Stock-Market Research Dispatch — 2026-05-27

Auto-generated by daily research routine. Not a source; not ingest-pending material. Summary of what changed in the wiki today.

Top of mind

Nvidia Q1 FY2027 was materially higher than the wiki believed — and the moat may be actively defended, not passively eroding. The wiki carried $44.1B as Nvidia's Q1 FY2027 figure; the actual was $81.6B (+85% YoY), with Data Center alone at $75.2B (+92% YoY) and DC networking at $14.8B (+199% YoY). Q2 guide is $91B, the first time in this cycle Nvidia's guidance has exceeded Street consensus. Three straight quarters of acceleration. The correction is significant: the thesis was underestimating both the scale and the momentum of Nvidia's current revenues.

Simultaneously, the Nvidia-Groq $20B acqui-hire (December 2025) surfaced as the most important counterevidence yet to cuda-moat-erosion-at-inference: Nvidia spent $20B to neutralize what Senators Warren/Blumenthal confirmed was "genuinely faster and more energy-efficient" inference hardware. This is not a company watching its moat passively erode — it's a company paying $20B to pre-empt the erosion. Conviction revised medium → low-medium. The structural tension is real (TPU, Trainium, wafer-scale all live), but Nvidia's willingness to acquire its inference competitors reshapes the timeline.

New wiki pages

Significant wiki updates

Nvidia Q1 FY2027 corrected to $81.6B (nvidia): All downstream thesis pages updated. AMD-Meta $60B deal (12GW committed) filed with AMD entity. Nvidia Feynman chip uses dual foundry — Intel 14A for I/O die (~25% of chip area), TSMC for compute tile. This is the first confirmed Nvidia-Intel 14A customer signal. Intel picks-and-shovels angle improves; intel-18a-yield-vs-tsmc updated.

Copper: ICSG flips to 150K MT deficit — Goldman is now the isolated outlier (copper-supercycle-ai-data-centers): The prior calibration (May 18) downgraded copper on Goldman's 490K surplus call. That downgrade is reversed. ICSG's May 2026 projection is a 150K MT deficit, the first ICSG deficit since 2009. LME hit a record $6.65/lb on May 13. Goldman's revised surplus is now ~310K MT above the authoritative international body. Conviction medium → medium-high. FCX Grasberg acid ban confirmed; SCCO more exposed (acid ban) so SCCO conviction slightly downgraded.

AMKR CoWoS thesis graduates from hypothesis to open (amkor-cowos-tsmc-overflow-osat): Q1 2026 $1.68B (+27%), HDFO platform 5+ customers, AI/HPC packaging "tripling" by 2028, Investor Day $6.7B→$11B/2030. TSMC outsourcing to Amkor AND ASE confirmed. Graduation remaining: explicit earnings disclosure of CoWoS revenue from TSMC contracts.

IREN/Nvidia DSX: Nvidia official press release confirms (iren-nvidia-dsx-ai-factory-operator): Not a rumor — "NVIDIA and IREN announce strategic partnership to accelerate deployment of up to 5 gigawatts of AI infrastructure." The "preferred AI factory partner" graduation criterion is substantially met. Remaining: Sweetwater Phase 1 (500MW) construction start + first Nvidia chip order tranches.

HBM shortage extended to 2028 (hbm-supply-bottleneck): IDC now calls it a "Global Memory Shortage Crisis." Blackwell Ultra requires HBM3E; Rubin requires HBM4. SK Hynix supplies 2/3 of Nvidia's HBM4 for Rubin. Intel-SK Hynix iHBM (30% thermal reduction) is a new form factor. The shortage is not a 2026 story — it runs through 2028 on current trajectories.

Watchlist

Undervalued candidates (today)

• amkor-cowos-tsmc-overflow-osat — AMKR (~$77.22). Q1 beat, HDFO platform growing, AI packaging tripling, Investor Day $11B/2030 target. TSMC CoWoS outsourcing to Amkor is structural. Valuation context unknown (WebFetch blocked), but the thesis is substantially confirmed while the ticker remains absent from mainstream CoWoS discussions.
• corning-ai-optical-connectivity — GLW. Nvidia named as anchor customer for a 10x US optical capacity expansion. Corning is not in any AI picks-and-shovels analysis this wiki has seen. If AI cluster optical demand materializes at the pace Nvidia's capex implies, GLW could be a deeply undercovered beneficiary.

New chains (hypothesis-stage)

• corning-ai-optical-connectivity — GLW. Graduate: Corning Q2 2026 earnings call disclosing AI-specific optical connectivity revenue >5% of optical communications segment.
• ai-power-delivery-ics-consolidation — MPWR, VICR, ADI. Graduate: second M&A transaction in AI power delivery category, or MPWR/VICR calling out AI compute power management as >15% of revenue.

Contradicted / weakened

• cuda-moat-erosion-at-inference — Nvidia-Groq $20B acqui-hire (December 2025) is direct counterevidence. Nvidia is actively defending the inference moat. Conviction medium → low-medium. The structural long-run erosion thesis (TPU, Trainium, wafer-scale) remains; but the near-term re-rate risk from inference alternatives has been substantially reduced by Groq's neutralization.

Calibration logged

1. Copper ICSG flip — Goldman surplus was an isolated outlier; ICSG 150K MT deficit confirms structural thesis. Conviction upgraded medium → medium-high. Pattern: overconfidence/sell-side-authority.
2. CUDA Groq inversion — Nvidia's $20B Groq acquisition is active moat defense, not passive erosion. Conviction revised medium → low-medium. Pattern: narrative-inertia — counterevidence that lives outside the thesis's usual source pool was slow to integrate.

Open questions for human review

1. AMKR: add to watchlist? CoWoS thesis now substantially confirmed. AMKR at ~$77.22 is the only named ticker in the CoWoS overflow hypothesis. Worth pulling a full valuation snapshot and deciding whether this is actionable.
2. GLW: research priority? Corning is the most novel uncovered beneficiary surfaced today. The optical connectivity 10x build is Nvidia-anchored. Should this get a full autoresearch pass before the next Corning earnings call?
3. Nvidia $91B guide: thesis implications? At $81.6B Q1 and $91B guide, Nvidia's revenues are scaling well beyond anything the wiki modeled. Does this change the conviction on downstream theses (CoWoS, HBM, picks-and-shovels) by accelerating timelines?
4. Apple-Intel deal: 19+ days and still preliminary. The deal is taking longer to finalize than expected. Is this normal negotiation lag, or a signal the deal terms are under revision? VLSI June 14-18 is the next hard catalyst.
5. Podcast feeds (11th consecutive 403): Persistent network policy block. This is a container configuration issue, not a content issue.

What I looked at

• Step 1 topics: AMKR CoWoS Q1 2026; China MRL/REO quota/NdPr prices; HBM supply bottleneck update; nuclear AI datacenter PPA update; Section 232 Taiwan quota/cost penalties; weekly macro scan (Nvidia correction + AMD + Feynman); Apple-Intel deal May 27; Samsung foundry May 27; copper supercycle update; regulatory/antitrust (Groq/DOJ); M&A and partnerships scan
• Step 2 macro buckets: Semis & AI infrastructure (Nvidia/AMD/Feynman); M&A and partnerships (Corning/ADI-Empower/Tata-ASML/IREN)
• Podcast episodes transcribed: 0 (all RSS feeds HTTP 403; 11th consecutive failure)
• Sources promoted: 11 (all from May 26-27); sources ingested: 11
• Valuation snapshots refreshed: 0 (WebFetch 403 — 11th consecutive)
• New hypothesis chains opened: 2 (corning-ai-optical-connectivity, ai-power-delivery-ics-consolidation)
• New entities created: 2 (amkor/AMKR, amd/AMD)
• Calibration logged: 2 (copper ICSG flip, CUDA Groq inversion)