Good morning. Tape's ugly underneath.
Friday's GS TMT Mo' pair dumped 875BPS — one of the biggest single-day pullbacks on record — and the bleed has continued: NVDA down 8 of 10 sessions, GOOGL three straight weeks red, LITE/AAOI "parabolic with rollover" per channel. But SOFTWARE was +20% IN MAY, best month since 2002. Not a regime change — a rotation. And positioning's stretched, "running out of buyers" per HF exposure reads. Buyer's strike setup, not a breakdown.
Futures leaning soft, no fresh prints to flip the script (earnings calendar thin today). Asia: SoftBank $87B / 5GW France commitment overnight — biggest AI infra deal in Europe — with Schneider Electric building an industrial power cluster in Dunkirk. That's the European behind-the-meter template at scale. Brent +1.8% ON US-IRAN TENSIONS — bad for European DC economics, reinforces the bypass thesis.
Three threads for the day: (1) Memory cycle elongation — Goldman raised SK Hynix 2027 op profit +22% TO $266B and Samsung +21% TO $352B. Peak just got pushed into 2027-2028, not normalized. HBM still the binding constraint. Trade's still on. (2) Bottleneck cascade rotating — memory → grid+capacity. META may have the most incremental GW online in 2026 of any lab, ahead of OpenAI/Anthropic. Intel 18A internal demand now squeezing external supply to Apple/Google/NVDA/AWS. Zero-sum foundry shift. (3) Behind-the-meter as European template — Schneider's Dunkirk cluster is proof of concept. If it scales, the European DC cost curve bends. Big if.
Inference is now the named next bottleneck: $66B → $292B BY 2029 at 45% CAGR. Mix shift from training to inference favors networking and memory density over raw compute. Capex thesis is rotating, not breaking.
We'll hit up TTAN first (highest conviction, ARW 4), then IOT — then get to the industrial/automation cluster with HON and AUR.
TTAN — No fresh signal in the feed. Clean trades-vertical SaaS name, lighter coverage from us for a reason. The longer-dated call is AI features embedded in the dispatch/workflow stack, but that's a 2027+ monetization story, not a 2026 catalyst. Pairs thematically with IOT on industrial data but no edge today.
IOT — No direct news, but the thematic setup is improving. Physical-AI narrative just got a capital reorg tell (Sora→robotics at OpenAI), and Schneider's Dunkirk "industrial production cluster" framing positions behind-the-meter monitoring as a critical layer. Samsara is the cleanest public vehicle for fleet/asset telemetry at scale. Catalyst watch: any industrial data partner win, particularly in DC operations or energy.
ADBE — No fresh news, but the enterprise inference pricing signal from the feed ("10x speed at 20-50x premium per token" per channel check) is a green shoot for any software company running proprietary inference workloads. Firefly + GenStudio is the monetization test. If enterprise WTP holds, Adobe is a beneficiary on the AI attach curve. Catalyst: next print, any commentary on AI ARR or attach rates.
AAPL — Multiple reads, and the net is mixed. (1) MacBook Neo REVISED UP 5M→10M shipments — real number move, and a vote for Apple's ability to absorb the memory cost squeeze via SKU mix shift. Most OEMs can't do this. (2) NVDA's N1X Arm-based chip for Windows PCs is a DIRECT CHALLENGE to Apple Silicon in AI PC — if MSFT/ARM/NVDA ecosystem gets validated at volume, Apple loses the "best AI PC" narrative. (3) Intel 18A external supply flows to Apple first (per Lip-Bu) — foundry relationship continuation is bullish. Net: hardware ASP resilient, AI PC defensibility under siege. Sub-$600 "extinction event" plays to Apple's premium positioning.
HON — The Schneider Electric Dunkirk call is the cleanest read for HON today. Schneider positioning as the integrated behind-the-meter / industrial power player for a 5GW European DC program is EXACTLY the lane Honeywell wants to own domestically. Process solutions, gas turbines, battery storage are the natural stack for US behind-the-meter DC projects. No fresh ticker news but thematic setup is strong. Catalyst: any US utility-scale behind-the-meter announcement in 2H26.
AUR — No direct signal. Autonomous trucking is the long-duration physical-AI play and Sora→robotics reorg is a sentiment tailwind (capital flowing to embodied AI). Aurora is pre-revenue scale so the thesis is binary on commercialization timeline. Watch for OIDA/regulatory milestones and PACCAR/Volvo/Uber Freight partner news. No edge on the feed today.
RBRK — No fresh news, but the thematic setup is real. Inference market scaling $66B → $292B BY 2029 AT 45% CAGR means data gravity compounds. Rubrik sits at the intersection of data management and AI infrastructure — backup/recovery for inference pipelines is a non-obvious beneficiary as enterprise AI workloads scale. Catalyst: any AI workload customer win or commentary on data protection demand at next print.
1. Hearing optical/photonics has gone "parabolic" with $AAOI and $LITE specifically showing "signs of rolling over." Watch for late summer/early autumn pullbacks to the 40-week moving averages. Technical signal, not fundamental. Question: does the inference bottleneck thesis keep the optical premium intact through 2H, or do we get a multiple compression with the tape?
2. Channel check: ABF/IC substrate prices +20%, contract DRAM/SSD +200-300% since mid-2025. PC OEMs downgrading memory/storage specs, prioritizing higher-end SKUs to pass costs. This is the squeeze template playing out in real time. MacBook Neo upward revision is Apple-specific ability to mix-shift, not a market-wide signal.
3. $32K MI300X listings on eBay and $1.99/GPU-hr rental showing up. Signal that AMD supply is loosening at the retail/rental level, OR a single vendor moving inventory. Not enough to call a trend but worth watching as a demand-softness early indicator at the bottom of the market. Distribution-only read, not a chain sizing call.
4. Word is META may have the MOST INCREMENTAL GW COMING ONLINE IN 2026 of any AI lab — potentially ahead of the 6GW OpenAI+Anthropic target floated on Dwarkesh. MSL team beginning to ship coincides with the capacity ramp. Dark horse frontier challenger narrative if it lands. [scale-mover if true, but conviction is low until we see model output]
5. Hearing Intel 18A internal demand is "rising quite quickly" per Lip-Bu — don't expect external supply except Apple/Google/Nvidia/AWS. Zero-sum foundry allocation shift in Intel's favor. Panther Lake Q-ship at 7X volume is the demand confirmation. Apple is the winner on the external allocation side; everyone else gets squeezed.
6. Enterprise inference pricing: "10x speed at 20-50x price premium per token" per one channel check. Early enterprise WTP test for ultra-low latency. This is the pricing signal that determines whether inference capex holds or pulls back — bookmark it. If it holds, the entire AI infra complex stays bid.
7. MLCC hoarding in distribution despite ODM demand falling — supply resources reallocating to AI applications. Sub-$600 PC segment "extinction event" expected. AI is the marginal buyer of every component category. Watch for any component maker guiding to AI-mix shift as a defense narrative.
8. Glass core substrates: per Semi reporting, initial production could begin ~2028 in selected high-performance applications. Larger package sizes, finer interconnects, improved dimensional stability vs ABF. Multi-year packaging capacity enabler. Structural 3-5yr duration, but scale-mover if it unlocks the 2028+ packaging ceiling that's currently the binding constraint.
9. OpenAI robotics: "Sora team became robotics team" per tweet, plus active hiring push. Physical-world AI commitment with no revenue timeline. Narrative for now, but capital reorgs are leading indicators — same template as MSL at META. Watch for any first product reveal or commercial pilot.