Skip to report
Next edition in 1630 min Sep 23, 23:00 - Sep 24, 12:45 Lisbon
Premarket Alpha Post-Market Alpha

Daily Alpha · X

The clearest verified X items: the 30-year at 5.44%, Muse retailer integrations, record Taiwan export orders, and Wright's warning on diesel export curbs.

1270posts scanned
241with a ticker
133verified claims
2220ranked voices

Themes on this desk

Long-end yields

The 30-year Treasury yield rose to 5.44%, its highest since June 2004 and near a 500bp move from the 2020 low, while the S&P Global US composite PMI rose to 58.4, a five-year high.

Memory pricing

TSMC is reported to plan 3-6% wafer price increases from January 2027, and Winbond said NOR flash may see 20-30% quarter-over-quarter ASP hikes every quarter to end-2027.

Meta Muse

Walmart, Best Buy, Gap, Sephora and Wayfair are integrating with Muse, and Zuckerberg says Meta expects to make money from Muse through a small transaction fee.

Ticker heat

?
Z-score shows how far this edition's mention count is above or below the ticker's own average across 20 comparable earlier editions. +2.7σ means mentions are 2.7 standard deviations above that average — simply, the ticker is being discussed much more often than usual. It measures attention, not a bullish signal or expected return. Ticker heat shows up to 12 tickers with at least 8 unique posts; gold begins at +2σ, and σ is hidden when history is too thin.
Sentiment →
HO=F -3.0%
last $4.50
8 posts · 10 voices +3.9σ
base 2.4
META +0.2%
last $727.39
58 posts · 28 voices +3.0σ
base 15.2
GOOGL +1.5%
last $343.21
13 posts · 13 voices +1.8σ
base 7.9
TLT -0.2%
last $77.53
22 posts · 19 voices +1.3σ
base 13.4
CRWV +0.9%
last $89.40
11 posts · 6 voices +1.1σ
base 4.8
MSFT +0.8%
last $516.80
9 posts · 6 voices +1.1σ
base 5.4
NBIS +4.5%
last $242.72
20 posts · 11 voices +0.9σ
base 14.2
IREN +2.7%
last $41.76
18 posts · 7 voices +0.9σ
base 8.8
AMZN +1.2%
last $251.31
8 posts · 7 voices +0.3σ
base 7.1
AMD +2.9%
last $633.25
10 posts · 8 voices -0.1σ
base 10.8
NVDA +1.5%
last $234.22
23 posts · 17 voices -0.4σ
base 37.1
USO -2.3%
last $146.59
16 posts · 5 voices -1.0σ
base 27.6

Top voices by smart followers and alpha score

Market Radar →
@unusual_whales
263 smart Alpha #983

Meta Muse agent commerce integrations span major retailers

unusual_whales reports $META announced Walmart, Best Buy, Gap, Sephora, Wayfair and many more are integrating with Muse to enable new shopping experiences, alongside no-camera smart glasses and VR spectacles with the Muse AI agent integrated into the glasses.

Meta, $META: Walmart, Best Buy, Gap, Sephora, Wayfair, and many more are integrating with Muse to enable new shopping experiences

A broad retailer integration list turns Muse from a chatbot into a commerce distribution channel, giving Meta a monetizable agentic shopping surface and giving partner retailers a new AI-driven demand channel.

Watch Whether Meta and named retailers confirm the integrations and disclose commercial terms or GMV contribution, and whether additional large retailers join.

Source →
day change +0.2%
since call -1.4%
@jukan05
97 smart Alpha #6

ABF substrate demand inflecting on CPU capacity needs; Kinsus earnings confirm

Jukan (@jukan05) argues ABF substrate stocks are rallying broadly because the market is realizing how much ABF substrate capacity CPUs will require, and cites Kinsus reporting strong earnings the same day as corroboration.

Looks like everyone is realizing just how much ABF substrate capacity CPUs will need. Kinsus also reported strong earnings today.

ABF substrate is a bottleneck input for advanced CPU/GPU packaging; a demand-driven re-rating would lift substrate suppliers and their upstream materials before it shows up in consensus estimates.

Watch Kinsus and peer substrate makers' next quarterly revenue/margin guidance, and whether ABF substrate pricing or capacity utilization data confirms tightening.

Source →
day change —
since call —
@bubbleboi
51 smart Alpha #185

TPU vs GPU inference gap attributed to MoE routing, with hybrid attention seen favoring systolic arrays

@bubbleboi argues TPUs/systolic arrays lagged Nvidia GPUs on inference mainly because they struggled with MoE routing, now fixed via XLA and custom kernels that densify routing; he expects a 'great schism' where hybrid-attention architectures run disproportionately better and cheaper on TPUs, Trainium and similar hardware, and expects the next Gemini to use hybrid attention.

But the real takeaway to me is it’s looking likely we get a great schism where certain architectures that use

If hybrid attention shifts inference economics toward non-Nvidia accelerators, it pressures Nvidia's inference share narrative while benefiting Google's TPU stack and other systolic-array silicon.

Watch Whether the next Gemini model discloses hybrid attention and whether third-party benchmarks show materially lower cost per token on TPUs/Trainium versus GPUs.

Source →
day change +1.5%
since call +1.3%
@the_ai_investor
34 smart Alpha #197

Taiwan August export orders hit record $102.96B, up 71.4% y/y, with CoWoS and leading-edge capacity tight into next year

Taiwan's August export orders crossed $100B for the first time at $102.96B, up 71.4% y/y, the 19th straight monthly gain; electronics orders rose 83.9% to $45.75B, ICT more than doubled to $34.21B, US orders jumped 88.9% to $42.13B, and the trade ministry forecasts September up 50.7-55.1% and full year around $1.1T, with CoWoS and leading-edge capacity staying tight into next year.

Taiwan's August export orders crossed $100B for the first time, hitting $102.96B, up 71.4% from a year ago.

Record AI/server-driven order growth and tight advanced packaging capacity support continued revenue momentum across the Taiwan supply chain and constrain competing accelerator supply.

Watch September export order data versus the 50.7-55.1% forecast and any CoWoS capacity expansion announcements from TSMC.

Source →
@tengyanAI
28 smart Alpha #432

Korean memory makers refuse ~$18B power prepayment, citing cycle uncertainty

Korea's power utility asked Samsung and SK Hynix to prepay roughly ₩25T (~$18B), about five years of power bills, to fund grid buildout for their chip clusters; both refused, reportedly because they cannot be sure the memory supercycle lasts five years. The author contrasts this with Samsung customers seeking 10-year supply deals and SK Hynix's CEO saying demand runs ahead of supply 'even beyond 2030,' concluding memory makers are happy to sell long contracts but not to sign one.

reportedly because they can't be sure the memory supercycle lasts five years.

The refusal is a rare revealed-preference signal from the supply side: the companies with the best information on memory demand are unwilling to underwrite a five-year horizon, which cuts against the bull case that the AI memory upcycle is structurally extended.

Watch Micron's Sep 30 report and whether ~$22B in customer deposits and commitments hits the balance sheet this quarter, as the author flags.

Source →
day change -2.5%
since call -0.4%