💻 AI infrastructure and semis

• Arm has been one of the cleanest ways to express the Agentic AI trade. The core logic is that Arm server CPUs fit the agentic workload stack better at the architecture and feature level, which helps explain why Arm just made a new high after the mid-April setup call. The same CPU-linked basket already saw strong follow-through: intel and 澜起科技 more than doubled; arm nearly doubled; amd、深南电路、海光 were up by several dozen points. 1

• Nvidia Vera CPU beating the latest x86 CPUs matters. It reinforces the view that CPU competition inside AI systems is shifting, and that incumbents in traditional x86 are no longer coasting. At the same time, Intel reportedly working on an 18A CPU to compete with the MacBook Neo shows the fight is moving into process and product co-design, not just raw branding. 2

• AMD signaling it needs more Taiwan capacity for the next 1, 2 and 3 years is a strong demand tell. The bigger read-through is that $TSM and advanced packaging are no longer side constraints; they are now foundational choke points across the AI semi stack. If AMD is still pushing partners to ramp faster here, supply remains tight where the real bottlenecks are. 3

🚀 Space and defense

• $RKLB buying Geost for about $275M already looks accretive on the strategic side. Less than a year later, that capability is helping win a $90M U.S. Space Force GEO satellite program, which suggests the M&A was not just story-telling but a fast track into higher-value national security payload, surveillance, and missile-warning work. 4

• The long-game bull case is that a Tesla + SpaceX combination feels increasingly plausible because the stack is converging: robotics, autonomous vehicles, energy storage, LLMs, batteries, rockets, chips, compute, connectivity, solar, and defense. The more those future-defining verticals start to overlap, the more an Alphabet-esque parent structure makes strategic sense. A future SpaceX IPO would likely be a major event if that path stays separate. 5

🏢 Enterprise AI adoption

• Big-company AI transformation is still jammed by internal friction, not model quality. The practical bottlenecks are ugly and slow: even getting Claude and similar tools onto employee computers can take months and requires fighting through IT controls; plugging legacy software into AI workflows often turns into fragile patchwork; and organizational coordination around AI remains messy. Net takeaway: enterprise AI rollout is real, but the install cycle is slower and more painful than top-down narratives imply. 6

🇯🇵 Japan tech takeaway

• The MLCC price hike signal was an early and useful read-through for Japan exposure, and 6981 was a key name tied to that thesis. The takeaway is less about the news itself and more about the edge: niche supply-chain datapoints like component pricing can front-run the move, but execution risk is still huge if the position gets shaken out too early. 7