After reading A/S-1, as a part-time AI investor, I closed the book with a long sigh: I’ve seen people show off, but never like this—so impressive! Holding $120 billion in cash, plus another $100 billion from the IPO—that’s $220 billion in cash. Who else?
@mediastorm_en A single video took down three million-follower influencers among Apple fans and caused a 10% return rate for the Duo. The current Apple Store Duo—I calculated that each Duo must be folded at least twice by 2,000 people per day, totaling 5,000 folds. The screen would be done in two days. https://t.co/5tkjiVHJYk
@stephzhan Zhan spent 11 years at Sequoia and was one of the youngest investors Sequoia ever hired. She has just announced the launch of her own early-stage AI fund while continuing as a partner at Sequoia. Her new fund’s LPs include Sequoia, Jeff Dean, Sachin Katti, and her Stanford professor and TSMC chief scientist advisor, Han-Shen Huang. She has invested in Skild AI, Reflection AI, and Recursive. In early-stage AI investing, Stephanie’s edge is her willingness to bet before others understand the opportunity.
The man named Alexander Wang has arrived: Meta releases Muse Spark Open, an open-weight model—free to download and commercially usable. To run it locally, you’ll need two Mac Studios with 512GB of RAM each, budgeting $10,000. On October 15, Meta released Muse Spark Open, making its model weights directly downloadable for free under the Muse Model License, including for commercial use. Meta calls it “the strongest open-weight model ever released in the U.S.” Since the Muse series launched, the flagship models—Muse Spark and Muse Wonder, released three days ago—were only accessible via Meta’s products and APIs; only the smaller Muse Glimmer had open weights. This is the first time Meta has released open weights for a large-scale Muse model. The model contains a total of 816 billion parameters, but only 42 billion are activated per token (MoE architecture), enabling fast computation—though the entire model must still fit into memory. It supports a context length of 1 million tokens, processing text, images, video, and audio, and outputs text, with a primary focus on code generation and enabling AI to operate autonomously over extended periods.
Broadcom CEO Chen Fuyang said today at the DBS forum in Singapore that many AI investors are like gamblers placing bets at a casino; the biggest bottleneck for AI in the future may not be chips, memory, or electricity, but capital and interest costs. Intel’s Chen Liwu followed up by saying that right now, the issue is capital and cost of funds; going forward, everyone will need to focus on returns. The question is: with financing interest rates of 9–11%, who will survive until the day returns materialize? And who will truly win?
Core Logic Summary After Reading: Will yield issues delay “Light into Rack”? No—they will only change the entry method. CPO is soldered in and not repairable; NPO (pluggable) allows field replacement, making it more practical from an engineering standpoint. Regardless of whether CPO or NPO is adopted, laser sources remain essential. Copper cables hit clear physical limits at higher speeds; optical integration into racks is a trend, not an option. The real bottleneck remains InP laser chips, not the optical engine itself. This is why long-term capital is still largely on the sidelines—no one can accurately predict how long the shortage will last. Two risks more significant than yield: China’s CW lasers are oversupplied: Even if not directly entering the U.S. market, they will depress global prices through other channels. Low-power sources like VCSELs: Could bypass the moat around high-power InP lasers. Lumentum itself is investing in this path, indicating the moat is not absolute. Simple Conclusion: Short-term: CPO volume remains limited; contribution to revenue for assemblers like Hon Hai is still minor. Medium-term: NPO will become a practical transitional path, shifting optical engines from “soldered” to “pluggable.” Long-term: InP laser capacity expansion requires 3–5 years; shortages and pricing pressure will persist for some time, but Chinese capacity and alternative technologies remain the biggest uncertainties.
Catch-all alert: If you're still using the Southeast Asia version of Ledger and are a real person with a blue checkmark, please read this immediately! 🙋♂
Mu's operating revenue exceeds Apple's. Can you believe it? https://t.co/rievkYgCI0
Meta hires Virtue AI founders in June 2026 Meta fires Virtue AI researchers in Oct 2026