. Twelve months later, "more than 50%" of everything. Some of that is a genuine broadening — AI clusters are now pulling on flip-chip PMIC, MCU, power discrete and memory — but a 4–5x jump in a year is mostly a definitional widening, not a demand datapoint. Investors should treat the >50% number as a directional narrative, not a measurable exposure.
Silicon photonics: from hobby to capex line
Silicon photonics quietly got promoted. On the August 2026 call, Bai described it as an area the company was already in at the device level and where “we are seeing the early stages of doing that” — Bai Peng, Chairman and President · 2026-08-13 — exploration language. On this call it hardened into a capital allocation decision: “the power management and silicon photonics, those are the two areas that we want to add significantly more capacity” — Bai Peng, Chairman and President · 2026-11-05 in 2027. That is a real change of state — from "we make some MCU and power chips that end up inside an optical module" to "we are building capacity for the silicon content of the module."
And it is being announced into a hostile tape. The market's optics complex has been one of the worst places to own anything: co packaged optics names are down roughly 36% over 90 days, and AI data centers has 57 decliners against 7 advancers over the same window — down about 31%. Hua Hong is adding capacity into a subsector the market has spent a quarter de-rating. That is either conviction or a timing error; there is no middle.
What got quietly upgraded — and what is stuck on repeat
Two reversals are worth noting. First, power discrete. In February 2026, Bai called it the platform with “the biggest pressure in terms of growth” — Bai Peng, Chairman and President · 2025-11-06, squeezed by low barriers to entry and silicon carbide pricing. This quarter it grew 19.5% year-on-year and is receiving selective price increases. The drag became a contributor.
Second, the memory product outsourcing thesis. Analysts have now asked about DRAM/NAND splitting the peripheral logic die onto a separate wafer for at least three consecutive calls. Bai's August 2026 answer was that it was “still probably in the early stages of this technical transition” — Bai Peng, Chairman and President · 2026-08-13; his answer this quarter is materially identical — collaboration projects ongoing, volume unclear. It is a genuinely large optionality, and it is genuinely unmonetized. Thematically it is confirmed broadly: memory cost inflation and DDR4 supply constraint show up across this week's reporters, from AOSL to NTGR — memory tightness is real, and its spillover into specialty foundry is the mechanism Hua Hong keeps describing without yet having the purchase orders to prove.
What management did stop leaning on is the book ratio as a bragging right. Where earlier calls touted orders at 1.5–2x capacity, Bai now warns against using it: “I want to caution you to use the bill book ratio because for our industry, this is not probably a typical target... this is also something of the pricing” — Bai Peng, Chairman and President · 2026-11-05. Precisely right — a book-to-bill inflated by underpricing is not a signal. The cleaner disclosure is that at least half of orders cannot be filled.
The bet embedded in the guide
Underneath the guidance sits a pricing assumption: another 10–15% average increase in 2027, on top of 10–15% this year, taking gross margin to ~25% or above. Bai's own caveat is the honest part — “our crystal ball probably don't look very far beyond” — Bai Peng, Chairman and President · 2026-11-05, and historically these supply demand imbalance cycles correct in 2–3 years once new capacity lands. The new capacity being committed across the industry right now, including by Hua Hong, is the very thing that ends the cycle.
So the shape of this report is: a genuinely transformed income statement, a fab integration that is accretive, a average selling price cycle management has skillfully exploited, and a narrative layer — "more than half AI," silicon photonics, memory logic die — that is running materially ahead of the disclosed numbers. The quarter was excellent. The story around it is doing more work than the P&L yet supports.