📌 X Insight Update[x_fin] (2026/07/23 19:16)

Original Insight Summary

$AMD has a credible long setup because coding AI can speed up ROCm development and narrow the gap with CUDA. The key logic: if AI tools can write and optimize more of the software stack, CUDA’s moat gets less absolute. AMD’s investment in Anthropic and planned use of Claude Code for chip design and software reinforces that flywheel. 1
Chinese open-source AI is not automatically bearish for $NVDA. It can actually reinforce NVIDIA’s dominance because many Chinese models are still trained on NVIDIA GPUs and then released in forms that expand the downstream ecosystem. More open models can mean more CUDA-native demand, not less. 2
DeepSeek working with Huawei and potentially securing about 16,000 Huawei AI chips matters because it points to a domestic AI-stack buildout. The bigger read-through: AI code generation plus languages like TileLang could rapidly lower the entry barriers created by the CUDA ecosystem. 3

Memory-stock panic selling on the YMTC market-share headline looks overdone. The warning was about not underestimating YMTC, not a signal to dump the whole memory trade. The core setup remains tight: NAND is still in short supply, and that shortage is expected to persist. 4

Texas Instruments looks positioned for share gains, not just a cyclical bounce. 2Q26 revenue rose +13% q/q and +23% y/y to $5.46B, gross margin hit 61.4% (+340bp q/q), and the growth base widened from industrial and data center into automotive. 3Q26 guidance of $5.65B–$6.15B, midpoint $5.90B, implies +8% q/q and above-seasonal momentum. 5
STMicro is showing an accelerating recovery with an AI-datacenter kicker. 2Q26 revenue rose +12.7% q/q and +26.0% y/y to $3.49B, gross margin reached 34.8% (+100bps q/q), and the QSep guide of $3.70B implies +6.2% q/q and +16.2% y/y at 37% GM (+220bps q/q). The data-center target being 2x vs 1Q26 makes the recovery story broader than a plain cyclical rebound. 6

The Pop Mart put-selling mistake is a clean lesson in contract specs and hidden leverage. Selling 22.5万张 Hong Kong put contracts was initially treated like U.S. options at 100 shares per contract, but HK contracts map to 200 shares. Actual exposure doubled. If fully assigned, the cash needed would be close to HK$70B to take 45M shares, enough to become the third-largest shareholder. Simple mechanics, huge tail risk. 7
BitMEX shutting down while U.S.-regulated perpetual futures go live marks a regime change. On May 29, the CFTC brought perpetuals under U.S. federal regulation and approved two platforms, Coinbase and Kalshi, triggering sharp weakness in CME and CBOE. The punchline: the product is entering the U.S. mainstream just as the original pioneer exits. One era ends; another venue captures the flow. 8

BTC’s biggest edge is better framed as availability, not abstract “freedom.” Data availability. Tool availability. Market availability. Crypto’s moat is that retail users can access options apps, market data, and trading infrastructure with far fewer gates than traditional finance. That accessibility is the real liquidity engine. 9

The so-called TACO trade has less room to work if the U.S. appears to be backing down in front of Iran. Iran-linked pressure is escalating on two fronts: U.S. military casualties and Houthi attacks on Red Sea tankers after U.S. warnings over oil-shipping attacks. That setup caps the “always de-escalate” assumption and keeps geopolitical risk premium alive. 10

Driven looks useful for Hong Kong equity research because its data quality appears stronger than general-purpose models. The new Skill Store also adds workflow leverage by letting users install community-uploaded skills. For stock research, specialized data beats generic chatbot output. 11

#NVDA #AMD