Tag: HBM

TL;DR

• ASML reported a record €11.7B in Q3 2026 net sales with 57.5% gross margin, raised guidance again, and still saw shares wobble because bookings and macro sentiment didn’t signal fresh upside beyond an “AI forever” setup already in valuations. [1][2][8]
• Management said customers want lithography slots locked through 2029–2030, yet 2027–2028 output will still trail demand—good for pricing power and upgrade revenue, but unit ceilings cap upside surprise. [2][3]
• The next constraint sits upstream: Zeiss optics and 12‑inch photomasks dictate when High‑NA EUV moves from press releases to high‑volume wafers in Veldhoven, Hsinchu, Hwaseong, and Chandler. [4][5]

What the source said

Barron’s noted ASML shares slipped even as the company raised guidance and highlighted multi‑year tool demand tied to AI workloads. Management said customers are trying to secure systems into 2029–2030 and reiterated that supply in 2027–2028 won’t fully meet demand. The piece framed a scramble for scarce lithography capacity and asked why the stock didn’t rally if the AI boom has legs. The answer: investors want proof that foundries serving hyperscalers will keep ordering at this pace and that ASML can convert backlog to shipments on time. [1][3]

Why it matters

• TSMC, Samsung, Intel Foundry, Micron, and SK hynix depend on 2027–2028 tool deliveries to hit roadmaps for N2/N3, HBM, and 1β/1γ DRAM; if ASML’s constrained supply persists, lead times extend, ramps slip, and capex tilts to productivity add‑ons. That dynamic flows through to Nvidia’s customers and rivals who buy capacity via those foundries. [2][3][4]

• ASML’s Installed Base Management (service + upgrades) was ~€3.289B in Q3, or ~28% of net sales, acting as a cycle buffer and margin support while unit ceilings limit volume beats. Zeiss SMT remains the critical path for EUV and High‑NA scale; if optics output lags, High‑NA volume timelines slip everywhere. [2][5]

Original analysis

Consensus says “Q3 proved AI capex is intact, ASML lifted the outlook, and shares faded on ‘priced‑in’ strength—buy the dip.” My read: the selloff is rational because the quarter codifies a physical production ceiling through 2028, so upside is paced by bottlenecks, not by demand euphoria. The stock multiple won’t expand if units can’t grow faster than the chain allows, even with 57.5% gross margin tailwinds. [1][2][3]

Back‑of‑the‑envelope checks (FY 2026 math)

• ASML guides 2026 sales to €44–45B; Q1–Q3 total ~€29.8B (8.8 + 9.3 + 11.7). To reach €44.5B midpoint, Q4 needs ~€14.7B—within the €14.2–15.2B guide, implying no heroic volume stretch; mix and margins decide upside. [2]
• Mix reality: Q3 Installed Base Management was ~€3.289B, ~28% of net sales (3.289 / 11.662). That high‑margin, repeatable stream helps EPS but signals throughput upgrades, not a step‑function jump in system units. [2]
• YoY lens: 2025 net sales were €32.7B; €44–45B implies ~35–38% growth. CFO Roger Dassen said 2026 is up ~36% versus 2025, matching the midpoint and shifting focus to execution. [7][6]

2×2 (2027–2028): Tool Access vs. Demand Certainty

• Preferred access + hyperscaler‑anchored demand: TSMC, Samsung memory/HBM, Intel Foundry on select nodes; first in line and willing to pay for productivity, using High‑NA timing to set price/perf. [2][4][6]
• Preferred access + speculative demand: Second‑tier logic ramps tied to smaller AI customers; risk over‑ordering into 2028 if inference ROI tightens.
• Constrained access + hyperscaler‑anchored demand: DRAM players outside the top two; likely to lean on DUV immersion plus upgrades while waiting on EUV slots. [2]
• Constrained access + speculative demand: Mainland China fabs under export rules; DUV for mature/ mainstream logic continues, but leading‑edge roadmaps remain throttled. [6]

Named‑stakeholder breakdown: TSMC, Samsung, Intel, Zeiss

• TSMC: Secures early 2028 slots and can time High‑NA toward ~2030 mass production per Bloomberg, aligning with customer economics instead of hype. [4]
• Samsung: Targets earlier High‑NA (~2028) for memory/logic and can arbitrage HBM throughput via upgrades while rivals queue. [4]
• Intel Foundry: Already running High‑NA on select Intel 18A layers tied to Core Ultra Series 3—a signaling edge to court leading‑pitch customers. [6]
• Zeiss SMT: Single‑threaded bottleneck on EUV/High‑NA optics; any slip at Zeiss delays everyone’s ramp. [5]

Contrarian read: multiples vs. unit ceilings in 2027–2028

• “Can’t meet demand” boosts pricing and margins but suppresses the prospect of a unit surprise, capping multiple expansion; with Samsung near ~2028 and TSMC near ~2030 for High‑NA volume and tool prices around $400M, most 2027–2028 wafers will ride Low‑NA EUV + DUV upgrades. That’s solid economics but not a narrative catalyst for hypergrowth. [4]
• Two overhangs persist: roughly ~20% China revenue mix stays policy‑sensitive if DUV controls tighten, and any incremental restriction would hit systems mix more than service. Those risks argue for patience versus a chase. [6]

What others are missing

Coverage focuses on orders through 2030, but the gating factor is the plumbing: Zeiss mirror/lens throughput and 12‑inch mask availability. High‑NA productivity hinges on multi‑year optics scale‑up at Zeiss SMT and on the September 2026 industry push—ASML, TSMC, Samsung, and Intel—to standardize 12‑inch photomasks that cut shot count and control line‑edge roughness. Until those two inputs are proven at volume, “orders into 2030” won’t translate to “wafers at cost” in Hsinchu or Hwaseong. That mismatch explains why 2027–2028 remains the hard ceiling, not a communications issue. [4][5]

What to watch next

1. By Q1 2027, ASML details a plan to lift 2028 Low‑NA EUV and DUV immersion output above the previously indicated ~+30% capacity step, citing added optics lines at Zeiss and laser subsystem suppliers. [2][5]

2. By Q4 2027, at least one non‑Intel production program (e.g., TSMC or Samsung) publicly confirms initial High‑NA usage on revenue wafers beyond test chips or demo lines. [4][6]

3. Through FY 2027, ASML reports gross margin ≥56% as Installed Base Management offsets system unit ceilings, sustaining EPS even if chip stocks wobble. [2]

My take

The quarter is excellent, but bottlenecks rule the tape: output caps in 2027–2028 plus long‑lead optics and mask constraints mean compounding, not explosions. That tilts the setup toward owners who value service and upgrade annuities over unit spikes. I expect a re‑rate once High‑NA proves cost‑effective at volume and Zeiss optics scale visibly. Until then, buy weakness, but don’t pay for unit ramps the factory can’t ship; keep an eye on Zeiss, 12‑inch masks, and the ~20% China mix. [2][4][5][6]

Sources

[1] ASML Earnings Point to Soaring AI Chip Demand. Why the Stock Isn’t Budging. — Barron’s (https://www.barrons.com/articles/asml-earnings-stock-price-d0973c5c) — Stock reaction vs. raised guidance and multi‑year demand.

[2] ASML reports €11.7 billion total net sales and €5.0 billion net income in Q3 2026 — ASML Investor Relations (https://www.investor.asml.com/news-releases/news-release-details/asml-reports-eu11.7-billion-total-net-sales-and-eu5.0-billion-net-income-q3-2026) — Official Q3 results, Q4 guide, FY 2026 outlook, and statement that 2027–2028 supply won’t fully meet demand.

[3] Chip tool giant ASML nudges up outlook as AI-driven demand outruns supply — Reuters via Euronext Live (https://live.euronext.com/en/financial-news/chip-tool-giant-asml-nudges-outlook-ai-driven-demand-outruns-supply) — Confirmation of raised outlook, orders out to 2029–2030, and capacity shortfall through 2028.

[4] ASML, TSMC, Samsung, Intel Back 12-Inch Masks for AI Chips — Bloomberg (https://bloomberg.com/news/articles/2026-09-08/asml-tsmc-samsung-intel-back-12-inch-masks-for-ai-chips) — High‑NA adoption timing (Samsung ~2028, TSMC ~2030) and tool cost; mask initiative details.

[5] ASML’s EUV Expansion Runs Through Zeiss — EE Times (https://www.eetimes.com/asml-euv-expansion-runs-through-zeiss/) — Optics as the bottleneck; dependency on Zeiss SMT for EUV/High‑NA scaling.

[6] ASML Q2 2026 Investor Call Transcript — ASML (PDF) (https://ourbrand.asml.com/asset/1fd3908a-0381-47b5-9b69-a3094f656651/2026_07_15-ASML-Transcript-investor-call-Q2-2026.pdf) — China ~20% of sales, 30% 2027 capacity step for Low‑NA EUV/DUV, Intel Foundry’s early High‑NA use on 18A.

[7] Q4 2025 and Full‑Year 2025 Financial Results — ASML Investor Relations (https://www.investor.asml.com/news-releases/news-release-details/q4-2025-financial-results) — 2025 baseline revenue (€32.7B) for 2026 growth math.

[8] ASML Stock Falls Following a Slight Sales Miss in Q3 — Yahoo Finance (https://finance.yahoo.com/markets/stocks/articles/asml-stock-falls-following-slight-112840272.html) — Same‑day share reaction and “priced‑in” narrative post‑earnings.

TL;DR

• Apple, Amazon, Reddit, Micron, SK Hynix, and Roblox were the stocks that explained today’s market tape: AWS-driven optimism battled iPhone-margin angst, while AI memory suppliers traded like utilities-in-the-making and ad/social names paid a “Google tax.” [1][3][4][5]
• The real tell: cash flow and contracts. Amazon’s AI push juiced AWS 37% YoY but flipped free cash flow negative; Micron locked in $22B of multi‑year take‑or‑pay deals, shifting memory from cyclical to contracted. [3][6]
• Consumer platforms are now distribution‑fragile. Reddit’s post‑print wobble on search‑referral fears and Roblox’s spring guidance cut show what happens when Google changes the hose. Expect valuation dispersion to widen. [9][10]

What the source said

Barron’s grouped the day’s action around a handful of bellwethers—Apple, Amazon, Reddit, Micron, SK Hynix, and Roblox—to explain why U.S. stocks ended a bruising July on a mixed but improving note. It highlighted a tech rebound that still had to swim against Apple’s drop, while AI‑infrastructure names steadied sentiment. The framing was less about a single earnings print and more about how these companies’ moves mapped onto the market’s push‑pull between AI capex optimism and consumer tech skepticism. In short: Amazon and chips lifted; Apple weighed; social/ad names chopped. [1]

Why it matters

• Stakeholders with the most to win or lose aren’t just the obvious shareholders. Hyperscalers (Amazon and peers), memory suppliers (Micron, SK Hynix), and their debt and equity financiers are tied together by multi‑year AI buildouts. A pause or re‑acceleration cascades instantly through contracts, capex plans, and margins. [3][6][7][8]
• On the other side sit ad‑dependent consumer platforms (Reddit, Roblox) and even Apple’s hardware complex, whose fortunes hinge on distribution and unit economics. One algorithm tweak or one quarter of component inflation can erase a year’s worth of product work—or a quarter’s beat. [4][9][10]

Original analysis

The bifurcated tape behind “stocks that explain today’s market”

The tape split cleanly into two economies: AI infrastructure sellers (Amazon’s AWS and chip supply chain) and consumer‑facing platforms (Apple devices, Reddit/Roblox engagement). Infrastructure got validation; consumer names got questions.

• AWS’s 37% YoY growth and Amazon’s disclosure that AI+chips are already a $25B annualized business tell you demand is real—but the bill is too. Amazon’s 12‑month free cash flow swung to a $7.6B outflow as it front‑loads capacity, which is why the stock can surge and still raise financing eyebrows. [3]
• Memory suppliers have quietly derisked. Micron says customers have committed $22B via five‑year take‑or‑pay deals—language this industry simply didn’t use in past cycles. That’s a structural pivot from “price taker” to “capacity allocator.” [6]
• Apple’s quarter was strong on the surface—$109.4B revenue, EPS $2.02, records across iPhone, Mac, and Services—but management also flagged rising memory costs into September. That pushes investors to ask: how much AI‑era component inflation can Services’ margin cushion offset? [4]
• Social/UGC businesses are distribution‑fragile. Reddit’s results beat, but sentiment turned on search‑referral risk; publishers across the web are considering restricting Google access as AI Overviews cannibalize clicks. That’s not a one‑day story; it’s a potential structural CAC hike for the whole open‑web ad economy. [10]

Back‑of‑envelope calculations (two quick cuts)

1. AWS AI+chips scale: If Amazon’s “AI+chips” sub‑business is running at a $25B annualized rate today and grows a conservative 35% in the next 12 months, that’s ~$33.8B by Q2 2027. At a 25% segment‑level operating margin, that implies >$8B of operating income from this sub‑line alone, enough to mask volatility elsewhere—but only if capex intensity tapers. [3]

Calculation: $25B × 1.35 ≈ $33.8B; $33.8B × 0.25 ≈ $8.45B.

2. Memory contracts as earnings floor: Micron’s $22B of five‑year, take‑or‑pay commitments average ~$4.4B/yr of revenue visibility. At a 45% memory-cycle gross margin (illustrative), you’ve got ~$2.0B gross profit “pre‑sold” annually—even before spot upside. This is why memory beta has started to compress. [6]

Calculation: $22B ÷ 5 ≈ $4.4B; $4.4B × 0.45 ≈ $2.0B.

A 2×2 that actually predicts behavior

Axes: Demand visibility (high ↔ low) vs. Distribution risk (low ↔ high).

• High visibility / Low distribution risk: Micron, SK Hynix (multi‑year supply deals; ADR broadened ownership). Playbook: raise capex methodically, pre‑sell capacity, manage price discipline. [6][7][8]
• High visibility / High distribution risk: Amazon (visibility on AI demand via backlog; but customers can optimize consumption). Playbook: sprint now, prove ROI later; expect cash flow volatility. [3]
• Low visibility / Low distribution risk: Apple (giant installed base; unit demand tied to macro and BOM inflation). Playbook: squeeze Services and ASPs to buffer input‑cost cycles; watch memory pricing. [4]
• Low visibility / High distribution risk: Reddit, Roblox (ad or bookings funnels exposed to search and policy shifts). Playbook: reduce Google reliance via native search and direct acquisition; reprice ads toward logged‑in, high‑intent inventory. [9][10]

Contrarian read

Consensus: “The AI infrastructure trade is back—own hyperscalers and HBM suppliers; avoid ad‑driven platforms.” My pushback: the market is underpricing the permanence of memory supply contracts and overpricing the linearity of hyperscaler free cash flow. Take‑or‑pay deals make Micron/SK Hynix earnings more utility‑like than the market credits; in contrast, AWS’s stellar growth is already showing up as cash burn. That inversion—contracted suppliers vs. capex‑heavy buyers—means the traditionally “riskier” memory names could deliver steadier EPS beats than the buyers funding them over the next 12 months, even if AI demand stays hot. [3][6][7][8]

What others are missing

The step‑function change isn’t “AI demand up.” It’s contract structure and distribution power. On supply, five‑year take‑or‑pay agreements shift memory from a spot‑exposed commodity to booked capacity—de‑risking earnings through 2027 for Micron and, indirectly, for SK Hynix now that it can tap U.S. capital at scale via ADRs. On demand, open‑web distribution is being rewritten by AI Overviews. Reddit’s wobble is the forward case for any platform leaning on Google for new users: acquisition costs rise, and growth quality tilts to logged‑in, high‑intent cohorts. Coverage that stops at “big movers” misses how these two mechanics—contracts upstream, distribution downstream—will drive multiple expansion for suppliers and a valuation headwind for ad‑dependent platforms. [6][7][8][10]

What to watch next

1. By Q4 2026, Micron lifts disclosed take‑or‑pay/backlog commitments to $30B+ across HBM/DRAM/NAND, confirming a structural derisking of memory earnings. [6]

2. By Q2 2027, Amazon states AI+chips (within AWS and adjacent) at a >$35B annualized revenue run‑rate, even as TTM free cash flow remains below its 2025 level due to continued data‑center buildout. [3]

3. By Q4 2026, Reddit reports flat to negative U.S. DAU YoY but >15% ex‑U.S. DAU growth and rising ad ARPU, reflecting a pivot toward logged‑in, high‑intent traffic as Google referrals dilute. [10]

My take

I’d rather underwrite contracted bandwidth than open‑web eyeballs in this tape. Memory just crossed a one‑way door: multi‑year commitments, ADR access for SK Hynix, and hyperscalers effectively pre‑paying for supply. That is not how the last memory super‑cycle ended. Meanwhile, every consumer platform that can be “answered” by AI will relearn the cost of rented distribution. I’m buying the derisked supplier earnings (Micron/SK Hynix on pullbacks) and treating AWS‑led rallies as proof of demand, not proof of cash discipline. For consumer internet, the burden of proof is now distribution resilience—not last quarter’s beat.

Sources

[1] Apple, Amazon, Reddit, Micron, SK Hynix, Roblox, and More Stocks That Explain Today’s Market — Barron’s (https://www.barrons.com/articles/stock-movers-e7d1ed18) — Sets the day’s framing: tech rebound vs. Apple drag;

TL;DR

• Alphabet earnings are the market’s Rorschach test: a higher 2026 capex guide could either reignite the Magnificent Seven trade or finally force investors to price the cash‑flow hit that AI spending is already inflicting on hyperscalers like Alphabet, Microsoft, and Amazon. [1][4]
• The semiconductor drawdown isn’t just “profit taking”; it’s a confidence wobble triggered by policy risk from the U.S. Commerce Department and competitive shocks like Moonshot’s Kimi K3, which sharpen questions about ROI on ever‑bigger AI data centers. [2][3][7]
• My edge: Alphabet’s capex mix and disclosure will be the tell—if the language tilts toward in‑house accelerators and infrastructure ownership in 2026–2027, merchant semis won’t get the reflexive relief rally people expect. [4][5]

What the source said

Yahoo Finance frames Alphabet’s results as a “moment of truth” for tech investors weighing a revived Magnificent Seven trade versus “buying the dip” in semiconductors, citing a widening one‑month performance gap between the SOX and the MAGS cohort. The article quotes Evercore ISI’s view that hyperscalers’ forward free cash flow could turn negative this quarter, and it argues Alphabet’s print and 2026 capex outlook could either lift both groups or deepen the split. The piece flags export rules, tariff headlines, and China‑based model releases (Kimi K3, DeepSeek) as catalysts that stoke fears U.S. firms are overspending on AI. The advice: watch Alphabet’s 2026–2027 capex signal and the stock’s reaction on the day of the report. [1]

Why it matters

Two sets of investors are exposed right now. First, anyone who chased the AI hardware trade through SOX constituents is learning how policy whiplash (export controls, tariff noise) and model‑cycle surprises can compress multiples in a week. Reuters pegs the SOX down more than 11% from its June record as of mid‑July 2026; South Korea’s KOSPI, a chip‑heavy barometer, slid more than 20% from its late‑June peak into a bear market before episodic bargain‑hunting bounces. [2][6]

Second, Alphabet, Microsoft, Amazon, and Meta shareholders own the other side of the AI buildout: the capex and the cash flow. FactSet highlights that hyperscaler spending increasingly outruns internal cash generation, with rating agencies already punishing balance sheets (S&P cut Oracle to BBB‑ in July 2026, citing surging AI capex and negative FCF). Alphabet’s guidance cadence on 2026–2027 will indicate whether the runway is smooth or a funding gap still needs bridging, which matters for credit spreads and equity risk premia. [4]

Original analysis

Back‑of‑envelope: what Alphabet’s capex implies

• Fact pattern: On its Q4 2025 call (Feb 4, 2026 transcript), Alphabet guided 2026 capex to $175–$185 billion, primarily for AI compute, technical infrastructure, and Cloud growth; management also flagged faster 2026 depreciation from prior ramps. [5]
• Simple math: Use the $180B midpoint and assume a 5‑year straight‑line life for a blended basket of data centers, servers, and networking. Annual depreciation ≈ $180B ÷ 5 = $36B.
• Cash consequence: To offset $36B of incremental non‑cash expense and its downstream cash costs (power, ops), Alphabet would need incremental operating cash flow of roughly the same order. At a 30% operating margin (Alphabet reported ~30% in Q4 2025), that implies ~$36B ÷ 0.30 ≈ $120B of additional annual revenue over a steady‑state base to hold FCF flat—directional, but scale‑setting. [5]
• Why this matters: Even if the real mix and useful lives differ, the size signals that one or more quarters of pressured or even negative free cash flow is plausible as assets ramp, a dynamic credit markets and rating agencies are already sensitized to across the hyperscaler set. [4][7]

2×2: “Capex Trajectory” × “Model Edge”

• Axes

  • X: Capex trajectory in 2026–2027 (raising vs. pausing) as implied by Alphabet’s guide and commentary. [5]
  • Y: Model edge in 2026 (frontier differentiation rising vs. compressing) amid Kimi K3 and DeepSeek pressure on capability‑per‑dollar. [3][1]

• Quadrants

  • Flywheel (Raising × Rising): Alphabet guides up and demonstrates widening AI product monetization in Search and Cloud, with explicit attach metrics or pricing anecdotes. This pulls MAGS higher; merchant‑semi relief is muted unless Alphabet signals more third‑party silicon procurement.
  • Cash‑burn trap (Raising × Compressing): Guidance up, but external shocks (e.g., Kimi K3 narrowing capability/cost gaps) keep monetization lagging; market punishes semis and hyperscalers, and spreads widen for issuers tapping debt to fund AI builds. [3][4]
  • Wait‑and‑see (Pausing × Rising): Capex restraint plus visible product velocity favors MAGS; SOX stabilizes but trails as investors rotate to software yield over hardware volume.
  • Value rotation (Pausing × Compressing): Both trades suffer; capital chases power, grid, cooling, and memory vendors with nearer‑term pricing power such as Vertiv, Schneider Electric, SK hynix, and Micron Technology.

My base case for the print: Alphabet’s language will keep capex elevated and emphasize infrastructure ownership across data centers, networking, and power integration. That stance is bullish for Alphabet’s moat and for select suppliers like HBM memory and advanced packaging, but it’s not a “lift all semis” catalyst—especially with BIS export‑policy noise resurfacing in Washington and investors newly alert to the cash cost of AI. [4][7]

What others are missing

Coverage obsesses over A100s vs. TPUs and the SOX chart, while the overlooked variable is competitive efficiency: open‑weight models like Moonshot’s Kimi K3 narrow capability gaps at lower API prices and push buyers toward “good‑enough” inference. If good‑enough AI improves faster than customers’ willingness to pay in 2026, ROI thresholds on hyperscaler‑owned capacity ratchet higher, and utilization needs to run hotter before expansions pencil. That dynamic hits merchant compute volumes first (customers sweat assets longer) and compresses the window for hyperscalers to translate capex into sticky revenue. Nature’s reporting that Kimi K3’s launch swamped capacity and narrowed the U.S.–China performance gap frames this as an industry‑wide pricing and payback story, not just a China headline. [3]

What to watch next

1. By October 31, 2026, Alphabet’s FY2026 capex midpoint will remain at or above $180B, with FY2027 qualitatively guided “higher” again in prepared remarks or Q&A (transcripts/IR pages will make this falsifiable). [5]
2. By January 15, 2027, the U.S. Commerce Department will publish at least one new tightening step or formal clarification that restricts AI‑chip shipments to Chinese firms or their overseas subsidiaries (Federal Register, BIS notices, or congressional testimony). [7]
3. By December 31, 2026, the KOSPI will have rallied at least 10% from its July 8, 2026 close of 7,246.79 as retail “buy‑the‑dip” flows and chip export strength intermittently counter AI‑spending angst (index data checkable). [6]

My take

Alphabet earnings are the fulcrum for the AI trade, but not because a bigger 2026 capex number rescues semis. If management doubles down on owning the stack, I’d stay overweight Alphabet and be selective in semis: HBM suppliers and advanced packaging ride the build; merchant compute faces share risk where TPUs or custom silicon take wallet. The SOX can bounce on positioning, but sustained upside likely waits for either a clearer monetization arc at the hyperscalers or policy relief from Washington and Brussels. Until then, favor the cash engines that control their own destiny—and make everyone else pay to plug in. [4][5]

Sources

[1] Alphabet earnings offer a moment of truth in this tech stock battle — Yahoo Finance (https://finance.yahoo.com/technology/article/alphabet-earnings-offer-a-moment-of-truth-in-this-tech-stock-battle-121514421.html) — Frames the “moment of truth,” the MAGS vs. SOX divergence, and Evercore ISI’s FCF warning.

[2] Chip stocks hit rocky patch. What’s next? — Reuters via Investing.com (https://www.investing.com/news/economy-news/chip-stocks-hit-rocky-patch-whats-next-4787795) — Documents the July semiconductor selloff and cites the SOX down >11% from its June record.

[3] Does China’s latest AI model finally equal US rivals? What scientists think — Nature (https://www.nature.com/articles/d41586-026-02281-2) — Reports on Moonshot’s Kimi K3 launch, demand spikes, and the narrowing performance gap.

[4] Hyperscalers Tap External Financing as AI Capex Outruns Cash Flow — FactSet Insight (https://insight.factset.com/hyperscalers-tap-external-financing-as-ai-capex-outruns-cash-flow) — Explains the financing strain; notes S&P’s Oracle downgrade tied to surging capex and negative FCF.

[5] Alphabet (GOOGL) Q4 2025 Earnings Call Transcript — The Motley Fool (https://www.fool.com/earnings/call-transcripts/2026/02/04/alphabet-googl-q4-2025-earnings-call-transcript/) — Provides Alphabet’s 2026 capex guide ($175–$185B) and commentary on depreciation and AI infrastructure priorities.

[6] South Korea’s KOSPI drops 20% from June record close as chipmakers drag — Reuters via MarketScreener (https://www.marketscreener.com/news/south-korea-s-kospi-drops-20-from-june-record-close-as-chipmakers-drag-ce7f5ed8de8ef522) — Establishes the KOSPI’s bear‑market move and chip‑led volatility shaping “buy‑the‑dip” episodes.

[7] Regulatory action on chips, AI is coming, Commerce official says — Reuters via MarketScreener (https://www.marketscreener.com/news/regulatory-action-on-chips-ai-is-coming-commerce-official-says-ce7f5edddb8bf121) — Signals pending U.S. export‑control actions on AI chips, a key variable for hyperscalers and semis.