Skip to content
Back to articles
AI/Market14分

NVIDIA/TSMC/ASML Q1 2026 Earnings: AI Semiconductor Supply Chain Analysis

木村 啓介Senior Semiconductor Analyst
2026-04-2114分
NVIDIATSMCASMLSemiconductorEarningsJapan Tech

What Q1 2026 Earnings Reveal About AI Semiconductor Reality

The third week of April 2026 brought results from all three core players in the AI semiconductor supply chain — and every one of them beat market expectations. NVIDIA's data center segment revenue hit $52.1 billion, up 71% year-over-year and 9% sequentially, comfortably above the $49.8 billion consensus. Gross margin came in at 75.8%, improving 0.6 percentage points quarter-over-quarter, confirming that pricing discipline has held even amid the shift to the Blackwell generation.

TSMC ramped its 2nm node (N2) into mass production in early 2026, with advanced processes (N3 and below) now accounting for 34% of Q1 revenue. ASML shipped 18 EUV tools in the quarter, including three High-NA EUV systems, pushing its backlog to a record €40.5 billion. Read together, these three sets of results paint a fairly clear picture of both the durability of AI infrastructure spending and how it is flowing through to Japan's equipment sector.

NVIDIA: Mix Shift from H200 to B200/GB300

NVIDIA's B200 (Blackwell) shipment mix surged from 41% in Q4 2025 to 63% in Q1 2026. The B200 HGX board carries an average selling price of around $42,000; the GB300 NVL72 rack system runs roughly $3.4 million per rack, and the CFO noted that rack-system gross margins are approximately 4 percentage points higher than board-only sales. Estimated GB300 rack shipments in Q1 reached roughly 1,820 units, implying around $6.2 billion at the rack level.

Inventory health is the key point. Allocation letters to hyperscalers are essentially booked out four quarters ahead, and emerging cloud players such as CoreWeave, Lambda, and Crusoe have been given delivery windows extending into the first half of 2027. There is no visible sign of speculative inventory build-up heading into Q2.

Hyperscaler Capex: Five Players Combining for ~$580 Billion

Aggregating 2026 calendar-year capex guidance, the picture is: Google $118 billion, Microsoft $135 billion, Meta $102 billion, AWS $147 billion, Oracle $78 billion — a combined $580 billion. That is up 41% from $410 billion in 2025. The AI infrastructure share is estimated at an average of 68% across these five, implying roughly $394 billion directed at AI-related spend.

Oracle and Meta stand out for front-loading their budgets. Oracle plans to deploy 63% of its full-year capex in the first half, driven by GPU procurement under the Stargate contract with OpenAI. Meta is simultaneously ramping its custom silicon MTIA and buying large quantities of NVIDIA GB300 — a clear dual-track strategy. Google's TPU v6e production on TSMC N3P is gaining traction, and the internal workload share running on TPUs has risen to 52%.

TSMC N2 Capacity: Structural Demand-Supply Imbalance

TSMC is targeting monthly N2 wafer output of 80,000 wafers by end-2026, but reservations are already filled through Q2 2027. The allocation roughly divides as: Apple ~35% for M5/A20, NVIDIA ~22% for the Rubin generation, AMD ~12% for MI400-series, with Qualcomm, Broadcom, and Google splitting the remainder. Rubin sampling is expected in the second half of 2026, with volume production likely in H1 2027.

High-NA EUV adoption has been pulled forward by three to four quarters from original plans — the primary driver of ASML's record backlog. At roughly $380 million per tool, more than twice the price of a conventional EUV system, full-year 2026 shipments are expected at 13–15 units. The productivity gains from eliminating DUV steps at 2nm and beyond are modeled to lift TSMC's gross margin by around 2 percentage points over the medium term.

Stock Reactions and Hedging Strategies

On the first trading day after the earnings releases, NVIDIA gained 6.8%, TSMC's ADR rose 4.2%, and ASML was up 3.1%. However, the decline in implied volatility was limited — the options market continues to price in further upside surprises before the next earnings round. ETF inflows into SOXX and SMH totaled $3.4 billion in the third week of April alone.

Hedging approaches that worked well include: (1) long NVIDIA against ATM puts on the Philadelphia Semiconductor Index (SOX) to isolate stock-specific alpha; (2) long TSMC versus short Samsung ADR as a foundry market share pair trade; and (3) pairing hyperscaler-capex longs with long-dated U.S. Treasuries (TLT) as a macro counter-weight. The third combination proved particularly effective in mid-February when recession fears flared.

Ripple Effects on Japanese Names: Renesas, Advantest, Disco, Tokyo Electron

Japan's semiconductor equipment and materials sector is among the clearest beneficiaries of this cycle. Advantest's SoC tester V93000 EXA has become essential for Blackwell and Rubin-generation validation; its Q1 orders came in 58% above the prior year, beating estimates. Disco maintained dominant share in CMP and dicing/grinding equipment tied to CoWoS packaging and raised its full-year revenue guidance.

Tokyo Electron (TEL) is embedded in the 2nm workflow with its Telius and Tactras etch tools; the advanced logic share of TEL's revenue mix has reached 44%. Screen Holdings (SPE) retains meaningful presence in cleaning processes around EUV. Renesas is winning not on the AI chip itself but as an NVIDIA-certified supplier of power management ICs for AI servers — its non-automotive segment grew 23% in Q1, its strongest pace in years.

From an investor perspective, the TOPIX Semiconductor sub-index is up 14.3% year-to-date versus 9.1% for the SOX, outperforming on the combination of yen weakness and domestic investment (TSMC Kumamoto, Rapidus Chitose). As long as those tailwinds persist, this outperformance is likely to carry through H2 2026.

Three Watchpoints for H2 2026

First: NVIDIA Rubin sample shipment timing. As long as Q3 2026 does not slip, downward pressure on B200 pricing should remain contained. Second: TSMC N2 yield. Whether the current yield in the high-60% range reaches 75% by end of the second half or slips to Q1 2027 will reshape the cost curve for the entire semiconductor industry. Third: hyperscaler "AI ROI" disclosure. As Microsoft has begun breaking out Azure AI gross margin detail, investors are entering a phase where they will scrutinize returns on deployed capital more rigorously.

For Japanese names, yen appreciation is the only meaningful headwind. For hedged positioning, an equal-weight basket of the three major equipment names — TEL, Advantest, and Disco — continues to offer an attractive risk-reward profile.

Let's solve your technical challenges together.

KGA IT Solutions delivers AI, cloud, and DevOps expertise to address your specific challenges.

Contact Us