Zero One Daily Intelligence Brief — June 11, 2026

TSMC flags price hikes after record May as SK Hynix eyes US$14bn US listing

Top Market Signals

TSMC follows record May with price-hike signals

TSMC (2330 TT) reported May revenue of NT$416.98bn, up 30.1% YoY and 1.5% MoM, a new monthly record and a third straight month above NT$400bn, while its CFO told the BBC the company is not ruling out price increases as inflation pushes up operating costs.

January-May revenue reached NT$1.96tn, up 30% YoY. April and May together brought in NT$827.7bn, roughly two thirds of the quarter's guided range of US$39bn to US$40.2bn, and Taiwanese institutional analysts now see 2Q26 revenue reaching the top of guidance if June holds at current levels. Commercial Times reporting puts 2Q26 monthly wafer capacity at 160,000-175,000 wafers, with customer queues showing no meaningful relief.

The pricing commentary is stacking up. CFO Wendell Huang told the BBC that inflation is raising costs and that TSMC has not ruled out charging more, though any increase would not be a sudden multiple of current prices. Chairman C.C. Wei told the 4 Jun shareholder meeting the company is preparing to discuss further price increases with customers. Supply chain sources cited by Commercial Times expect advanced-process foundry prices to rise moderately in 2H26, with the tightest node, 3nm, estimated at around a 15% increase.

Huang also said the most advanced process will stay in Taiwan, that overseas fabs were built on customer demand rather than government pressure, and that moving the semiconductor manufacturing ecosystem to the US would take five to ten years or longer.

Linked stocks: 2330 TT, TSM US, NVDA US, AMD US

Sources: 台積5月營收拔尖 Q2衝高標, 通膨推升成本…台積財務長黃仁昭：不排除漲價, 連續3個月破4千億 台積電5月營收創新高

Google's Intel order covers packaging, not wafers

The Google order that lifted Intel (INTC US) earlier this week is taking clearer shape: The Information reports Google has booked Intel for more than 3 million TPUs (tensor processing units, Google's custom AI chips) in 2028, and Taiwanese reporting indicates the work is advanced packaging, with the chips themselves still fabricated at TSMC.

According to The Information, citing four people familiar with the matter, the order follows months of testing of Intel's advanced packaging. SK Hynix is separately testing whether its high-bandwidth memory works reliably with Intel's packaging, and NVIDIA is evaluating Intel to build a future processor that fuses four GPU dies into one unit, tied to its Feynman architecture due in 2028.

Liberty Times reporting cites foreign brokers saying the TPUs will still be fabricated at TSMC, with the compute die on TSMC's 2nm process and the I/O die on 3nm, leaving Intel responsible only for EMIB advanced packaging because TSMC's own packaging capacity is short. TSMC's CoWoS (chip-on-wafer-on-substrate, the packaging step that assembles AI chips) has been oversubscribed for more than two years; Tom's Hardware notes NVIDIA is expected to absorb about 60% of global CoWoS demand this year, with Broadcom and AMD taking another 26%. At the 4 Jun shareholder meeting, C.C. Wei said "It will be a long time before we can meet customer demand."

Our 9 Jun 2026 Blade flagged Google's reported 3-million-chip order reviving Intel's foundry story. Today's detail narrows what was won: a packaging qualification that relieves the CoWoS bottleneck while TSMC keeps the wafer work.

Linked stocks: INTC US, 2330 TT, GOOGL US, 000660 KS

Sources: Tom's Hardware, 連續3個月破4千億 台積電5月營收創新高

SK Hynix eyes a US$14bn US listing

SK Hynix (000660 KS) is preparing a US listing as soon as August, Reuters reports, with other reports putting the potential raise at up to US$14bn, which would give the HBM leader direct access to the US capital pool now chasing AI memory names.

The plan centers on an American depositary receipt (ADR) debut. The logic is straightforward: DigiTimes reporting notes the US market accounted for nearly 65% of SK Hynix's 1Q26 revenue of KRW52.6tn (about US$34.9bn), driven by high-bandwidth memory demand from AI infrastructure.

The listing lands alongside aggressive capacity commitments. Chairman Chey Tae-won told Nikkei Asia the company plans to triple wafer capacity by 2034, and trade publication The Elec reports SK Hynix will mass produce 375-layer NAND by year-end.

Linked stocks: 000660 KS, MU US, NVDA US

Sources: Google News, Google News, Google News

NVIDIA cuts memory per chip to stretch supply

NVIDIA (NVDA US) has decided to halve the memory capacity of the SOCAMM modules (the LPDDR-based memory boards) on its next-generation Vera Rubin Superchip, research firm TrendForce reports, because suppliers cannot allocate it enough LPDRAM in 2027.

TrendForce frames the move as a supply-side compromise, not a demand cut. Samsung, SK Hynix, and Micron have initially planned LPDRAM allocations covering only about 60% of NVIDIA's requirement, with limited room to raise output even as all three expand capacity. NVIDIA's answer is to lower per-module capacity and ship more modules, protecting unit share while hedging against continued tightness.

Our 6 Jun 2026 Blade flagged the market rumor that NVIDIA's Vera Rubin platform could carry less on-board memory, a report that sent storage stocks tumbling that week. TrendForce's note confirms the change and quantifies the supply gap behind it.

The standalone Vera CPU rack, which NVIDIA is pushing hard for inference workloads, is one of the main drivers of that LPDRAM demand. TrendForce estimates AI servers could become the largest single outlet for LPDRAM between 2028 and 2030, overtaking smartphones, as automotive and other ASIC players also begin adopting the memory type.

Demand visibility for memory makers now extends years out: even NVIDIA, the customer every supplier prioritizes, cannot secure full allocation at its planned specifications.

Linked stocks: NVDA US, 000660 KS, MU US, 2408 TT

Sources: TechNews

Super Micro raises US$7bn to cover order surge

Super Micro (SMCI US) will raise US$7bn through a series of equity and equity-linked transactions to buy components, telling investors the proceeds will go toward fulfilling roughly US$39bn of advanced AI server orders from more than 20 customers.

The package includes about US$3.75bn of depositary shares and US$1.25bn of common stock sold to underwriters, plus an at-the-market program of up to US$2bn starting after 3Q26. The stock fell about 8% in after-hours trading on the announcement, per CNA citing Reuters, and Commercial Times reporting has the shares down 28% in Wednesday's session as the broader chip selloff deepened.

The raise is part of a wider pattern of AI infrastructure players tapping equity markets at scale: CNA notes Alphabet launched an expanded share offering last week targeting US$84.75bn. A US$39bn order book against a US$7bn component bill underlines how working-capital-intensive the AI server build-out has become, with suppliers financing inventory quarters ahead of delivery.

Linked stocks: SMCI US, NVDA US, 2317 TT

Sources: 美超微將啟動70億美元融資計畫　滿足AI伺服器需求, Google News

OpenAI eyes 10GW Ohio campus backed by NVIDIA

OpenAI is in talks to lease a 10-gigawatt data center campus on federal land in southern Ohio, The Information reports, with a build cost of at least US$500bn and NVIDIA (NVDA US) expected to provide both the hardware and financial guarantees.

The campus would be developed by SoftBank's SB Energy on US Department of Energy land and would be the largest facility of its kind. Under the reported 20-year lease, OpenAI would control the equipment and begin paying rent once the facility starts operating, with the first phase slated for 2028. NVIDIA is expected to guarantee both OpenAI's lease and SB Energy's financing, extending the pattern of the chipmaker underwriting its largest customer's expansion.

The financing backdrop is getting more complicated. Yahoo Finance reports SoftBank tried to borrow US$6bn against its OpenAI stake before the talks stalled, and SoftBank shares have fallen 9.7%. Separately, OpenAI's CEO has told staff to expect an IPO within the next year, The Information reports.

Linked stocks: NVDA US, 9984 JT, SMCI US

Sources: 傳 OpenAI 洽租俄亥俄州巨型資料中心，輝達擬出資支持, Google News

先聲 First Word — Exclusives from Chinese-Language Sources

Ennoconn launches tender offer for Austria's Kontron

Ennoconn (6414 TT) announced at a late-night press conference it will acquire Kontron common shares at EUR23.5 per share (about NT$859), triggering a mandatory tender offer under German securities law after its stake crossed the 30% threshold, with completion targeted by end-August. The industrial PC group, a Kontron investor since 2016, is consolidating a European industrial AIoT and physical AI supplier active in over 30 countries, spanning smart manufacturing, Europe's 2G-to-5G rail communications upgrade, and avionics. (11 Jun 2026) Source: 樺漢公開收購Kontron 打造實體AI全球版圖

Taiwan's listed companies post a record NT$5.5tn May

Taiwan's listed companies reported combined May revenue of NT$5.5tn, up 38.04% YoY and 0.79% MoM, a third straight record month, with 126 companies setting monthly records versus 65 a year earlier. Analysts cited by Commercial Times see 2Q26 earnings potentially doubling YoY off a low base from last year's FX losses, evidence the AI cycle is now lifting the broad Taiwan market rather than a handful of names. (11 Jun 2026) Source: 上市櫃營收估5.5兆 最旺5月

Buyers lock up Nanya's 3Q26 DDR4 output

DigiTimes reports Nanya Technology's (2408 TT) 3Q26 DDR4 capacity is becoming even tighter, with its four major private-placement shareholders and large cloud service provider customers all locking in supply, and some large customers offering NCNR (non-cancellable, non-returnable) commitments to secure long-term capacity. Contract prices are set to keep rising in 3Q26, a sign legacy DRAM scarcity is hardening into contractual structures rather than spot-market spikes. (11 Jun 2026) Source: 南亞科DDR4私募大股東「底部支撐」　3Q26加劇緊缺推升合約價