Autoresearch: Semis & AI infrastructure macro scan — May 23-26 2026

Generated by /autoresearch on 2026-05-26. 3 rounds of search (full). All content from search snippets (WebFetch 403). Treat as raw material. Context: vault/projects/stock-market

Summary

The dominant finding is NVIDIA Q1 FY2027 (reported May 20, 2026): $81.6B revenue (+85% YoY), data center $75.2B (+92% YoY), with Q2 guidance of $91B — above Wall Street estimates for the first time this cycle. This is three consecutive quarters of acceleration, and the wiki's NVDA entity currently shows "$44.1B" (Q4 FY2026 data), making the Q1 actuals a significant gap to fill. The second major uncaptured finding is the AMD-Meta $60B / 6GW deal (February 24, 2026) — the largest single hardware procurement in Magnificent Seven history, no AMD entity page exists in the wiki. Third: Nvidia Feynman (2028) will use a dual-foundry approach — TSMC A16 for the GPU die, Intel 14A for the I/O die with EMIB packaging (DigiTimes-sourced, unconfirmed but multiply corroborated), directly reinforcing the tsmc-saturation-to-intel-anchor-stack thesis with Nvidia as a new Intel 14A anchor.

Findings

Nvidia Q1 FY2027: blockbuster results, three straight quarters of acceleration

Nvidia Q1 FY2027 (Feb–Apr 2026, reported May 20, 2026):

• Total revenue: $81.6B (+85% YoY), beating consensus $78.9B (TIKR, May 2026)
• Net income: $58.3B vs estimate $42.9B
• Data Center: $75.2B (+92% YoY, +21% QoQ) — record (Futurum, May 2026)
  • DC compute: $60.4B (+77% YoY)
  • DC networking: $14.8B (+199% YoY) — InfiniBand/Spectrum-X Ethernet/NVLink
• Q2 FY2027 guide: $91B vs Street estimate $86B — first time this cycle guidance has exceeded Street consensus (Kiplinger, May 2026)
• 25x dividend increase + $80B buyback — cash generation materially accelerating
• Blackwell (including GB300 Ultra) accounts for large majority of DC compute; adopted by "every major hyperscaler, every cloud provider, and every model builder"
• China DC: zero revenue in Q1 FY2027, vs $4.6B in Q1 FY2026 — H20 charge was prior-period event; ongoing zero China guidance
• Jensen Huang: "Demand has gone parabolic. Agentic AI has arrived."

Wiki gap: The nvidia entity currently reads "Q1 FY2027: $44.1B (+69% YoY)" — this is Q4 FY2026 data. Q1 FY2027 actuals ($81.6B) need to update the entity. The data center networking figure ($14.8B, +199%) confirms the InfiniBand/Spectrum-X play as a real and fast-growing revenue line alongside compute.

AMD MI400 and the Meta $60B deal: real but structurally different from NVDA

AMD unveiled the full Instinct MI400 series at CES 2026 (January 2026):

• MI455X (flagship), MI440X (enterprise), MI430X (sovereign/HPC) — CDNA 5 architecture
• 432GB HBM4, 19.6TB/sec memory bandwidth
• MI455X powers the Helios rack-scale platform: up to 3 AI exaflops per rack (shipping Q3 2026) (Tom's Hardware, CES 2026)
• MI500 (CDNA 6) planned for 2027, claiming 1,000x performance increase over MI300X (Data Center Dynamics, 2026)

AMD-Meta $60B deal (February 24, 2026):

• 5-year, $60B agreement; 6GW of custom AMD MI450 GPUs (co-engineered with Meta) starting H2 2026 (Jon Peddie Research, Feb 2026)
• Performance-based warrants: Meta can acquire up to 160 million AMD shares
• First 1GW deployment begins H2 2026; MI450 = CDNA 5, 432GB HBM4, 20 PFLOPS FP8
• Meta is "starting to train on AMD" — training (not just inference) is the use case (SemiAnalysis, 2026)

AMD total committed GPU deployments from hyperscaler deals: ~12GW (Meta 6GW + Microsoft/OpenAI deal ~6GW combined). MI400 described as "potentially competitive with Nvidia VR200 NVL144" in H2 2026. However, NVDA's $75.2B data center Q1 vs AMD's $7.2B-run-rate GPU line makes this a 10:1+ revenue gap.

Wiki gap: No AMD entity page exists. AMD-Meta deal is the largest single merchant-silicon hardware procurement outside Nvidia and is not captured anywhere in the wiki.

Nvidia Feynman (2028): dual-foundry TSMC A16 + Intel 14A

Nvidia's Feynman architecture (2028 production, 2029 deliveries):

• GPU die: TSMC A16 (or custom N2 variant) — confirmed as primary
• I/O die: Intel 18A or Intel 14A — partial adoption with Intel's EMIB packaging (WCCFTech, 2026)
• Estimated 75% TSMC / 25% Intel silicon split (igorslab.de)
• Memory: HBM5 (4-layer 3D heterogeneous memory stack: register, L1/L2, stacked SRAM, HBM5 video memory) (NADDOD Medium, March 2026)
• Intel 14A mass production timing: 2028 — aligned with Feynman launch
• Status: DigiTimes-sourced, not officially confirmed by Nvidia or Intel; described as "low risk" approach by analysts since I/O die assignment doesn't compromise main GPU die volume if Intel yield misses (Notebookcheck, 2026)

Implication for Intel IFS thesis: Even 25% of Nvidia Feynman I/O dies at Intel 14A would be a significant volume anchor. Combined with Tesla (first 14A customer confirmed), Apple A21 (~60M/yr, 2028), and potential AMD foundry work, Intel 14A has a multi-customer pipeline forming for 2028. This extends the tsmc-saturation-to-intel-anchor-stack mechanism by one generational step — if confirmed, Nvidia becomes the fifth major Intel 14A anchor (after Tesla, Apple, Google EMIB, Terafab).

VLSI 2026 (June 14-18): Intel 18A-P and TSMC A16 specifications

Intel 18A-P vs baseline 18A (TrendForce, May 2026):

• +9% performance at same power, OR -18% power at same performance
• +50% thermal conductivity (reduced thermal resistance) — Apple M7 thermal headroom implication
• Enhanced via additional logic VT pairs, skew corner tightening, new LP + HP devices
• SRAM Vmin matched + improved logic NBTI

TSMC A16 vs N2P (Tom's Hardware, 2026):

• 8-10% faster OR 15-20% power reduction + 8-10% density gain vs N2P
• Gate-all-around (GAA) transistors + Super Power Rail (SPR) backside power delivery
• Risk production 2026; volume late 2026/2027 (some reports cite a slip to 2027 full mass production)
• Nvidia Feynman is reportedly the "first customer for high-volume manufacturing"

CoWoS roadmap extension (new detail from TSMC symposium):

• Current: 5.5-reticle-size CoWoS
• 2028: 14-reticle-size CoWoS packages
• 2029: 48x compute leap, supporting 24 HBM5E stacks (Tom's Hardware, 2026)
• CoWoS ASP now approaching $10,000/wafer — parity with 7nm advanced logic nodes (TrendForce, April 2026)

Hyperscaler capex: $725B confirmed, no new signals in May 23-26 window

No new hyperscaler capex announcements in the May 23-26 window beyond what's in the wiki. The confirmed figures:

• Amazon: $200B 2026 (Q1 confirmed)
• Alphabet: $180-190B (raised from $75B post-Q1)
• Meta: $115-145B
• Microsoft: ~$120B
• Combined Big-5: ~$725B — up 77% from $410B in 2025 (Futurum, 2026)
• 75% of aggregate capex = AI-specific ($450B)

This is already well-captured in the wiki. No incremental news this window.

Contradictions and open questions

• NVDA wiki entity needs updating: Wiki shows "$44.1B" (Q4 FY2026); Q1 FY2027 actual was $81.6B — a ~85% gap. The entity nvidia needs the Q1 FY2027 actuals + Q2 guide ($91B).
• AMD entity page missing: The $60B AMD-Meta deal is the single largest AMD customer win ever, and AMD is now >12GW committed from hyperscalers. No AMD entity page in the wiki. This is a significant gap for the competitive dynamics section.
• Feynman Intel 14A not confirmed: DigiTimes-sourced. Multiple outlets corroborate but Nvidia has not confirmed. Treat as high-probability hypothesis, not confirmed thesis step.
• A16 timeline ambiguity: One Tom's Hardware article says "A16 slips to 2027" for full mass production; others say "volume late 2026." The distinction matters for Feynman's 2028 launch readiness.
• AMD vs Nvidia competitive read: AMD's $60B Meta deal + 12GW commitments is impressive in absolute terms but NVDA's $75.2B data center in ONE QUARTER dwarfs AMD's entire AI GPU trajectory. The market share shift is real but the magnitude is ~10:1 in favor of NVDA through at least 2027.
• HBM5 supply: Feynman requires HBM5 (not HBM4). SK Hynix, Samsung, and Micron would all need to ramp HBM5 by 2027-2028. No HBM5 production ramp visibility yet — this extends the HBM bottleneck thesis another generation.

Provenance

Rounds run: 3 of 3 (full)

Sub-questions by round:

Round 1 (broad survey):

1. Nvidia Q1 FY2027 earnings results, Q2 guidance, and H20/China impact
2. AMD MI400 / Broadcom XPU / next-gen AI accelerator announcements May 2026
3. New hyperscaler capex commitments or AI infrastructure announcements May 23-26 2026
4. Advanced packaging / CoWoS / HBM supply chain developments May 23-26 window
5. VLSI 2026 pre-conference Intel 18A-P vs TSMC A16 preview disclosures

Round 2 (drill-down):

1. Nvidia Q1 FY2027 revenue breakdown (DC vs other), Blackwell ramp — targeted: full earnings detail for wiki
2. AMD MI400 hyperscaler customer wins — targeted: competitive dynamics vs NVDA
3. TSMC A16 production timeline + Nvidia Feynman HBM5 requirements — targeted: next-gen supply chain chain

Round 3 (resolve remaining uncertainty):

1. Nvidia Feynman dual-foundry Intel 14A — confirmed or rumor? — targeted: Intel thesis step extension
2. AMD Meta $60B deal scale — real figure? — targeted: competitive dynamics magnitude

Anchor source: No Grokipedia entry (HTTP 403 — consistent with container network policy).

URLs fetched: 0 (all HTTP 403). Content from search snippets only.

Tools used: WebSearch (7 queries). WebFetch: all 403. Generated: 2026-05-26