650 Group Underlying Assumptions; PC Units Finally Decline; IG Downgrades of MSFT/ META; Gokul TSM Preview Notes

Research evidence excerpt

650 Group Underlying Assumptions; PC Units Finally Decline; IG Downgrades of MSFT/ META; Gokul TSM Preview Notes

Allow NVIDIA H200 Imports, but Approvals May Fall Below 200K Chips. Beijing is reportedly set

to let firms like Alibaba, ByteDance, and DeepSeek import NVIDIA H200 chips, but the cap looks like it’ll land under 200,000

units — less than half the volume requested by the companies earlier this year — and even those chips would be restricted to

AI training, with inference workloads pushed toward domestic alternatives like Huawei’s. It’s a partial reopening of NVIDIA’s

China market rather than a full one: real revenue upside, but capped volume and usage restrictions limit how much this

offsets the roughly 95% high-end market share NVIDIA lost when export controls first hit - Trendforce

• SpaceXAI, Cursor Launch Grok 4.5 AI Model for Finance, Legal Applications. SpaceXAI has unveiled a new artificial

intelligence model called Grok 4.5, built in partnership with AI coding startup Cursor, to gain ground on rivals like Anthropic

PBC and OpenAI. It’s designed to handle difficult, long-running tasks, including software engineering, legal and financial

services, and has increased cybersecurity capabilities. The release comes as part of a broader effort by SpaceXAI to catch up

in the AI race and attract more business customers, particularly in finance work to gain clients on Wall Street - Bloomberg

• The AI Superfans Companies Count On to Convert the Skeptics.Companies (Ropes & Gray, Citigroup, Mars/Kellanova)

are formalizing internal “AI champion” programs — employee volunteers who get early access and training in exchange for

one-on-one peer persuasion — because group demos don’t move adoption but personal conversations do; results are notable: