Semiconductor Spotlights Index

The Spotlights are SemiconductorX's company and program-level supply chain deep-dives. Where the Sectors pillar maps semiconductor demand by end market and the Chip Types pillar maps supply by device category, Spotlights follow the specific companies and programs whose supply chain decisions, market concentration, or geopolitical position make them disproportionately important to understanding the silicon economy. A handful of companies — TSMC, NVIDIA, ASML, Wolfspeed — exercise supply chain leverage far beyond their revenue rank because they occupy sole-source or near-sole-source positions in critical supply chains the entire industry depends on. Others — Tesla Terafab, Huawei HiSilicon, SpaceX — are building vertically integrated silicon programs whose outcomes will define competitive landscapes through 2030 and beyond.

Spotlights organize into four categories: Foundry & Equipment (the manufacturing supply chain); AI & Compute (the dominant demand-side companies); Automotive, Robotics & Industrial (electrification and physical AI silicon programs); and Sector Programs (mission-specific supply chain case studies). Each spotlight covers supply chain position, key products, foundry dependencies, geographic concentration, geopolitical exposure, supply chain bottlenecks, and outlook. The navigation logic: Sectors identify the constraint, Chip Types map the device supply chain, Spotlights profile the companies whose decisions determine whether and when the constraint resolves.

Foundry & Equipment

The foundry and equipment tier sits at the top of the semiconductor supply chain hierarchy. These companies do not design the chips the world uses — they manufacture them, or supply the equipment required to manufacture them. Their leverage derives from concentration: TSMC manufactures approximately 90% of chips at sub-5nm, ASML supplies 100% of EUV lithography scanners, and no credible alternative to either exists within this decade. Supply chain analysis at any other tier is incomplete without understanding the constraints these companies impose on the system.

• TSMC: TSMC makes ~90% of sub-5nm chips and is the single most critical node in the supply chain; NVIDIA overtook Apple as its #1 customer in 2025 at ~19% of revenue and holds ~60% of CoWoS allocation, and CoWoS capacity (120-130K wafers/month targeted for end 2026) is the binding AI GPU packaging constraint. N2 is in volume production with both Taiwan N2 fabs sold out for 2026, 2026 CapEx runs $52-56B, Apple, NVIDIA, AMD and Broadcom concentrate the customer base, and the $165B Arizona buildout is accelerating (Fab 21: Fab 1 in production, Fab 2 N3 pulled in to 2027).
• Samsung Semiconductor: Samsung is the most vertically integrated semiconductor company in the world, with Memory, Foundry and LSI under one structure; its HBM4 comeback made it first to ship (February 2026) with a 4nm base die and 1c DRAM, ~30% of NVIDIA Vera Rubin allocation, pricing parity with SK Hynix and the OpenAI Titan chip HBM4 exclusive, and HBM4E reaches 4.0 TB/s at 16 Gbps (GTC 2026). SF2 (2nm GAA) yields of 50-60% target 60-70%, >50% of Pyeongtaek foundry capacity has been redirected to HBM4 base die over external customers, Taylor, Texas Fab 1 ($25B) began pilot operations in February 2026, Tesla AI6 is a $16.5B anchor, Groq 3 LPU runs on Samsung 4nm and Pyeongtaek P4 is converting to 1c DRAM.
• Intel Foundry: Intel is the only US-headquartered company making sub-5nm chips and the only credible near-term US-domiciled alternative to TSMC for advanced-node fabless customers; Intel 18A (GAA RibbonFET plus PowerVia backside power) has been in high-volume manufacturing (HVM) at Fab 52 Arizona since January 2026 at ~60% yield against a ~70-80% profitability threshold, and Panther Lake launched at CES 2026 with a 180 TOPS NPU. The Terafab partnership (April 7, 2026, an IFS anchor arrangement) is the most commercially significant IFS win since IDM 2.0, NVIDIA invested $5B, Microsoft, Amazon and Apple are discussed IFS customers, the Ohio fab slips to 2030+, 14A brings High-NA EUV in 2027 and Crescent Island marks an AI inference pivot.
• Wolfspeed: Wolfspeed emerged from a prepackaged 91-day Chapter 11 on September 29, 2025 with debt cut about 70% ($4.6B), closed the Durham 150mm device fab in November 2025 and consolidated production at Mohawk Valley 200mm; it remains the world's #1 SiC substrate supplier at 33.7% share against SICC (17.1%) and TanKeBlue (17.3%). AI datacenter SiC revenue rose 50% sequentially in Q2 FY2026, CHIPS Act 48D credits of $698.6M arrived in December 2025, and nine-market demand convergence creates structural demand that no bankruptcy removes.
• ASML: ASML is the sole supplier of EUV lithography scanners, the most concentrated equipment chokepoint in the supply chain, and its ~50-60 EUV systems a year set the physical ceiling on global leading-edge fab capacity addition. High-NA EUV EXE:5000 (~20 systems a year at $380M each) defines sub-2nm, Carl Zeiss SMT optics is the binding sub-supply constraint, EUV export restrictions (China ban) rest on Netherlands/US coordination, and TSMC, Samsung and Intel share the EUV allocation.
• GlobalFoundries: GF is the world's largest specialty foundry by revenue after exiting the leading-edge race in 2018, making the specialty-process chips TSMC will not, with no commercial alternative, and it carries defense and government supply chain qualification. Fab 9 (Malta, New York) SiGe BiCMOS is the primary Western source for mmWave radar and defense RF ICs and also serves 5G mmWave and 6G research, and 22FDX FD-SOI at Dresden serves ultra-low-power IoT.

AI & Compute

The AI and compute tier covers the companies whose silicon decisions are driving the largest demand signals in the semiconductor supply chain today. NVIDIA's GPU concentration mirrors TSMC's foundry concentration on the demand side: ~80% of AI training accelerator revenue concentrated in one company, manufactured at one foundry, packaged with one advanced packaging process, using one primary HBM supplier. The hyperscaler custom ASIC programs and AMD's MI400 series represent the structural response — but they compete for the same upstream TSMC, CoWoS, and HBM supply chains rather than relieving them.

• NVIDIA: NVIDIA holds ~80% of AI training accelerators, a $1 trillion Blackwell plus Vera Rubin order backlog through 2027, ~19% of TSMC revenue as its #1 customer and ~60% of TSMC CoWoS allocation; Vera Rubin (TSMC N3, eight HBM4 stacks, 576GB Superchip) was cut from 2M to 1.5M units on HBM4 delays and ODM mass production moved from June to September, with HBM4 split SK Hynix 70% and Samsung 30% and Micron HVM confirmed at GTC March 2026. The CUDA ecosystem is the durable moat; other threads are the proprietary CoWoS-L packaging bottleneck, the Rubin Ultra roadmap, Groq 3 LPU ($20B acqui-hire, Samsung 4nm, 1-2K racks in 2026), Space-1 Vera Rubin orbital compute (GTC 2026), China share near zero after the April 2025 H20 ban, the $5B Intel stake and a $2B Marvell investment.
• AMD: AMD is the credible NVIDIA alternative and the most successful server CPU turnaround in semiconductor history, 100% dependent on TSMC: MI350X (CDNA 4, TSMC N3P, 288GB HBM3e) is in production, the MI400/Helios rack (TSMC N2, 432GB HBM4, 72 MI455X, UALink, 3 exaFLOPS, Q3 2026) is its most competitive AI platform yet, EPYC Turin holds ~40% of x86 servers and Venice (256-core Zen 6, TSMC N2) follows in H2 2026. OpenAI 6GW (with an equity option) and Meta 6GW custom MI450 (19.6 TB/s HBM4) are the largest non-NVIDIA AI chip commitments ever, Oracle is lead partner, AI accelerator share is projected to rise from 4-9% to 18%, and Strix Halo, Radeon RX 9000 and the XCD plus IOD chiplet architecture round out the portfolio.
• Huawei / HiSilicon: Huawei is the defining case study in export control policy and domestic self-sufficiency: Ascend 910C (SMIC N+2, ~60% of H100 inference) shipped 805K units in 2025, CloudMatrix 384 surpasses NVL72 aggregate PFLOPs at 4.1x the power, and Ascend 950PR (1.56 PFLOPS FP4) plans 750K units in 2026 with ByteDance at $5.6B, with Ascend 920 also on the roadmap. HBM binds output more than SMIC wafer capacity (CXMT ~2M stacks in 2026 against >1M die capacity), the TSMC die bank (2.9M dies via Sophgo) is exhausted, Kirin 9000S/9020 on SMIC N+2 DUV-only 7nm proves domestic manufacturing at a ~7nm-equivalent ceiling, Beijing issued a GPU directive in September 2025, and the April 2025 H20 ban leaves Ascend as the only legal AI accelerator for Chinese buyers.
• ASML: ASML is the most cross-referenced missing spotlight across mfg-process-nodes.html, fab clusters and all leading-edge foundry spotlights, and its EUV scanner production rate, the binding physical ceiling on all leading-edge capacity addition, is referenced more often on SX than any other fact without a home page. EUV uses CO2 laser and tin droplet plasma, High-NA EUV EXE:5000 ramps at ~20 a year at $380M each, Carl Zeiss SMT optics is the sub-supply constraint, and TSMC, Samsung and Intel are the primary EUV customers.
• Apple Silicon: Apple is TSMC's second-largest customer (~18-20% of revenue) and the anchor N2 customer with >50% of initial supply pre-purchased, Apple Intelligence NPU demand drives N2 independently of raw performance competition, and four 2nm chipsets are in development for the iPhone 17/A19 cycle. The C1 modem (iPhone 16e, first in-house 5G modem) displaces 20-25% of Qualcomm revenue, the most commercially significant supply chain transition in smartphone history, and A18/A18 Pro and the M4 family run on N3E.
• Broadcom: Broadcom dominates Ethernet switch ASICs and is the largest custom AI ASIC co-design partner for hyperscalers (Google TPU v5/v6 under a 2031 long-term agreement, Meta MTIA, OpenAI Titan), with ~15% of TSMC CoWoS allocation and rising toward a top-3 customer. Its custom ASIC programs add structural TSMC and CoWoS demand and do not substitute for NVIDIA, with which it partners on NVLink Fusion.
• Qualcomm: The dominant 5G modem IP licensor and Android flagship SoC supplier faces the most commercially significant customer disruption in its history as Apple's C1 modem displaces 20-25% of handset revenue, while its IP licensing model stays structurally durable regardless of hardware share. Snapdragon X Elite (Windows ARM PC) and Snapdragon Ride (a $45B+ automotive pipeline) are the diversification bets, Snapdragon 8 Elite runs on TSMC N3E and Samsung 2nm discussions are ongoing.

Automotive, Robotics & Industrial

The automotive and industrial tier covers the companies whose silicon programs are driving the most structurally complex supply chain dynamics in the physical economy. Tesla's Terafab vertical integration program, the humanoid robot analog/mixed-signal supply gap, Infineon's SiC dominance in automotive power, and the space compute semiconductor stack are all generating demand signals the semiconductor industry was not designed to accommodate at current scale or on current timelines.

• Tesla EV & AI: Tesla runs the most vertically integrated automotive semiconductor program, a three-chip family of AI5 (dual-sourced at TSMC Arizona and Samsung Taylor), AI6 (Samsung Taylor SF2P, dedicated captive, $16.5B through 2033) and AI7/D3 (rad-tolerant, Terafab orbital target). Terafab (Tesla/SpaceX/xAI JV announced March 21, 2026, Intel joining April 7 as IFS anchor, $20-25B, North Campus Giga Texas) is the most discussed new fab of 2026, Optimus drives humanoid semiconductor demand independently of automotive programs, and speculation about a Tesla acquisition of Intel raises Committee on Foreign Investment in the United States (CFIUS) questions.
• Humanoid Robots: Each robot carries 300-400 analog and mixed-signal ICs, dominated by motion control and sensing (GaN motor drive ICs, magnetic position encoders, MEMS IMUs, force-torque sensors, BMS ICs, current sense amplifiers), from supply chains not sized for million-unit production, making this the most undercovered semiconductor demand story of 2026-2030; the inference SoC gets the headlines, and the proprioceptive layer is the binding supply chain constraint. GaN joint drive supply (EPC, TI LMG, Navitas) has a gap, ams-OSRAM AS5047P encoders are concentrated, force-torque sensor supply is a void, IMUs come from Analog Devices ADIS and Bosch BMI088, and the platforms are Tesla Optimus (Austin, Texas), Figure (Sunnyvale, California), Agility Robotics (Corvallis, Oregon), 1X (Moss, Norway) and Unitree (Hangzhou, China), with China bifurcating the humanoid supply chain.
• Data Center: Hyperscaler custom ASICs (Google TPU v5p/v6, Amazon Trainium2/Inferentia2, Microsoft Maia 100, Meta MTIA v2) are structural TSMC N5/N3 and CoWoS demand additions that compete with NVIDIA for the same supply chain, and the OpenAI Titan chip is a Broadcom co-design on TSMC N3. AWS AI revenue ran at a $15B rate in Q1 2026 ("ascending rapidly"), CoreWeave and Meta signed a $21B compute agreement, AMD MI450's first 1GW deployment begins in H2 2026, hyperscaler capex is tracked through 2027, and AI cluster power density is driving GaN PSU and 48V rack adoption.
• Mobileye: Mobileye is the largest ADAS SoC supplier by unit volume, with EyeQ5 (TSMC 7nm) and EyeQ6H (TSMC N5, AEC-Q100 automotive qualified) referenced across SX automotive pages and REM crowdsourced HD mapping as a moat. It is an Intel subsidiary facing strategic tension, SuperVision L2+ is in volume production, and Chauffeur L4 robotaxi competes with NVIDIA DRIVE Thor, Qualcomm and Tesla FSD for the next platform generation.

Sector Programs

Sector programs focus on specific supply chain programs, platforms, or ecosystem dynamics rather than a single company. They cover the supply chain story of a technology platform as a system — including multiple companies, multiple device categories, and the convergence dynamics that make the program supply-chain-significant at the system level.

• Starlink / SpaceX: The SpaceX D3/AI7 orbital compute chip is the most ambitious commercial-foundry-for-space program, Terafab targets 80% of its output at orbital AI compute, and SpaceX vertical integration (launch vehicle, satellite, chip design) gives it a deployment pathway no other orbital compute program can replicate. NVIDIA's Space-1 Vera Rubin Module (GTC March 2026, 25x H100 AI compute, no ship date with thermal unsolved) validates the commercial orbital compute category, IGX Thor is radiation-approved and in orbit now with Planet Labs, Jetson Orin completes the three-tier NVIDIA space stack, and Aetherflux plans the first datacenter satellite in Q1 2027.
• SiC Nine-Market Convergence: Nine demand markets draw on the same SiC substrate funnel at once: EV traction inverters, EV onboard chargers (OBC), EVSE DC fast chargers, BESS power conversion systems (PCS), solar string inverters, industrial VFDs, solid-state transformers (SST), datacenter UPS and humanoid robot joint drives, served through automotive OEMs, BESS operators, solar inverter OEMs and AI datacenter operators. Wolfspeed's September 2025 Chapter 11 emergence and Mohawk Valley ramp are the live stress event, the 150mm to 200mm transition is the primary volume multiplier, SST is the sixth emerging demand wave, AI datacenter SiC diversifies supplier revenue, and SICC and TanKeBlue lead the Chinese rise, with growth rates and per-supplier capacity status tracked.
• China Semiconductor Bifurcation: Bifurcation spans foundry (SMIC vs TSMC), DRAM (CXMT vs Samsung, Micron, SK Hynix), NAND (YMTC), equipment (NAURA and AMEC vs ASML, Applied Materials, Lam Research, KLA), EDA (Empyrean vs Synopsys and Cadence) and AI compute (Huawei Ascend vs NVIDIA), each layer reinforcing the others toward a separate supply ecosystem, the most structurally significant geopolitical supply chain event of the decade. SMIC N+2 is capped at DUV-only manufacturing, the Beijing GPU directive followed the H20 ban, gallium and germanium export controls give China counter-leverage, domestic equipment makers include NAURA, AMEC, SMEE and SiCarrier, and 28nm is the geopolitical policy frontier.
• Tesla Terafab: The Tesla/SpaceX/xAI JV announced March 21, 2026 is the most discussed new semiconductor manufacturing announcement of 2026, and Intel joined April 7 as manufacturing and packaging partner, most credibly as an IFS anchor customer arrangement with Intel 18A plus Foveros packaging as the production backbone. The $20-25B cost is not in Tesla's 2026 capex, the North Campus Giga Texas site shows ground clearing and 5.2M sq ft permits, 80% of output targets orbital AI compute, and the Tesla Terafab Supply Chain page details the three-chip architecture, Samsung Taylor captive arrangement, Intel partnership structure, AI6 vs AI7 distinctions, Optimus integration and acquisition speculation with CFIUS analysis.

Related Coverage

AI Data Center Spotlight | AMD Supply Chain Spotlight | ASML Supply Chain Spotlight | China Semiconductor Bifurcation | Fabs as Semiconductor Consumers | Huawei / HiSilicon Supply Chain Spotlight | Humanoids & Robotics Spotlight