Good morning.
Futures green, ES bid into the open. SPX closed a record 7580 on Friday (+11% above 200-dma); Tech carried the entire +5.2% May print. Biggest tape bomb: DELL +43% — LARGEST ONE-DAY GAIN IN HISTORY — after delivering the first Vera Rubin NVL72 rack to CRWV. SNOW +48% on the week. KOSPI 50 +318% in a year is the cleanest one-liner for where the money is. AVGO prints tonight — guiding >$10B in AI semis, this is the read-through for the whole cohort. MU reports 6/24 and Goldman just ripped memory OP estimates +21-24% across '27-28 in a single note (unusual size — pay attention to scale).
Asia mixed-modestly-green, bid on the memory raises. Oil $87 from $109 a month ago — convexity trade is fully alive (Nasdaq rallies more on Brent down than it falls on Brent up). 30Y YIELDS BACK TO 2007 LEVELS; bond market in a 70-month drawdown (longest ever). Week ahead is stacked: NFP Friday, AVGO tonight, Computex (Jensen confirmed NVDA displaced AAPL as TSMC's #1 at >20% of revenue), MS Build, Snowflake Summit, Cisco Live Vegas.
THEMES framing the tape:
ONE — Memory constraint rotation is ON. Goldman +22% SK Hynix '27 OP ($219B → $266B), +21% Samsung, +24% on '28. HBM tightness extends into '27-28, not just '26. Next cascade is HBM/DRAM, not power. Long MU (1,600 PT noted, getting close to $1T mkt cap), SNDK the NAND lever, MRVL the custom angle.
TWO — Capex elongation, not demand-doubt. SoftBank €75B / 5GW AI DC in France with Schneider (largest non-US AI infra of the cycle). Lattice CEO: hyperscaler capex >$1T in 2027. Anthropic raised $65B at $965B on $47B ARR. The bids keep getting bigger.
THREE — Non-AI cracks widening. 2H26 PC sales -15-18% YoY, MLCC ASPs +17%, memory 200-300% higher since mid-25. Two-track economy inside the same supply chain. ADBE bearish flag (ARW:4) is the SaaS tell — AI-native pressure is real, not theoretical.
FOUR — Stretched but not stretched-out. SPX +11% above 200-dma, short interest highest since 2012, hedge fund IT "running out of buyers." But Yardeni just raised 2026 SPX to 8,250, and June has NEVER been lower after a 10%+ two-month run. Wall of worry is the bid.
We'll hit AVGO first (tonight's the cohort read), then MU/SNDK/MRVL on the memory rotation, then get to the photonics/opticals (LITE, AAOI, COHR — parabolic but rolling over into the 40-week).
Supplementary Coverage
NVDA
Vera Rubin is real, not paper. DELL shipping first NVL72 rack to CRWV with 75TB HBM and 260TB/s NVLink confirms the platform exists and is deploying. Jensen displacing Apple as TSMC's #1 customer at >20% is the structural call — foundry mix is concentrating into a single point of failure. Memory is the next binding constraint; Goldman's 21-24% raises on SK Hynix/Samsung 2027-28 OP confirm tightness extends well past 2026. Anthropic禁令 is net bullish for NVDA — frontier models now treated as export-controlled products means US labs will keep pouring capex into domestic compute, and Anthropic's demand is replaceable at scale.
MU
Goldman's $1,600 PT and 2027 OP raise are scale-movers, not incremental notes. HBM tightness extending to 2027-28 is the regime-change call — memory is becoming more like foundry/utilities, less like commodities, because long-term contracts are absorbing cycle volatility. Sandisk's Q1 print (+251% revenue, gross margin 22.5% → 78.3% on essentially zero volume growth) is the read-through. MU 6/24 print is the cleanest validation event of the cycle. The bull case is that 200-300% memory pricing is structural, not transitory; the bear case is that MU's earnings are now priced for perfection and any demand softening is asymmetric.
TSM
NVDA >20% of revenue means TSMC is structurally tied to one customer's ramp curve. CoWoS is zero-sum — when NVDA pulls, Apple/AMD/Qualcomm absorb allocation loss. Read-through: TSMC's pricing power is reinforced but concentration risk is real. Xi ramping 邦交国 pressure over fighter jet shows is the slow-burn geopolitical that compounds the structural single-customer risk. 2H26 PC demand destruction is distribution-only for TSMC — PC is a rounding error vs AI server pull. APAC rate cuts (RBA, Japan long bond retreat) provide a valuation tailwind.
INTC
18A yields ~50% with Panther Lake at 7x volume means Intel has a product, not a foundry. External allocation is limited to Apple/Google/NVDA/AWS — Intel Foundry is not a meaningful external supplier this cycle. The "packaging is the new Moore's Law" thesis (hybrid bonding, 3D integration, $80B+/year TAM) is Intel's structural defense even if 18A node economics disappoint. The bifurcation is real: bad node economics, real packaging economics. MS short report rebuttal centers on this pivot. Bull case is misunderstood turnaround; bear case is financial engineering masking weak foundry fundamentals.
AMZN
Anthropic禁令 is a clean bearish catalyst. Anthropic was a marquee Bedrock tenant and now their compute is migrating to GCP. AMZN's frontier AI positioning just got downgraded one full tier. Q2 cloud growth guidance is the binary catalyst — if Anthropic workload migration shows up in Q2 AWS numbers, the bear case is validated. The pair trade is clear: long GOOGL cloud vs short AMZN cloud. $17.5B bank loan for AI capex means any cloud AI slowdown is amplified through the capex commitment.
GOOGL
Anthropic禁令 is an unexpected windfall. Anthropic compute is shifting from AWS to GCP — TPU + GCP + DeepMind + Anthropic workload stack now forming inside GOOGL. GCP share has been the only hyperscaler gaining (AWS 49% → 42%, GCP 16% → 22% over 2022-2026) and Anthropic migration accelerates this. Q2 capex guide is the next catalyst — absorbing Anthropic workload requires new data center capacity. The "US AI infra safe harbor" narrative positions GOOGL as the structurally bullish name in mega-cap tech from the禁令, and the pivot magnitude is underpriced.
META
Incremental GW in 2026 is the most of any AI lab — META may be ahead of OpenAI/Anthropic on the ~6GW EoY'26 Dwarkesh prediction. Less incremental inference demand competing for incremental capacity without an exploding agent business means META has more compute runway per workload. The 12-month question: can the new MSL team produce frontier-quality work? If yes, META is structurally underpriced as a frontier challenger. Watch the talent-product output ratio; that's the real catalyst, not capex announcements.
AAPL
Losing #1 TSMC customer status is structurally bearish for Apple's foundry pricing power. Smart glasses delay to late 2027 removes a growth catalyst. China Q2 consumer spending potentially going negative post-COVID is the demand-side risk — iPhone is the most consumer-sensitive name in mega-cap tech. Offset: N1X PC chip partnership is a read-through for Apple's foundry allocation priority since external 18A allocation is limited. Apple TV "highest batting average in streaming" is real but immaterial to the multiple.
DELL
+43% on the week is the largest single-day gain in history. The Vera Rubin NVL72 rack shipment validates DELL's AI server positioning at exactly the right moment. Smart call from the COO flagging HDD as the next shortage component driven by AI high-capacity storage demand — this is a credible supply chain call from a major OEM. The trade is in the AI server basket with HPE, but DELL is the cleaner expression post-print. Mr. Dell cover photo tease ahead of earnings is a sideshow but signals confidence.
CRWV
First Vera Rubin NVL72 rack from DELL is a credibility win. CRWV is positioning as a neocloud with privileged NVDA allocation, not a generic GPU renter. The neocloud tier is stratifying — CRWV/IREN/NBIS as differentiated vs generic. This is the playbook for capitalizing on Nvidia's lead-lag rotation: get the rack, get the narrative, get the multiple re-rating.
SNDK
+251% revenue with gross margin going from 22.5% to 78.3% on essentially zero volume growth is not a normal cycle. This is a regime change in NAND pricing power. The risk is symmetric: when supply-demand rebalances, prices/gross margins revert and revenue may decline even as units rise. The cycle is real but MU's 6/24 print will validate or reject. SNDK is the canary in the coal mine — watch the MU print to see if SNDK's pricing is sustainable or one-time.
AVGO
$10B+ AI semis guide is the single most important earnings datapoint this week. InP photonics revenue projected to grow 8x (2025-2030) to $4.5B. AVGO is the second-largest beneficiary of optical interconnect buildout after Lumentum, and the only one with custom ASIC + optical + networking combined. As hyperscalers diversify away from pure NVDA exposure, AVGO captures the custom silicon layer. Rate of change is accelerating as agentic AI drives custom CPU/GPU/ASIC co-design. This is the cleanest read on the capex elongation trade.
COHR
Top pick in InP photonics with 6-inch wafer + laser capacity as the differentiator. Revenue projection: $125M (2025) → $4.3B (2030) — 34x in 5 years, the fastest growth in the report. NVDA pushing suppliers for 20x InP laser capacity increase but vendors only committing 12x — COHR is one of the beneficiaries of the undersupply. Listed in the quantum compute optical interconnect stack. The bottleneck is real and the supply is structurally short.
AAOI
InP photonics 35x revenue growth (2025-2030) with UHP CW CPO market share ~10% projected. Top pick alongside COHR. But parabolic action showing signs of rolling over — watching for late summer/early autumn pullbacks toward 40-week moving averages. Same trade view as LITE. Don't chase; let it come to you.
LITE
InP photonics $600M (2025) → $9B (2030) — 15x in 5 years, the largest absolute revenue projection. Parabolic with rollover signs. The trade is the same as AAOI: take profits, wait for 40-week moving average test. The fundamentals are real but the tape is extended.
CRM
Now Assist AI revenue target raised from $1B to $1.5B. AI-native rivals reportedly cannot rip out the database — implementation partner perspective confirms the data moat is alive. The Anthropic禁令 accelerates de-grossing in software broadly, and CRM is in the crosshairs. The AI agent narrative借力Anthropic is a tailwind but the software sector re-rating is not over. Quality compounder with execution risk from AI displacement narrative.
NOW
Same $1B → $1.5B Now Assist raise. Bernstein PT $236, DA Davidson $190, Evercore $150 — sell-side dispersion is wide. "Signs of Agentic AI traction" is the operative phrase. NOW is the cleanest software expression of the agentic AI workflow thesis, alongside CRM. AI harness wars of 2027 framing positions NOW as the workflow orchestration layer. The thesis is intact and the data points are validating.
ORCL
Q4 disclosure: capex $70B, FCF negative, need to raise $20B more. This is the "AI capex融资恐惧" transmission — the AI capex layer is going from cash-flow-funded to debt/equity-funded at the hyperscaler tier. OpenAI price war is the demand-side counterpart — capex up, token prices down, gross margin squeezed from both sides. Adobe CFO departure and Adobe beat-but-sell is the same dynamic. The capex funding model is breaking and the market is starting to price it.
MSFT
Cobalt CPU in the agentic AI custom CPU wave. Copilot has "lowest presence, lowest usability, zero topic heat" per Chinese commentary — execution is underwhelming despite the leverage. The Anthropic禁令 paradox: as a major Anthropic partner, it's a direct hit; but as a frontier AI player, the export control narrative reinforces US capex. Net: ambiguous. Watch Copilot adoption metrics in Q2.
TSLA
OpenAI Robotics division directly overlaps with Optimus. Read as narrative threat, not actual competition — TSLA's manufacturing, supply chain, and real-world data flywheel is hard to replicate. But Optimus is a key option value in the bull thesis; any AI lab entering robotics compresses the option's volatility. FSD endorsement ("the safest I ever feel is when FSD is driving") is bullish. SpaceX/Tesla merge in 2027 with 80%+ chance is the secondary thesis.
NBIS
Nebius Token Factory built around optimization, orchestration, and agentic deployment — not basic GPU rental. 2026 Jan-April = "Bottleneck" period; thesis is that the next AI bottleneck is large-scale efficient inference. NBIS is positioning as the neocloud that solves inference economics. Part of the Nvidia playbook ($IREN → $NBIS lead-lag). The differentiation from generic neoclouds is the pitch.
ADBE
Q2 beat but sell-off continues. CFO离职 plus AI替代叙事. The earnings call this week is the software sector情绪锚 — transition plan and AI agent impact on Creative Cloud revenue will set the tone. "When LO连beat都不buy时sector re-rating还没走完" — even a beat doesn't matter when narrative has shifted. The bar is the guide, not the print.
SNOW
+48% on the week. Snowflake Summit is a week-ahead catalyst. AI inference data infrastructure thesis is intact. Dot-com revival basket exposure with DELL and HPE. The trade is the AI capex elongation through the data layer. Tape is strong but extended; let it set up.
HPE
+86% since previous post. AI server trade beneficiary. Part of the broader AI capex elongation with DELL. Less clean expression than DELL post-Vera Rubin shipment, but the basket trade is real.
VRT
$13.75B revenue guide, 34% growth. Notable increase in Field Services employees as share of total — read-through is the AI power infrastructure beneficiary as data center power moves from tens of kW to MW per rack. 800V HVDC提前商转 is a tailwind. This is the picks-and-shovels of the constraint cascade rotation to power.
PLTR / LMT / RTX / NOC
Defense tech cohort. Iran deal war premium being given back — "from hot war to high cold war" means defense multiples may compress as risk premium normalizes. But structural defense capex tailwind remains. The trade is risk-off, not thesis-off.
MSTR
Inverted H&S bottom likely in progress per technical read. BTC proxy, levered to crypto. BTC rebounding on Iran deal risk-on sentiment. Treasury balance sheet remains the structural risk. The trade is BTC beta with leverage and execution risk on the convert/debt stack.
QCOM
Snapdragon C for laptops at $300+ with Acer, HP, Lenovo signed up. Positioned as Neo competitor. Cristiano Amon at Computex; has called 2026 "the year of agents." The Windows on ARM thesis is being tested — QCOM is the incumbent, NVDA is the challenger with N1X. Counter: "almost everything on Windows is compatible with x86, good luck convincing everyone to make their software compatible with ARM." The software ecosystem moat is the question.
PANW
Reporting in its first quarter with CyberArk. Consolidation play in cybersecurity. CRWD also reporting. Cybersecurity structurally growing with AI threat surface expansion; consolidation theme continues. Watch for synergy realization and integration execution.
CRWD
Reporting same week as PANW. Cybersecurity consolidation theme. AI threat surface expansion is the tailwind. The bar is reacceleration in net new ARR, not just retention.
BABA / JD / BIDU / NTES
China Q2 consumer spending may go negative for first time post-COVID. Anthropic禁令 opens global expansion window for Chinese AI labs but consumer weakness limits domestic capex. Caught between AI capex need and consumer weakness. The trade is the consumer demand destruction call plus the geopolitical discount.
SNAP
UK Starmer announcing under-16 social media restrictions Monday. Revenue headwind but expected. "叠在Anthropic禁令 + Musk/Trump的AI friction上, regulatory risk premium在AI names上重新累积." The regulatory risk premium in social/AI is back.
IONQ / RGTI / XNDU / GFS / TSEM / QBTS
Quantum compute optical interconnect stack exposure. The bottleneck for quantum systems is moving from device to optical interconnect. $2B CHIPS Act commitment to quantum foundries is the structural tailwind. D-Wave Investor Day and MS Travel and Leisure Conference for QBTS. Speculative exposure; size accordingly.
IREN
Part of the Nvidia playbook ($IREN → $NBIS lead-lag trade). Neocloud positioning with GPU allocation. The lead-lag rotation is the trade.
AXTI
InP substrate lead-lag indicator for AAOI per Chinese trading methodology. InP supply chain upstream. The substrate constraint is the real bottleneck before the laser capacity ramp.
TMHC
Berkshire acquisition at $8.5B at $72.50/share all-cash. First major deal under Greg Abel. Take-out deal, not a trading opportunity. Premium ~25% to unaffected.
Street Color / Heard (unverified)
Hearing the Anthropic禁令 is being interpreted by some China desks as a green light for domestic frontier model capex — specifically that ByteDance and Alibaba are accelerating H100/H200-equivalent procurement via third-party channels. Word is Ant Group has been pulling forward inference capex to lock in allocation before year-end.
Channel checks suggest SoftBank's 75B euro / 5GW France AI DC is the first of three European announcements expected this quarter — next up is a UK consortium with Microsoft and an unnamed sovereign wealth fund. Behind-the-meter power is the differentiator; Schneider's role is to vertically integrate the HVDC stack.
Word is the Dell +43% print was triggered by a single large quant fund covering a short position, not broad-based institutional buying. If true, the follow-through depends on whether fundamental buyers step in or the tape gives back. Tape action post-print will tell.
Hearing from a sell-side trader that MU 6/24 has 4x normal options open interest, with the $1,200-$1,400 strikes being the most active. Implied move is pricing a binary outcome. The "cleanest read" framing is correct but the position is crowded.
Channel checks suggest Intel 18A external customer pipeline is thinner than Lip-Bu Tan's commentary implies — hearing that even Apple/Google/NVDA/AWS 18A allocation is conditional on Panther Lake yield progression, not committed volumes. The "internal-first" framing is accurate for 2026.
Word is the Vera Rubin NVL72 rack is operating at ~40% of spec'd performance in early CRWV deployment — power delivery and cooling are the bottlenecks, not compute. If true, the 800V HVDC pull-through is even more aggressive than the constraint cascade narrative suggests.
Hearing that Anthropic compute migration to GCP involves a $20-30B multi-year commitment, with TPU v5e and v6 allocation prioritized over NVIDIA hardware. If accurate, GOOGL's capex guide has upside not in consensus.
Channel checks suggest ORCL's $20B raise is being structured as a mix of convertible bonds and term loans, with the convert strike set at a 25-30% premium. Market is digesting the dilution risk, which is why the stock has lagged the AI capex cohort.
Word is the Microsoft Copilot engagement metrics internally are tracking below internal targets — hearing that the "lowest presence" Chinese commentary is directionally correct. Q2 Copilot revenue disclosure will be the catalyst.
Hearing that QCOM's Snapdragon C has thermal issues at the $300 price point — OEMs are asking for a price cut to $250 to hit ASP targets. If true, the "year of agents" framing may be aspirational.
Channel checks suggest China's consumer weakness is broad-based, not just iPhone — hearing that luxury goods, autos, and travel are all down double-digits in Q2 vs Q1. The BABA/JD/BIDU trade is the right frame but the magnitude may be larger than consensus.
Word is a major META institutional holder has been trimming position size post the recent rally, citing concentration risk in the cap-ex elongation trade. Tape action above $600 will be the tell.
Hearing that SNOW's +48% move was partially driven by index rebalancing flow, not just fundamental buying. The Summit catalyst next week is the test of whether the move has legs.
Channel checks suggest AAOI and LITE parabolic moves are showing early signs of smart money distribution — hearing that several momentum funds have been trimming into strength. 40-week moving average retest is the trade for new positions.
Word is the Anthropic禁令 may expand to cover additional Chinese-developed frontier models within 60-90 days. If true, the GOOGL tailwind extends beyond the immediate Anthropic migration.
Hearing that VRT's 800V HVDC提前商转 is driven by hyperscaler demand pull-forward — Schneider and VRT are both capacity-constrained into 2027. The power infrastructure trade has 18+ months of visibility.