
Former Financial Analyst at a public equity fund, currently @MiniMax_AI @MiniMax_Code Product Manager
Shanghai
Joined October 2026
- Tweets37
- Following39
- Followers63
- Likes29
Can someone tell me exactly how many models Google has? Would the periodic table even be enough to cover them all?
tomorrow DeepMind will be dropping Gemini 4 Sodium, a fast yet powerful frontier model which costs just $11.75/M (introductory pricing, raises to $16.50 soon) and is their best model for various things and stuff, rivaling Argon, Carbon, Hydrogen, and Neon in several situations
37
Forecasts suggest Samsung’s HBM TSV capacity could rise from 180k wafers/month at end-2026 to 280k at end-2027, overtaking SK hynix’s 200k to 260k. Nvidia may account for 42% of 2027 HBM demand; Google’s share could rise from 22% to 31%.
Separately, an estimate cuts Southeast Asia’s data-center capacity forecast by 15% (~1.2GW) for 2028 and 2.5GW for 2030, with utilization at 65% and 72%, respectively.
46
Some channel checks suggest data centers due online in 2027–early 2028 remain largely on track, while the next wave may wait for policy clarity. Cooling and power-equipment orders could soften before recovering later in 2027.
There’s also talk of foundry price hikes around 10%–20% or more for some customers. Memory looks less uniform: conventional DRAM and HBM pricing may keep rising, while some forecasts point to flat-to-lower NAND contract prices through 2027.
48
1. True 3D over flat pixels: Atlas unifies generation and 3D reconstruction, outputting explicit depth and 3DGS with exact camera control.
2. Phone shots to Real-to-Sim: A few casual captures unlock cinematic bullet-time reframing and interactive 3D worlds for robotics.
We’re opening up access to Atlas to everyone today.
We’ve built apps and API endpoints to help you use Atlas any way you can imagine: bring your ideas for building a 3D world and come away with keyframes, video clips, or 3D Gaussian splats.
1
113
新燃气轮机在 2030 年前基本无供货?
Unconfirmed industry feedback suggests new gas turbines are largely unavailable for delivery before 2030. If accurate, long equipment lead times could make power—not just chips—a binding constraint on AI data-center expansion.
1
2
89
Why Chinese Big Tech can't create something like Muse and Grok?
teapawt.org/HanZJ_AIPM/status/2121…
53
Industry supply-chain checks suggest end-to-end yields from photonic wafers to qualified CPO optical engines are only around 30–40%, with key bonding equipment still maturing—potentially pushing volume shipments to late 2027 or 2028; these estimates remain unverified. In my view, Jensen Huang’s reported departure could also disrupt execution across NVIDIA’s frontier-technology roadmap.
43

















