Aya X Pulse · Semiconductors

Semiconductors · Weekly X Pulse

Sunday, August 16, 2026
Today's Heat Board

Tickers ranked by buzz on X (Twitter); bars show the bull/bear split.

BullishBearishMixed/Neutral

Tap a ticker to see its KOL chatter →

1NVIDIA45%
2Micron75%
3Applied Materials55%
4SK Hynix80%
5TSMC70%
6SMIC70%
7AMD65%
8SanDisk70%
9Intel55%
10BroadcomMixed
View the full board (19 tickers) →

Semiconductor Sector Weekly (2026-08-16)

Report date: 2026-08-16 (Sun) | Data cutoff: US close 2026-08-14, A-share/HK close 2026-08-14 | Sources: real-time X (Twitter) search (last 168 hours; industry signals over 14 days) + exchange closing snapshots for all price data. X posts are used for events and opinions only; every price/percentage change is independently verified against exchange data.

1. Key Events

  1. Applied Materials (AMAT) beat big on FQ3 and raised guidance, yet the stock closed down 5.1% Friday: revenue $9.12B, adj. EPS $3.50, both above consensus; Q4 guide of $10.25B ±$0.5B far above estimates; management said it is "further raising our Semiconductor Systems revenue expectations for calendar 2026." Good news, but the stock closed at $507.18 (-5.12%) — profit-taking at highs. @wallstengine @FirstSquawk 𝕏¹ 𝕏²
  2. WSJ: Nvidia cut its proposed $250B backstop for OpenAI's Ohio data-center project to under $120B, reportedly after pushback from its own investors — the week's biggest cautionary signal on the AI capex narrative. @HedgieMarkets (citing WSJ) 𝕏
  3. SK Hynix Chairman Chey Tae-won on CNBC: memory demand is roughly doubling this year, "customers asking for about double their prior orders"; capacity takes 4-5 years of lead time, so 2027 is when the shortage bites hardest. @firesidealpha 𝕏
  4. SMIC delivered sharply higher profit and revenue on robust legacy/specialty orders; co-CEO Zhao Haijun said Chinese internet companies' hardware investment is "above expectations" and compute chips are "in short supply." H-shares jumped 4.81% Friday to HK$70.80 on heavy volume; A-shares +2.65% to ¥132.87. @ChinaBeigeBook @cnfinancewatch 𝕏¹ 𝕏²
  5. J.P. Morgan (via @jukan05): HBM will be in even shorter supply in 2028 than in 2027, implying HBM configurations of 2028 AI accelerators may be scaled back. 𝕏
  6. Report that the US government urged Apple not to buy Chinese memory chips (CXMT/YMTC) — read on X as bullish for Micron/SK Hynix/Samsung; very high engagement (≈351k views) but no Reuters/Bloomberg confirmation found this week — treat as unverified. @aleabitoreddit 𝕏
  7. Reuters: Nvidia eyes a $3B investment in SB Energy under the OpenAI data-center deal (per The Information). @Reuters 𝕏
  8. Memory stocks led again Friday: SanDisk +7.39% ($1,641.11), Western Digital +4.41%, Micron +2.30% ($971.66); AMD surged 6.50% to $514.39 amid X chatter about an AMD×Google TPU-related tie-up (not confirmed by major media; price moves are exchange data). @Kay2289123 𝕏

2. Institutional & Media Coverage

Factual reporting

  • AMAT earnings (@wallstengine, @FirstSquawk, @unusual_whales, merged): $9.12B revenue / $3.50 EPS double beat, Semiconductor Systems at $7.04B; FirstSquawk noted "customers like SK Hynix and Micron race to build new facilities amid memory-chip shortages." Options volume had already surged to more than twice the 20-day average pre-print. 𝕏¹ 𝕏² 𝕏³
  • Reuters: Nvidia weighs $3B into SB Energy for OpenAI data-center power — AI capex spilling into power infrastructure. @Reuters 𝕏
  • Bloomberg (@business): Leopold Aschenbrenner's Situational Awareness fund added Micron and SanDisk in Q2 while trimming options positions in Nvidia, Broadcom and AMD — a data point for the "long memory, lighten mega-cap AI compute" rotation. 𝕏
  • @StockMKTNewz (citing Bloomberg): Foxconn (Hon Hai) revenue +54.2% YoY to $29.4B, net income $1.9B, above estimates — confirming AI-server assembly volumes. 𝕏
  • @ChinaBeigeBook: SMIC "recorded sharply higher net profit & revenue, cashing in on robust orders for legacy & specialty semiconductors as AI-driven demand squeezes production capacity worldwide." 𝕏

Opinions / ratings

  • BofA reiterated Intel Buy, PT $145 (last close $102.50): "Over time, we expect Intel to potentially address ~8-10% of the ~$380bn CY30 wafer foundry TAM," plus 25-30% of 2.5D/3D packaging TAM; server ASP +43% YoY to $1,200. @wallstengine 𝕏
  • J.P. Morgan (HBM model update): 2028 shortage worse than 2027 but better than May estimates — pricing tailwind for memory makers, configuration constraint for accelerator vendors. Via @jukan05. 𝕏

3. KOL Bull & Bear Views

Bullish

  • @MelvinInvests (Micron, ≈124 likes / 12k views): "Micron's HBM shipments are set to 10x by 2027... Its entire 2026 HBM output is sold out... DRAM prices have risen more than 200% since early 2025." 𝕏
  • @aleabitoreddit (Micron/Hynix/Samsung, ≈2k likes / 351k views): "US Government urges $AAPL not to buy Chinese memory chips. $MU, $SKHY, and Samsung are happy to hear this... CXMT/YMTC is price hiking too." 𝕏
  • @firesidealpha (SK Hynix, ≈404 likes / 270 bookmarks, CNBC interview clip): Chey Tae-won — "We are in the AI era, memory demand is high. AI is still a child (early)"; demand doubling this year, 2027 the tightest year. 𝕏
  • @Kay2289123 (AMD / industry structure, ≈185 likes / 39k views): AI chips are moving from "whose GPU has the most compute" to "who can assemble CPU + XPU + HBM + I/O + networking + optics into the best system... the next step for HBM isn't just price hikes — it's Custom HBM." 𝕏
  • @choblin29 (NAND, citing Counterpoint): "AI inference is now eating almost HALF of the world's NAND... Server SSDs are expected to consume more than 50% of all NAND bits by year-end." 𝕏
  • @cnfinancewatch (SMIC / A-share semis, ≈11-194 likes): SMIC co-CEO's "above expectations" hardware-investment and "short supply" compute comments; separately called Cambricon's H1 (revenue ¥5.996B +108%, net profit ¥2.311B +123%, per company results) "full validation of the domestic AI-chip champion thesis." 𝕏¹ 𝕏²

Bearish / Cautious

  • @jukan05 (HBM/accelerators, ≈636 likes): citing JPM — "HBM will actually be in shorter supply in 2028 than in 2027 — meaning that the HBM configurations of 2028 AI accelerators will also be scaled back." An implicit negative for downstream compute platforms. 𝕏
  • @HedgieMarkets (Nvidia / AI capex, ≈347 likes): "The $250 billion guarantee... got cut to under $120 billion after Nvidia's own investors pushed back" (per WSJ) — credit backstops along the AI capex chain are tightening. 𝕏
  • @leshka_eth (Nvidia/China, ≈59 likes / 14k views): "CHINA JUST COMPLETELY BANNED NVIDIA CHIPS... THE ENTIRE AI TRADE IS STARTING TO CHANGE" — note: cross-checked; no Reuters/Bloomberg/WSJ/FT account reported a "complete ban" this week. Retail-level circulation only — treat with caution. 𝕏

4. Buzz & Sentiment Shifts

  • Memory complex (Micron/SanDisk/WDC/SK Hynix): the hottest semis sub-sector on X this week, roughly 80/20 bullish. Drivers: Chey Tae-won's "demand doubling" remarks, JPM's tighter HBM model, and the unverified "Apple told to drop Chinese memory" story. Versus last week (HBM price hikes as the main line), sentiment got hotter and broadened from DRAM/HBM into NAND/enterprise SSD.
  • Nvidia & mega-cap AI compute: buzz stayed elevated but sentiment turned mixed (~45% bullish) — the WSJ-reported OpenAI backstop cut and the China "ban" rumor weighed; NVDA closed flat Friday (-0.06%), Broadcom fell 5.94%, while 13F data showed rotation into memory. The clearest marginal change from last week's uniform bullishness.
  • A-share / HK semis: SMIC's "short supply" earnings-call comments lit up sentiment — H-shares +4.81% on heavy volume, A-shares +2.65%, JCET +1.14%; Cambricon (-1.13%) and Hygon (-1.75%) consolidated post-results. Cross-market linkage was explicit: AMAT's memory-customer capex narrative ↔ Korean/Taiwanese memory & foundry tightness ↔ A-share foundry/OSAT sentiment.

5. First-hand Industry Signals (Last 14 Days)

Date/TimeCompanyEvent TypeOne-line EventStatusSource
08-11 06:32 ETTSMCAdv. packaging ramp/yieldCoWoS 5.5x-reticle in mass production with >98% yield on multiple AI products; SoIC 6μm hybrid bonding in HVM, shrinking to 4.5μm by 2029; 10 advanced packaging fabs, CoWoS capacity doubling annually for 3 yearsConfirmed (TSMC VP Jun He, public remarks)@dnystedt 𝕏
08-09 20:12 ETTSMCNew capacity (CoWoS)In talks to acquire two AUO fabs in Central Taiwan (7.5G/5G) for >NT$30B to convert into CoWoS capacityRumor (media, unnamed sources)@dnystedt 𝕏
08-13 21:09 ETTSMC / NvidiaLeading-edge node + packagingNvidia's next-gen Feynman platform goes straight to A16 (enhanced 2nm); SoIC capacity targeted at 50k wafers/month by end-2027; AP7/AP8 packaging fabs pulled forwardMedia report (DigiTimes)@dnystedt @jukan05 𝕏¹ 𝕏²
08-14 01:50 ETSamsungNew capacity (2nm)Considering converting NRD-K Line 2 from R&D to foundry production, expanding 2nm capacity for custom-HBM base dies and HBM5Rumor (Korean media)@jukan05 𝕏
08-13 09:59 ETNvidiaHBM generationRubin Ultra expected to skip HBM4 for 8-hi/12-hi HBM4E, up to 768GB per packageRumor (Aletheia sources)@jukan05 𝕏
08-14Nanya TechNew capacity (DRAM)Plans new 12-inch DRAM fabs in Yunlin and Pingtung (incl. sub-10nm for AI and customized DRAM); NT$300B Linkou 5A fab due for mass production in 2028Confirmed (company plan)@dnystedt 𝕏
~08-12Lam Research / ASEExpansionSemiEngineering weekly flagged Lam Research and ASE expansions; CPO system-architecture initiative launchedConfirmed (trade-media roundup)@SemiEngineering 𝕏

Interpretation (top three)

  • TSMC's CoWoS 5.5x reticle at >98% yield: advanced packaging has moved from "capacity bottleneck" to "large-size + high-yield" — directly enabling Rubin-generation mega-interposer designs. Combined with the AUO fab acquisition talks and AP7/AP8 pull-ins, TSMC clearly sees 2027-28 packaging demand exceeding current plans — a sustained order signal for bonding/test equipment and Taiwan's packaging chain.
  • Nvidia Feynman straight to A16 + Rubin Ultra to HBM4E: platform cadence keeps compressing (newest node, skipping HBM4 for HBM4E), deepening the bind to TSMC's leading edge and memory makers' newest generation; with JPM modeling 2028 HBM even tighter, HBM4E allocation becomes the 2027-28 accelerator battleground.
  • Samsung's NRD-K conversion to 2nm for custom-HBM base dies: corroborates this week's "Custom HBM" debate (@Kay2289123) — logic-process base dies are welding the foundry and memory races together; Samsung is betting on a "2nm + HBM5" combo, with yield progress the metric to watch.

Disclaimer: This report is compiled from public information and real-time X (Twitter) searches for informational and educational purposes only, and does not constitute investment advice. Views in cited posts belong to their authors; items marked "rumor/unverified" lack authoritative confirmation. Price data are the latest exchange closing snapshots and may be delayed. Invest at your own risk.

Want a daily X pulse for your own watchlist?

$10 free credit on signup · Just $5 per Mtoken · 30% off first top-ups

Start Free

This content is AI-generated from public posts on X (Twitter), for reference only and not investment advice. Investing involves risk.