Original Insight Digest

🤖 AI Models & Infrastructure

• MiniMax could become attractive on further high-volume downside. The setup: diluted market cap around 90 billion, post-unlock volume has clearly expanded, chips may be getting flushed, and valuation is already far below Kimi’s latest round. The core logic is that grok4.5 proves open-source SOTA can still be chased, making the huge premiums on GLM and DS look less justified. LLMs are still in a growth phase, not yet a zero-sum game. 1

• CPU demand is not just a new cycle. Part of it is catch-up capex. Over the past three years, AI chips were deployed at massive scale without enough matching CPU buildout, so today’s CPU demand is partly backlog repair rather than a purely structural shift. 2

• CPUs have become a fresh market focus because OpenAI and Anthropic signed large-scale cloud partnerships with Amazon, Google, and Microsoft. The read-through: AI workloads are pulling meaningful cloud CPU consumption, not only GPU demand. 3

• NVIDIA expecting memory shortages to persist for several years matters because NVIDIA has among the best visibility into global AI demand. If that view is right, memory remains a multi-year bottleneck, not a short inventory squeeze. 4

🚀 Robotics, Space & Defense Tech

• 1X Tech’s dexterous hand is a real surprise. If it is fully self-developed rather than co-built with another dexterous-hand company, the Norway-founded humanoid robotics player deserves serious respect. The production angle is not vaporware either: its NEO Factory in Hayward, California has weekly capacity of 200 units. 5

• Commercial space has moved from 0-1 into 1-10. Once that stage begins, absolute technical gaps matter less, capital floods in, and the sector can enter explosive growth. The caveat: current scale still cannot be compared with semiconductors. 6

• The Long March 10B recovery video suggests net-assisted recovery may not be necessary forever. The net is likely there to improve touchdown success at the final moment. At this pace, the frame could be removed soon, moving toward a SpaceX Falcon 9-style landing path. $SPCX 7

• Counter-drone systems look like a huge market. The demand runway is getting harder to ignore. 8

🏦 Macro Policy & Market Structure

• The Fed’s five reform working groups look like a chairman-led parallel policy design layer. Formal decision power still sits with the FOMC and the Board, but agenda-setting, knowledge production, data access, and public narrative power appear to be concentrating around the Chair and selected external experts. The structure even resembles a “central working group” style often used in Chinese institutional reform. 9

• rToken is trying to use a new wrapper to attack an old brokerage market. The pain points are clear: traditional U.S. equities are still locked into 9:30-16:00 ET, T+1 settlement, and no weekend or holiday trading. In a more volatile market, that legacy structure is increasingly out of sync with trader demand. Bitget rToken offers an alternative trading model. 10

₿ Crypto Infrastructure & Trading

• $BTC still looks technically constructive as the $60k ish key level continues to hold well. The market is defending the line instead of breaking down. 11

• GreeksLive is building core infrastructure for $BTC options trading, not just shipping surface-level tools. Options backtesting is not simply “input strategy, output PnL curve”; the engineering burden is much heavier. Building through a weak crypto cycle earns respect because infra gets built in the trough, not during hype. 12

• Binance stablecoin rails already look like financial infrastructure: $53 billion in reserves, 21 million merchants, and $500 billion+ in TradFi perpetuals. Its stablecoin reserve share is cited at 57%, leading the second-largest exchange by about $42 billion. The scale points to network effects, not just exchange balance-sheet strength. 13

📈 Equity Ideas

• $META looks cheap on valuation. The key point is simple: on P/E, $META is cheaper than most $SPX companies, giving it room for rerating if sentiment turns. 14

• $META only needs valuation repair to the $GOOG level to create meaningful upside. Compared with semiconductors that dump on every small headline, $META looks cleaner and more stable. 15