Zero One Daily Intelligence Brief — May 11, 2026

Apple hands Intel its first foundry win as AMD picks up TSMC capacity from MediaTek

Top Market Signals

Apple hands Intel first foundry win, TSMC keeps flagship

Apple (AAPL US) and Intel (INTC US) have finalised a preliminary deal for Intel to manufacture some Apple chips, per Wall Street Journal reporting Friday, after more than a year of negotiations; Taiwan analysts confirm TSMC (2330 TT) keeps Apple's flagship iPhone and 5G modem orders.

The deal lands the catalyst flagged in our 6 May 2026 Blade. That piece had Apple touring Intel and Samsung fabs as the news driver of Intel's first $100 close; the What to Watch entry from that day specifically named "Apple's official capacity commitment to Intel or Samsung" as the next confirmation. Today's WSJ scoop is that signal.

The economics on Intel's side are real. Per Wccftech via Liberty Times, Intel 18A wafer pricing comes in roughly 25% below TSMC's 2nm; 18A is also positioned as a way to soften Apple's exposure to current memory-price inflation and to reduce its supply-chain and tariff concentration. GF Securities and DigiTimes earlier reported that Apple's likely first 18A products are 2027 entry-level M-series chips, 2028 non-Pro iPhone chips, and an Apple ASIC code-named Baltra (2027/2028) using Intel's EMIB packaging. Apple has already bought PDK samples from Intel for 18A-P evaluation.

The flagship socket stays at TSMC. Two Taiwan analysts cited by CNA, Liu Pei-chen at the Taiwan Institute of Economic Research and Lee Fang-kuo at Uni-President Investment Trust, both said TSMC's InFO and CoWoS advanced packaging plus stable yield record remain the binding constraint. Liu framed Intel and Samsung as "strategic reserve" rather than primary foundry alternatives. Lee made the more pointed observation that Apple's diversification is supplier-pushed, not buyer-led: TSMC's 2nm capacity is full of NVIDIA (NVDA US) AI orders, which is what is forcing Apple to a second source.

The cleanest read-through is the related Economic Daily News scoop the same day: Apple is giving its self-developed C2 5G modem chip volume entirely to TSMC at 2nm GAA. The modem will replace Qualcomm (QCOM US) silicon across iPhone, iPad, and Apple Watch from 2027 (estimated 100m+ units annually), with mmWave and satellite support added. TSMC backend is at capacity, so Apple's 5G modem testing has been routed to TSMC subsidiary Precision Test Systems (3374 TT), which has emergency-ordered 600 test tools to take the work, with new capacity ramping from 2027.

So Apple's diversification is real but bounded. Intel gets non-flagship volume on a second-source basis; TSMC keeps the M-series, A-series, and the new C2 modem at the bleeding edge.

Linked stocks: AAPL US, INTC US, 2330 TT, 3374 TT

Sources: 英特爾傳奪蘋果訂單 專家：台積電技術與良率仍是蘋果代工首選, 傳英特爾搶到蘋果訂單 外媒爆18A竟比台積2奈米便宜「這麼多」, 台積傳奪蘋果自研 5G 數據機晶片大單 精材跟著旺

AMD picks up TSMC capacity from MediaTek and Qualcomm

AMD (AMD US) is absorbing the TSMC (2330 TT) 4nm and 5nm wafer capacity that MediaTek (2454 TT) and Qualcomm (QCOM US) have just released, per Wccftech via Liberty Times reporting Saturday; the reshuffle accounts for roughly 20,000 to 30,000 wafers (15m to 20m mobile chips) and is being driven by entry- and mid-tier smartphone weakness.

The mechanism is the memory super-cycle. DRAM is now around 35% of an entry-level smartphone's bill of materials and NAND another 19%, so memory pricing alone has pushed combined memory cost above 50% of BOM, gutting the entry- and mid-segment handset markets. MediaTek and Qualcomm responded by cutting TSMC 4nm/5nm wafer starts.

AMD has stepped in. Lisa Su confirmed on the recent earnings call that 1Q26 growth was volume-led, with shipment increases across not only the high-end Turin server CPU line but also Genoa and the broader Zen core families.

The investor read-through has three parts. First, this is the cleanest mechanistic example of the AI memory super-cycle hitting consumer demand at the BOM level, not just at the headline. Second, TSMC's 4nm/5nm utilisation is unaffected; the wafers are simply rotating from a low-margin mix (mobile SoC) to a high-margin one (server CPU), which should be revenue-positive at TSMC even before the 2nm AI ramp. Third, AMD's data centre share gain against Intel now has a direct supply-side enabler on top of the demand-side TAM expansion flagged in our 6 May 2026 Blade (AMD 1Q26 data centre revenue +57% YoY; server CPU TAM tripled to USD 120bn).

The corollary signal: MediaTek and Qualcomm gross margin pressure into 2H26 should now be the consensus risk. Both firms had been seen as benefiting from the AI ASIC pivot; that thesis is intact, but the smartphone-side overhang is showing up in capacity actions, not just commentary.

Linked stocks: AMD US, 2454 TT, QCOM US, 2330 TT

Sources: 傳聯發科、高通空出台積電部分產線 AMD蘇姿丰撿到槍

Innolux's FOPLP joins TSMC at Longtan after SpaceX win

Innolux (3481 TT) is reportedly joining TSMC (2330 TT) on fan-out panel-level packaging (FOPLP) at TSMC's Longtan fab, per Economic Daily News, after Innolux's FOPLP first won a SpaceX supply slot earlier this year. Chairman Hung Chin-yang's shareholder letter declares Innolux's advanced-packaging position is "in the leading group."

This is a structural fit with the May 8 CoPoS exclusivity story. TSMC is building a closed advanced-packaging ecosystem one generation ahead; Innolux's FOPLP and TGV (through-glass via) capability slot directly into that pivot from wafer-level (CoWoS) to panel-level (CoPoS / FOPLP). The chip-first FOPLP volume at Innolux has scaled roughly 10x to over 40m chips per month, with full utilisation and "economic scale" reached, per Hung's earlier interview.

Three product wins are worth surfacing from the shareholder letter. First, Innolux is supplying a new GaN multi-die high-power management IC for SPS (Smart Power Stage) applications inside AI servers, with a follow-on roadmap. Second, automotive radar and gesture-control chip customers are validating Innolux's embedded packaging using low-Dk, low-Df dielectric materials. Third, Innolux's RDL interposer technology is now in validation with large-package customers for the 1- to 2-year horizon when AI/HPC chip die sizes scale beyond what wafer-level RDL can carry.

Innolux declined to comment on the TSMC partnership specifically, but Hung's framing on TGV ("clear competitive differentiation, integrated advantage from prior glass-substrate experience") plus the SpaceX reference point reads as a company that has crossed from technology demonstration to revenue generation.

The read-across is to the rest of Taiwan's panel-level packaging supply chain. AUO (2409 TT), the other large display maker, has flagged FOPLP exposure as a longer-dated optionality. Innolux is now demonstrably ahead. For TSMC, this expands the supplier base for CoPoS and FOPLP without breaking the exclusivity terms it imposed on equipment vendors (those terms cover next-generation tooling, not packaging-house JVs).

Linked stocks: 3481 TT, 2330 TT, 2409 TT

Sources: 群創 FOPLP 傳將與台積合作 董座喊先進封裝技術「已站在第一領先群」

CSP backlog hits $1.5 trillion, +142% YoY

The combined remaining performance obligations of the top three US cloud-service providers have reached USD 1.5 trillion, up 142% year-on-year, per cnYES synthesis of US CSP 1Q26 earnings; Google and Meta have raised 2026 capex guidance by 3% to 8%, and Microsoft's full-year capex run-rate is now around USD 190 billion.

This extends the May 6 Blade signal of "hyperscaler capex now ~USD 725 billion." The 1.5T backlog figure is the more useful number going forward because RPO is the contractual demand pipeline behind the capex spend, not the spend itself. A 142% YoY expansion in cloud-customer commitment is the cleanest piece of evidence for the demand-side durability of the AI infrastructure cycle through 2027 to 2028.

For the Taiwan AI-server supply chain (Foxconn 2317 TT, Quanta 2382 TT, Wiwynn 6669 TT, Gigabyte 2376 TT), the pull-through assumption now rests on much firmer ground. The capex revision direction is up across both Google and Meta, the two that were at risk of guidance cuts on the previous cycle peak.

The implication for memory and packaging is the same vector we have been writing for the past three weeks: as long as RPO keeps expanding, hyperscalers have no incentive to slow capex and every incentive to lock supplier capacity (cf. the SK Hynix funding offers in our 8 May 2026 Blade). The constraint is fab build-out timing, not orders.

Linked stocks: MSFT US, GOOGL US, META US, 2317 TT

Sources: 美系CSP資本支出全面上修，AI伺服器供應鏈大爆發：鴻海、廣達、技嘉、緯穎

NVIDIA's 2026 strategic-stake portfolio crosses $40 billion

NVIDIA's (NVDA US) AI-related strategic investments in 2026 now total over USD 40 billion, with its OpenAI stake the largest single position, per Benzinga reporting Saturday. The total includes the USD 2.1 billion equity-and-warrant deal with IREN (IREN US) flagged in our 8 May 2026 Blade.

The pattern thread connects three prior Blades. Our 13 April 2026 Blade flagged Nanya Technology's NT$78.7 billion (~USD 2.5 billion) private placement that brought SK Hynix subsidiary Solidigm, SanDisk, Kioxia, and Cisco aboard for a combined 10.19% stake. Our 8 May 2026 Blade flagged NVIDIA's USD 2.1 billion IREN equity commitment plus a separate USD 9.7 billion Microsoft cloud agreement at the same Texas Sweetwater campus, framing it as the largest single endorsement of "neocloud" GPU-rental businesses.

Today's USD 40 billion total is the rolled-up portfolio view. The throughline is that NVIDIA, hyperscalers, and memory-customer consortiums are no longer just buying chips; they are taking equity positions in suppliers, partners, and adjacent infrastructure operators. NVIDIA's deployment now includes both data-centre operators (IREN) and core compute partners (OpenAI), creating a vertically aligned set of dependencies that the rest of the supply chain has no easy way to replicate.

For investors, the question is no longer whether the AI capex cycle is "structural." It is whether the resulting supplier, partner, and infrastructure structures will reset the rest of the value chain's pricing power and competitive intensity for years. We will continue to track each new strategic stake as it lands.

Linked stocks: NVDA US, IREN US

Sources: Google News

先聲 First Word — Exclusives from Chinese-Language Sources

Compal pushes for AI-server first-tier with switch development

Compal (2324 TT) chairman Chen Jui-tsung is targeting the first tier of global AI-server ODMs and is now developing data-centre switches in addition to its server lines, with deals in place to supply Verda (Europe) and Datasection (Japan, USD 325m of NVIDIA B300 servers). Switch development moves Compal into Broadcom and Arista territory and is the first time a Taiwan ODM has gone vertical into the network layer. (10 May 2026) Source: 仁寶攻 AI 火力全開 切入交換器研發建置 拚躋身第一梯隊

AES-KY guides lithium-battery share of data-centre backup higher

AES-KY (6781 TT) chairman Soong Fu-hsiang flagged in the company's annual shareholder report that AI data-centre and 5G base-station backup-power demand will lift the lithium-battery share of AES revenue in 2026, with lithium continuing to displace lead-acid in base-station deployments. AES posted 2025 revenue of NT$16.0 billion (+60% YoY), net income NT$3.26 billion (+50.23%), EPS NT$38.2. The data-centre power-budget vertical is becoming a coherent investable theme separate from the server side. (10 May 2026) Source: AES 鋰電池銷售看增

Simplo chairman warns notebook memory shortage to drag 1H26 shipments

Simplo (6121 TT) chairman Soong Fu-hsiang's annual report says the notebook industry will face a memory-supply shortage in 2026 that may compress Simplo's notebook-related shipments, even as new businesses (e-bikes, data-centre backup, drones) hold up. Simplo 2025 revenue NT$82.15 billion (+2.7%), EPS NT$30.6. The notable detail is Simplo and AES-KY share a chairman, and both 1Q26 outlooks now treat data-centre backup as the swing growth vertical for the battery-module group. (10 May 2026) Source: 新普 NB 業務面臨挑戰 審慎樂觀看待新事業