📌 X Insight Update[x_fin] (2026/06/23 21:00)

🧠 AI Compute & Infrastructure

Masayoshi Son’s pushback on Elon Musk’s $SPCX orbital data center idea is a clear “earth-first” compute thesis: the AI race gets won by whoever builds massive terrestrial capacity first, because launch, maintenance, and communication costs eat up the power-saving upside. “He who strikes first wins.” 2

$WMT signing its first nuclear power agreement with $CEG for ~176MW from Constellation’s Dresden facility is another datapoint that hyperscalers and large corporates are rotating toward nuclear for reliable long-duration power, not just green optics. 5

$TSM pushing for a 5–10% price hike across advanced nodes, including 7nm, points to foundry pricing power coming back. The logic is simple: management saw memory peers enjoying higher pricing and wants to capture the same upcycle dynamic. 1
DRAM remains structurally tight if HBM keeps eating wafer capacity. Deutsche Bank’s model has demand running above supply every year through 2030. The key squeeze: HBM rises from 18% to 35% of total capacity, tightening both HBM and standard DRAM at the same time. 18
$MU earnings are set up to be messy after the last couple days of massive volatility. Memory names are now trading into the print with elevated event risk, not a clean setup. 16

The KOSPI -10% overnight crash looks scary on the tape, but the key context matters: the index only fell back to levels from ten days ago. That makes the move look more like a violent reset than a structural breakdown. 12
Korea’s selloff had a clear policy trigger. The market fell nearly 10%, the biggest one-day drop since March 4, and an intraday drop of more than 8% triggered a 20-minute circuit breaker. The proposed tax reform around unrealized gains appears to be the main catalyst, turning policy risk into forced de-risking. 6

SoFi Composer is not an “AI agent trading for you” product. The real angle is rules-based strategy building in plain language: users define weights, conditions, and filters, then backtest before execution. That makes it more of a retail quant workflow tool than a black-box trading bot. 13

$ZETA reclaiming $20 premarket after rising more than 8% reflects a serious enterprise AI catalyst, not just headline hype. The $PLTR partnership rebuilds Zeta Data Cloud on Palantir Foundry and places Athena’s AI decision engine on top of enterprise ontology and governance. Management’s target of more than $100M in annual revenue over the next few years gives the deal real monetization heft. 7

The market is still being treated as a dip-buyer’s tape, but the setup is getting tested: $QQQ was down nearly 3% premarket, $MU earnings are on deck, and any bounce needs to respect key levels rather than blind bottom-fishing. 19