Good morning.

Tape soft into the US open. NVDA down 8 of 10, LITE down 5 straight, GOOGL bleeding 3 weeks, GS Most Short +25% in 8 weeks, TMT momentum pair -875bps Friday. Positioning tapped (GS: hedge fund IT "out of buyers"); thesis hasn't cracked. Earnings slate: CRM and ORCL are the 📌 prints — enterprise software tells us if AI ROI converts to license rev or stays in PoCs.

Asia overnight, two big prints. GOLDMAN RAISED SK HYNIX 2028 OP PROFIT 24% TO $301B, SAMSUNG 23% TO $405B — structural floor under the memory super-cycle through 2028. SOFTBANK COMMITTED €75B (~$87B) FOR 5 GW OF FRENCH AI DCs, Schneider as industrial partner. KOSPI bid, Asia green. Overhang: US-Iran war keeping European power costs elevated — every French DC assumption fragile if oil stays bid.

Four themes framing the day:

1. Goldman just floored memory. $301B SK Hynix 2028 op profit makes the HBM thesis un-investable to short. MU cleanest US read; WDC, STX, SNDK ride. Bullish the chain through 2028, full stop.

2. Positioning stress, not demand crack. NVDA 8-of-10, LITE 5-session slide, Most Short +25% in 8 weeks — flows, not thesis. Optical parabolic (AAOI, LITE, COHR) flagging 40-wk MA pullbacks. PM Q: healthy rotation, or 2-3 month consolidation before the next leg?

3. SoftBank/Schneider France is the new DC playbook. €75B / 5 GW bypasses US grid queues; Schneider is the named industrial partner. VRT, GFS, power/cooling beneficiaries. SoftBank is the cycle's dominant financial architect — $30B+ OpenAI stake, plus Arm — and spending like it.

4. Intel 18A flips; TSM concentration crystallizes. 18A framed as supply-constrained to Apple/Google/Nvidia/AWS (not supply-starved) — first time. Panther Lake 7x volume on 50% yields, Intel eating most of its own output. Bullish INTC. JENSEN CONFIRMED NVDA NOW >20% OF TSMC REVENUE at GTC Taipei — Apple-to-Nvidia transition is Jensen-confirmed, not inferred. TSM structural winner; customer concentration risk the new overhang.

Inference is the structural rotation: $66B (2025) → $292B (2029), 45% CAGR. Vista Equity's 40x cost reduction on insurance claims ($8M → $200K with embedded context) = value accrues at the proprietary context layer, not the model. Bullish CRM, NOW, ADBE, DDOG.

We'll hit MU, NVDA, and INTC first, then get to optical (LITE, COHR, AAOI) and software (CRM, ORCL, NOW).

---

CORE ANALYSIS

ADBE

VERDICT: 11% CC GROWTH BEAT, STOCK AT 52-WEEK LOW. SOMETHING'S BROKEN. ADBE printed a clean top-line beat ($6.62B vs $6.45B) and AI-first ARR tripled YoY to $500M+. Yet the stock is at $218.80, within cents of its 52-week low, down ~13% in a week. Read: the headline beat is masking a real deterioration underneath. Net new ARR EX-SEMRUSH was DOWN 3% YoY. That's the number. Street is repricing the model lower AND the leadership is walking out the door at the same time — CFO Durn to MRVL on June 15, CEO search still open. PMs want to be patient here but this is a "show me" story until both seats are filled and the freemium pivot is proven not to be a multi-quarter ARR hole.

STREET VIEW

PTs CUT FROM A $300+ CLUSTER TO $195-285. Stifel did the most damage ($350 → $200, downgrade to Hold) — that's a -43% PT move. KeyBanc (Underweight) sits at the floor at $195. RBC stays most constructive at $285 (Outperform), DA Davidson $250 (Buy). Piper $240, Mizuho $245 — both Neutral. Consensus has clearly migrated from a $300 handle to a ~$240-260 range, reflecting peer multiple compression plus Adobe-specific growth angst. Two outright downgrades (Wolfe, Stifel), rest are PT trims. DA Davidson's 10x FY26 EPS framework is the new anchor for the "fair" case.

BULL VS BEAR

BULL CASE — Trading at 12.9x P/E for an 89% gross margin software franchise is generational value. AI-first ARR went from ~$165M to $500M+ in a year. Firefly ARR +50% QoQ. Acrobat/Express MAU at 850M (vs 700M prior year) — freemium is feeding the funnel, not killing it. FY27 ARR guide maintained despite $480M Semrush contribution, so the underlying trajectory isn't collapsing. DA Davidson framing it best: leadership turnover "creates potential for meaningful strategic changes" — new CEO can be the AI reset the stock needs.

BEAR CASE — Net new ARR ex-Semrush DOWN 3% YoY. That's the single most important data point in the print. Organic ARR guide was cut by ~$480M and individual subscription ARR expectations were lowered. Adobe is deliberately giving up near-term monetization for freemium funnel building — and you're paying 12.9x for a company in active digestion mode. Meanwhile, BOTH the CEO and CFO seats are open simultaneously. KeyBanc's Ader nailed it: "hard to step in" when you don't know who's running the place and the organic engine is decelerating. Stifel downgraded — $200 PT implies -9% downside from here.

"Between the good but not stellar organic performance and guidance and the news that CFO Dan Durn is departing, it is hard to step in. Now a CEO and CFO search is underway, and we expect shares to continue to lag the sector." — Jackson Ader, KeyBanc

WHAT'S NEW

• CFO Durn → MRVL (June 15). Rare for a $500M+ EV software CFO to leave for a semi. Street reading this as a yellow flag on the Adobe trajectory.
• Net new ARR ex-Semrush: -3% YoY. This is the buried lede. Stifel/Wolfe fixated on this.
• Freemium strategic shift confirmed. Deliberate trade of near-term ARR for user base. New disclosure language.
• $500M AI-first ARR disclosure (3x YoY). First time the company gave a hard number — helps the AI credibility narrative.
• Individual subscription ARR guide reduced. Creative Cloud line optimizations deferred. Both signal monetization being pushed right.

READ-THROUGH

• TMT software peer tape: MSFT, CRM, NOW all dealing with the same "is AI a tailwind or a disruption to existing ARR base" question. ADBE is the cleanest cut case in the cohort right now — the +300% AI ARR growth exists alongside -3% organic net new ARR. The market is choosing to underwrite the latter.
• CFO → MRVL signal: Durn going to Marvell is a useful read on the relative attractiveness of the AI compute vs AI software names right now. Capital and talent flowing to the picks-and-shovels layer.
• Positioning: ADBE screens as a deep value name on absolute multiples (12.9x P/E, ~7x EV/FCF per Piper) but no one on the sell-side is willing to be aggressive until the CEO search resolves. Crowded underweight risk is low; crowded long risk is also low. PMs sitting on the sidelines is the dominant posture.

---

TTAN

THE CALL

KeyBanc out today keeping TTAN at Overweight, $120 PT after sitting down with CFO Sherry and IR's Rechel. Stock sits at $68.24 — so a 76% bogey on a name that's been BLED OUT 35% YTD. KeyBanc's calling this a top pick, more positive on Max after the meetings. The setup is: AI tailwind, beaten-down multiple, durable growth, everyone already hates the stock.

THE THESIS

ServiceTitan just printed 25% Q1 revenue growth vs 19% guide — that's a 6-point beat, not a rounding error. LTM revenue hit $1.01B on 24% growth, so this is a real grower getting cheaper. The bull case is the Max product: management's data point is 2x subscription spend per customer once Max fully ramps. That's land-and-expand math PMs will underwrite. Pair that with internal AI efficiency (margin lever nobody's pricing in) and you've got the case for the mid-20s% growth in FY27/FY28 KeyBanc is modeling. Market's clearly not believing the durability.

"came away feeling incrementally more positive on the company's Max opportunity, which can drive meaningful ROI for customers and result in 2x greater subscription spend when ramped."

THE STREET

Tight Buy cluster post-Q1, PTs in the $103-125 range. TD Cowen at $125 (top of the range, flagged the strongest sub growth print in company history), Truist at $110, KeyBanc at $120, BMO at $103, Freedom Broker at $105. Six firms, zero sells, but ~$22 wide in PTs — which means real disagreement on Max ramp speed and sustainability of that Q1 beat. Consensus hasn't caught up to the bullish interpretation yet.

WATCH LIST

• Insider: CAO O'Connor sold 10K at $69.14. Small print, CAO-level not C-suite, but PMs tracking Form 4 flow will see it. Doesn't break the thesis.
• Catalyst path: Next print is the re-rating event. If FY27 guide comes in line with KeyBanc's mid-20s% framing, the multiple expands. If it guides to a deceleration, this stays dead money at $68.
• Sector read: If trade/small business software is going to work in 2H, TTAN is a clean expression — vertical SaaS with AI embedded, not bolt-on.

Not sure we can call the bottom here, but the r/r is starting to look interesting at this print.

---

MRVL

THE TAKE

B.Riley just ripped the PT to $345 from $240 — a 44% raise — and the Street is running in the same direction. 26 upward revisions, PEG at 0.15, and Jensen Huang handed them the ultimate cover line on stage at Computex. But the stock's up 230% YTD vs SOX +86% and indicated 1.5% lower after hours. The discovery trade is over. This is now a positioning story, and the easy money's been made.

"Marvell is the next $1 trillion company." > — Jensen Huang, joint appearance with CEO Matt Murphy, Computex

That quote is going to be in every AI deck for the next two quarters. B.Riley is leaning into it — citing the deepening NVDA partnership across Custom silicon, NVLink Fusion, and Celestial in-house optics. The trifecta narrative is real and Stifel ($321), Benchmark ($275) are aligned.

BULL VS BEAR

Bull case: Custom silicon + NVLink + optics = the picks-and-shovels second derivative to NVDA's rack-scale vision. Sub-1 PEG on a 230% YTD mover is genuinely rare. Huang on stage is a 6-9 month narrative tailwind.

Bear case: Cantor is the outlier at $220 Neutral — and that dispersion is the tell. When the high-low spread on a name running 230% is $125, the Street doesn't agree on what this is worth. Cantor staying sidelined after the AI infra framework pitch is the sober voice in the room. New CFO Dan Durn starts June 15 — fresh eyes on the model means the first earnings call sets the tone.

WHAT MATTERS NEXT

• S&P 500 inclusion June 22 — passive bid into the date is real, watch for the unwind after
• 605K options contracts with heavy call skew — that's positioning, not conviction
• Stock -1.5% AH despite the B.Riley print tells you the marginal buyer is exhausted at these levels

Own it, don't chase it. The Huang call-up is a multi-quarter narrative, not a multi-day trade.

---

Supplementary Coverage

NVDA

Verdict: The $20B debt print is the cycle's most important demand signal of 2026. NVDA — the cleanest AI capex balance sheet on the street — tapping 7 tranches means the internal cash flow ceiling has been breached. (Down 8 of 10 sessions, TMT momentum pair -875bps Friday, marginal buyer story is broken.) N1X is the prove-it moment at GTC Taipei: if it's a real product, memory/storage content per device shifts and the client TAM shape changes; if it's "late, power-bleeding, expensive," this is the second narrative crack after the debt. TSMC >20% concentration (Jensen-confirmed) is the structural chokepoint — every N3/N2 wafer and every CoWoS slot is now a single-customer bet. 4Q26 mid-Feb is the single highest-impact demand durability print on the calendar. Bogey: HBM3E volume + Rubin ramp + capex commitments from hyperscalers/sovereigns.

MU

Verdict: $1,000 into 6/24 print is the consensus bull case; $1,600 PT cited. DRAM/SSD contract up 200-300% since mid-2025, magnitude now displacing consumer configs (less 32GB/16GB DRAM, 4TB/2TB SSD). MLCC distributor hoarding is the analog signal — AI demand is cannibalizing non-AI supply. SNDK GM 78.3% (vs 22.5% YoY) shows the leverage. The July print is the cycle's most important memory event: inventory days, FY27 capex, HBM4E qualification/customer mix. The Chinese source flagged it best: trade is real, but "easy money is behind, r/r is no longer asymmetric." Watch the call for forward ASP commentary — that's where the print lives or dies.

GOOGL

Verdict: Cleanest "safe harbor" in the AI infra complex. Anthropic compute transferring AWS → GCP is a structural data point. Apple using Google Cloud + NVDA GPU for AI reset is the second confirmation. GCP share 16% (2022) → 22% (2026) — the trajectory is intact. Down 3 weeks, Most Short basket +25% in 8 weeks, hedge fund IT "out of buyers" — pure flows damage, no fundamental break. Classic flows-vs-fundamentals split. If anything breaks elsewhere, GOOGL benefits. The 3p for PMs: this is the name that institutional desks will rotate into when the AI tape gets indigestion.

AAPL

Verdict: MacBook Neo 5M → 10M units — Apple gaining share at the low end as the sub-$600 PC segment faces "extinction event." Apple is the only PC OEM winning share in a contracting segment. Apple using Google Cloud + NVDA GPU for AI reset is the second story: admission that in-house AI isn't keeping pace — incremental demand for NVDA/GOOGL, pressure on AMZN. Smart glasses delayed to late 2027 (Meta first-mover for another 12+ months). iOS 27 ChatGPT-style app syncing is the feature chase, not the feature lead.

META

Verdict: Legitimate frontier challenger within 12 months — most incremental GW coming online in 2026 of any AI lab. Could be ahead of the ~6 GW OAI/Anthropic EoY '26 estimate (Anthropic closing the gap via xAI sublease). Lower incremental inference demand (no agent explosion yet) = training capacity advantage as new MSL team ramps. Holds smart glasses first-mover for 12+ months. META is the under-owned training-side AI play.

AMD

Verdict: MI300X at $32K eBay and $1.99/GPU/hr rental with 1.5TB VRAM self-service, no contract, accepts stablecoins. Pricing leverage is capped. AMD is the second-tier chip where supply looks different from H100/B200. Stargate/OpenAI custom silicon threat = structural overhang — if NVDA's custom silicon share at CSPs grows, AMD loses the second-source slot. $1T mkt cap is live in market conversation but r/r is brutal into MI300X cap. Packaging-as-Moore's-Law is the bull case: EMIB revenue could be materially higher than Street, $80B+/year packaging business. Mixed read.

MSFT

Verdict: Capex-harvester, exposed both ways on AI regulation. OpenAI's $150M partner network + statements on China-linked ChatGPT accounts = preemptive compliance posture ahead of BIS July 15 ruling. The "voluntary program" vs "government directive" line is being erased. Xbox studios (Compulsion, Double Fine) negotiating independence = gaming being harvested to fund AI capex. N1X with NVDA/Arm at Computex is the ARM-Windows TAM question for the next 24 months. The Windows 11 AI PC refresh is the consumer-facing capex payoff.

AMZN

Verdict: Cleanest loser from the export ban cleanup + Anthropic transfer. Anthropic leaving AWS for GCP is direct customer loss. Apple multi-cloud pivot erodes Bedrock differentiation. AMZN is the hyperscaler complex's clearest "this isn't working" name. The GOOGL/AMZN spread is the cleanest pair trade in the AI infra complex right now. Bedrock re-rating is the catalyst to watch.

AVGO

Verdict: Earnings this week, >$10B AI semis guide is the bar. Computex is the platform to formalize. Second-name AI accelerator play (Google TPU, Meta MTIA potential). InP optical revenue $550M → $4.5B (2025→2030) per cited model — long-duration play, AI capex visibility through 2027+ supports the 8x trajectory. The $10B number is what the stock is priced for. Any miss on the guide and the multiple compresses fast.

INTC

Verdict: 18A narrative flip is the lead. Lip-Bu Tam: "Internal 18A demand is rising quite quickly so don't expect external supply except for a few strategic product (Apple/Google/Nvidia/AWS)." First time 18A has been framed as supply-constrained to named external customers, not supply-starved. Panther Lake shipping at 7x volume on 50% yields. NARRATIVE FLIP: "no one designing for 18A" → "external demand exceeds supply." N1X is the direct threat to client compute. INTC entry-level CPUs up 15% per supply chain note = pricing pressure starting to show. Positioning watch: INTC is being repositioned as a packaging shop, not a leading-edge logic shop — that's the bull case (EMIB, Foveros) and the bear case (logic TAM leakage to TSMC).

TSM

Verdict: NVDA >20% of TSMC revenue confirmed by Jensen at GTC. Apple-to-NVDA leadership transition is now Jensen-confirmed, not inferred from 10-K language. CoPoS dual-line: Longtan test line one European/US equipment, one Taiwan (Shin Yu/Wan Run/Hong Sung). If Taiwan-side validation passes, 本土设备链新一波成长cycle opens. Glass core substrates 2028 production is the next long-tail bottleneck. ABF prices +20% = real packaging cost base inflation. TSMC is the single most important structural name for the next 24 months.

LITE

Verdict: Parabolic move, watch for late summer/early autumn 40-week MA pullback. Down 5 straight sessions per Goldman positioning. InP market $1.9B (2025) → $22.75B (2030) with ~50% undersupply persisting. NVDA pushing ~20x InP laser capacity, most vendors committing only ~12x. LITE 2030 revenue forecast $600M → $9B. The destination is real; the path matters as much as where it ends. LITE is the cleanest public proxy for 1.6T cycle.

AAOI

Verdict: Fastest grower, most levered to parabolic risk. 2030 revenue $60M → $2.1B per InP model. UHP CW CPO share ~10%. AXTI-up-AAOI-follows rotation play. Smaller base = more levered to mean reversion. Watch the 40-week MA test into autumn — the move has been parabolic.

COHR

Verdict: Preferred name: 6-inch wafer + laser capacity gives the cost structure to win InP undersupply. 2030 revenue $125M → $4.3B per cited model. Quantum interconnect optical stack exposure noted. The "COHR + 6-inch" combo is the structural advantage vs single-axis competitors. Watch parabolic mean reversion alongside LITE/AAOI into autumn.

DELL

Verdict: +32-33% Friday, largest single-day gain in history. World's first Vera Rubin NVL72 rack delivered to CRWV — 72 Rubin GPUs, 36 Vera CPUs, 3.6 exaFLOPS FP4, 75TB HBM, 260TB/s NVLink. Dell is the rack-level integrator bellwether for AI capex cycle. COO flags HDD as next shortage — Japanese suppliers (Resonac/TDK/Nidec/Minebea Mitsumi) as beneficiaries. WDC was S&P 500's biggest winner on the day = storage cycle second-order pricing index-confirmed.

WDC

Verdict: Storage cycle second-order, not NAND — AI data lake physical storage story. S&P 500's biggest winner on the day. HAMR/MAMR capex discipline + international vendors exiting 2D NAND = structural HDD undersupply. International exit (Samsung/SK/Kioxia) creates transfer opportunity to Taiwan names (MXIM/2337/2344). "Market hasn't fully priced the subsidiary angle." This is the cleanest rotation play off AI capex demand without the multiple compression risk of memory spot.

SNOW

Verdict: +48% cited, Summit this week in SF. Part of the "AI capex monetization" week (Snowflake Summit + MSFT Build + Cisco Live). Cleanest data platform play for agentic AI workflow story. The 48% rally is the rotation into software that works WITH the capex cycle, not against it. Watch for monetization proof points from the keynote.

CRM

Verdict: First publicly priced 'agentic AI vertical consolidation' deal — Fin (Intercom) for $3.6B, ~18x markup on $200M raised. $3.6B/$200M = 18x is infra-level, not traditional SaaS 5-8x. CRM is the consolidation buyer, not the target. Data moat is the structural defense vs AI-native rivals. Expect 2-3 more deals of this size from CRM/MSFT/GOOGL. Intercom/Zendesk/Five9 face re-rating pressure.

ORCL

Verdict: $70B capex was the first signal that AI capex is straining the most cash-rich players. NVDA $20B debt is the confirm. ORCL still works as a "debt-financed AI capex beneficiary" — but the multiple has to absorb the leverage cost narrative.

HPE

Verdict: Reports this week, +86% since referenced post. Second-name AI server play behind DELL, Juniper gives networking exposure. The print is the second-leg test of the AI server trade — if HPE guides in line, the basket extends; if it disappoints, the rotation is over.

CRWD / PANW

Verdict: Cyber double test this week. CRWD alongside PANW's first CyberArk quarter. The question: is cybersecurity spend sustaining into the agentic AI era, or are agentic AI workloads creating new attack surface faster than security spend can track? Integration execution is the PANW catalyst. CRWD is the broader spend test.

CRWV

Verdict: First Vera Rubin NVL72 rack recipient from DELL. Bare-metal MI300X at $1.99/GPU/hr self-service. Per JPMorgan tech sentiment matrix specifically flagged. SpaceX IPO at $1.8T (down from $2T) is the structural event — if Day 3-5 holds $185+, late-stage private $2T valuation re-rates the entire private AI infra universe. CRWV is the cleanest neocloud public proxy for that re-rating.

RDDT

Verdict: Long thesis surfaced; AI training-corpus value is the real moat. Disintermediation risk to traffic is partially hedged by AI labs paying for the same data on the backend. Founder voting control = persistent governance discount. Net effect on labs: neutral to slightly positive (cheaper training data, lower traffic-based feedback loops).

SNDK

Verdict: Q1 2026 (non-fiscal): revenue $5.95B vs $1.695B YoY (+251%); GM 78.3% vs 22.5% YoY. Growth came from ASP, not volume. "Sell more, earn less" is the structural bear case for NAND as a category. When supply catches up, revenue falls even as units grow. This is the cycle risk that bulls are not pricing.

STX

Verdict: HDD supply bottleneck flagged inside Dell commentary. Same supply-side rationalization that drives SNDK/NAND higher applies to STX/WDC. HAMR/MAMR capex discipline + international vendor exits = structural undersupply vs AI-driven cold storage demand. The HDD/AI cold storage angle is under-modeled.

QCOM

Verdict: "Year of agents" Computex opener, Snapdragon C targeting $300+ price points with Acer/HP/Lenovo as design partners. Established Arm Windows laptop stack vs N1X. QCOM has the OEM relationships, N1X has the NVDA/AI halo. QCOM is the incumbent, N1X is the threat. Watch the OEM design wins for share signal.

IONQ / RGTI / GFS / TSEM

Verdict: Listed exposure to the four-layer optical interconnect stack behind quantum scaling. IONQ + RGTI pure quantum plays, GFS/TSEM also general foundry exposure. $2B CHIPS Act quantum foundry commitment noted. Qubit-count-to-interconnect transition is real but chain read-through limited to photonics vendor concentration. Low-conviction names for now — sized for spec positioning only.

NOW

Verdict: PTs $150-236 (Bernstein high), Now Assist AI revenue target raised $1B → $1.5B for 2026. Cleanest platform play for enterprise AI workflow automation. The AI revenue ramp is the proof point for the agentic AI thesis — if NOW can show $1.5B+ in AI revenue at scale, the whole category re-rates.

VRT

Verdict: $13.75B revenue guide, +34% growth; heavy Field Services hiring. Cleanest secular growth profile in the AI DC power/cooling chain. VRT is the infrastructure trade behind the capex trade. The hiring signal is the proof: can't deliver $13.75B without boots on the ground.

LSCC

Verdict: CEO calling 2027 hyperscaler capex "over $1 trillion" — most bullish call on record. Small-cap play on AI capex visibility through 2027. If even half the $1T number is right, LSCC is multiple-expansion candidate. The CEO is putting his reputation on the most aggressive capex call of the cycle.

ADSK

Verdict: AEC revenue +17% CAGR since 2020; RPO +15% CAGR since 2015. Construction/AEC software play with structural growth driven by data center buildout (DCs are AEC projects). ADSK is the underrated AI capex play — data centers require architecture, engineering, and construction software too.

TMHC / BRK.B

Verdict: Berkshire acquires Taylor Morrison for $8.5B cash, $72.50/share — Greg Abel-led first major M&A in years. Direct signal of where smart capital sees value: real economy housing exposure. Cash position getting deployed. The deployment pivot is the story — BRK.B is signaling that rates are workable, housing is the value play, and Berkshire is finally putting cash to work.

NKE / MSTR / HIMS / IREN / NBIS / ONDS / OPEN

Verdict: Technical/spec positioning only. NKE volume shelf, MSTR inverted H&S bottom, HIMS $26 target with 30% short, IREN $63 with NBIS rotation logic, ONDS/OPEN watchlist. IREN/NBIS rotation framework (Chinese methodology) is the most actionable — IREN-up-NBIS-follows is the trade to size.

VIAV

Verdict: Preferred name in InP supply tightness trade. Same 1.6T cycle exposure as LITE/AAOI/COHR. Watch parabolic mean reversion.

CSCO / TXN

Verdict: Part of the six-hyperscaler-plus-name basket averaging +158% YTD, adding $1.7T mkt cap. Cisco Live in Vegas this week — CSCO is the AI networking incumbent with the highest-quality franchise in the basket. TXN analog/embedded exposure. The basket trade has been the year's biggest beta to AI capex — the question is whether it extends or rotates.

STM

Verdict: No specifics, article coming soon. European specialty/auto semis name. Low priority for the AI tape but worth tracking for non-AI cycle exposure.

ROKU / FOX

Verdict: Fox buying Roku per CNBC — "elevates them into Disney and Netflix tier." Streaming consolidation continues. Not core tech tape driver per the rss analysis but a real read on media M&A appetite at current rates.

CAR / CSGP / BLBD / ARXS / LOAR / CEVA / DDOG / ONTO

Verdict: Niche moat signals, no read-through to the AI tape. CAR (Avis workshop moat), CSGP (Florance/CoStar culture), BLBD (school bus lock-in), ARXS (Qnnect growth), LOAR (latent pricing power), CEVA/DDOG/ONTO (no signal). Niche industrial roll-up M&A framework applies to LOAR/ARXS — small-cap, multi-X latent pricing power post-acquisition is the thesis.

---

Street Color / Heard (unverified)

Hearing that hedge fund IT is "running out of buyers" — gross/net exposure in TMT at multi-year highs while marginal demand has thinned. NVDA down 8 of 10, GOOGL down 3 weeks, LITE down 5, Most Short basket +25% in 8 weeks. The fundamental thesis has not cracked; the marginal buyer has. Whether this is healthy rotation or the first sign of a top is the binary for the week.

Channel checks suggest AMD MI300X bare-metal at $1.99/GPU/hr (1.5TB VRAM, self-service, no contract, accepts stablecoins) and $32K eBay listings — second-tier AI chip supply looks materially different from H100/B200 where rental queues remain tight. xAI struck a "favorable deal on a ST basis" with 90-day return optionality — a hedge structure that only makes sense when buyer expects to roll into tighter supply. Rental stability/increases are the operating leverage that justifies HBM and CoWoS capex.

Word is SoftBank/Schneider €75B (~$87B) commitment for 5 GW French AI DCs (3.1 GW phase 1 in Hauts-de-France by 2031) is the largest single European AI infra bet on record and the template for hyperscalers/neoclouds looking to add GW outside the 3-7 year US grid queue. ~600 MW/year net new EU capacity. The Schneider vertical is the second-order trade.

First public signal from Lip-Bu Tam: 18A framed as supply-constrained to named external customers (AAPL/GOOGL/NVDA/AWS), not supply-starved. First time this framing has surfaced — narrative flip from "no one designing for 18A" to "external demand exceeds supply." Panther Lake shipping 7x volume on 50% yields.

Hearing Meta has the most incremental GW coming online in 2026 of any AI lab, potentially ahead of the ~6 GW OAI/Anthropic EoY '26 estimate. Anthropic has since added the xAI sublease to close the 1 GW gap with OAI. Meta's lower incremental inference demand = training capacity advantage as new MSL team ramps.

Single-source tweet claims OpenAI's Sora team has been reassigned to robotics. Separate OpenAI Robotics recruiting post claims "rapid progress towards building AI that can help people in the physical world." First public signal, unconfirmed — but world-model expertise transfers to physical-world AI cleanly.

Hearing pre-training compute has not crossed the $1B/run threshold yet — labs have never used 300T+ tokens for pretraining or paid $4/hr for GB300. Softens the "training bubble" framing but doesn't address inference volume growth.

Channel checks suggest inference pricing is bifurcating: 10x speed at 20-50x token premium for latency-driven enterprise inference. The "AI harness wars of 2027" framing — browsers and user memory as the next platform battle — surfaced as forward thesis. Jevons paradox is the offset: per-task efficiency doesn't reduce aggregate compute if volume scales.

Word is glass core substrates being evaluated for high-end packages with initial production targeted ~2028 — "larger package sizes, finer interconnects, improved dimensional stability." Aligns with late-cycle TSMC N2/A16 packaging ramp. Long-tail structural bottleneck, 2-3 years out.

Hearing that Lattice Semi CEO made the most bullish hyperscaler capex call on record: >$1T for 2027 and "continues to grow." If even half the $1T materializes, LSCC is multiple-expansion candidate.

First public signal that AMD-Google partnership on HBM-equivalent memory architecture could be a structural advantage. AMD's $1T thesis is live in market conversation but the MI300X pricing cap is the bear counter.

TSMC CoPoS dual-line chatter (Chinese source methodology): Longtan test line one European/US equipment, one Taiwan-side (Shin Yu/Wan Run/Hong Sung). If Taiwan-side validation passes, 本土设备链 new growth cycle opens. Distribution-only at chain level but real signal.

Reddit (RDDT) long thesis surfaced in feed: training-corpus value "what makes Reddit's data-licensing position valuable to AI developers." Disintermediation risk to traffic is partially hedged by AI labs paying for same data on backend. Net effect on labs: neutral to slightly positive (cheaper training data, lower traffic-based feedback loops).

Hearing that AMD-Google HBM-equivalent memory architecture partnership is being re-evaluated by the Street. AMD-Google collaboration could give AMD a structural memory advantage vs SK Hynix/Micron — but MI300X pricing cap is the offset. $1T mkt cap thesis is now in conversation, r/r is the question.

Channel checks suggest HDD is the next shortage component (per DELL COO). Japanese suppliers (Resonac/TDK/Nidec/Minebea Mitsumi) flagged as beneficiaries. WDC was S&P 500's biggest winner on the day = storage cycle second-order pricing index-confirmed. The HDD/AI cold storage angle is under-modeled.

Hearing that Salesforce's Fin (Intercom) acquisition at $3.6B / 18x markup on $200M raised is the first publicly priced "agentic AI vertical consolidation" deal. 18x is infra-level, not traditional SaaS 5-8x. Expect 2-3 more deals of this size from CRM/MSFT/GOOGL.

First public signal that Stack overflow data-licensing is a "shovel" in the AI gold rush — same framework as Reddit training-corpus value. AI data licensing as a category is under-priced.

Word is that NVDA's custom silicon share at CSPs is growing, threatening AMD's second-source slot. Stargate/OpenAI custom silicon is the structural overhang — if NVDA's custom silicon share grows further, AMD loses the second-source slot at the most cash-rich buyer.

Hearing that SpaceX IPO at $1.8T (down from $2T) is the structural event for the entire private AI infra universe. If Day 3-5 of SpaceX holds $185+, late-stage private $2T valuation re-rates the entire neocloud complex (CRWV/CRWV-type names benefit).

Channel checks suggest Berkshire/Taylor Morrison $8.5B cash deal at $72.50/share is Greg Abel-led first major M&A for Berkshire. The capital deployment pivot is the real story — BRK.B is signaling rates are workable, housing is the value play, and the cash position is finally getting put to work.

First public signal that Apple is using Google Cloud + NVDA GPU for AI reset — admission in-house AI isn't keeping pace. Incremental demand for NVDA/GOOGL, pressure on AMZN positioning. Same framework as Anthropic transferring AWS → GCP.

Hearing that MSFT Xbox studios (Compulsion Games, Double Fine) in active negotiations to go independent — cost-cutting is the cash generation story for AI capex. Gaming being harvested to fund AI infrastructure is the capital allocation signal.

Word is that the Anthropic transfer from AWS to GCP is a "selective enforcement" read-through from the export ban aftermath. Trump officials meeting Anthropic per Politico; OpenAI launching $150M partner network + commenting on China-linked ChatGPT accounts as preemptive compliance posture. BIS July 15 scope could expand to OpenAI — that's the binary risk.