Wildreamz

In general statistics, the mean is the signal the outlier is the noise. In investing, the outlier is the signal, the mean is the noise.

12 years of Tesla AI hardware evolution HW1 to HW5 - this is the heart of Tesla FSD The biggest leap yet just taped out (April 2026)

Map of Chongqing's metro system in 3D. Chongqing is called the "mountain city". Building metro here is more difficult than in plain cities. Chongqing’s metro total length ranks 7th in the world, with over 550 km.

Amazing application of AI+AR/VR/XR!

@YffuL56 x.com/Wildreamz/stat…

Crazy how some people still don't understand this moment

@YffuL56 One of my favourite moment on a personal level.

@Google @googledocs Fixed. Thanks.

@Google @googledocs $PSTG recently changed their ticker to $P. Google Sheet formulas are breaking. Please kindly update, thanks!

*DEEPSEEK RAISING MONEY AT $10B-PLUS VALUATION: INFORMATION *DEEPSEEK LOOKS FOR DATA CENTER ENGINEERS IN INNER MONGOLIA Deepseek is raising money in an outside round for the first time at a $10B valuation. Ridiculously small compared to US AI companies. They’re looking to raise $300m and are also seeking data center engineers in Mongolia. I bet Mongolia’s crawling with DC engineers. Deepseek’s next-gen flagship model, DeepSeek V4 was initially expected to be released in February during the Chinese Lunar New Year but has been delayed. Release is now anticipated before the end of April.

Don’t count pluggable optics out just yet in the datacenter. XPO looks like the last hurrah before co-packaged optics takes over. $ANET $NVDA $AVGO nextplatform.com/connect/2026/0…

BREAKING: Iran officially reopens Strait of Hormuz

I am grateful to the NBPA for advocating on my behalf and to the NBA for their fair decision. It was so important to me to be present for the birth of my daughter in December and I appreciate Mark, Jeanie, Rob, JJ, and the entire Lakers organization for fully supporting me and allowing me to travel to be there. This season has been so special to me because of what my teammates and I have been able to accomplish, and I am honored to have the opportunity to be considered for the league’s end-of-season awards.

BREAKING: LUKA WON HIS 65-GAME APPEAL AND IS ELIGIBLE FOR SEASON AWARDS 👀 🪄

Interview with a $GOOGL employee explaining how hyperscaler pricing for big AI labs works and the age gap of production-grade hardware clusters: 1. The expert emphasizes that concerns about signing multi-year GPU contracts as newer hardware generations approach are largely overstated. The gap between a chip's announcement and its availability in production-grade clusters is typically 3-4 years, meaning contracted hardware remains relevant well into the contract term. The expert highlights H100 as an example, explaining that most AI workloads still run on it despite newer generations being available for some time now. 2. According to him, pricing for legacy GPUs does come down as newer generations arrive, but for large AI labs, this dynamic is largely irrelevant. Their contracts are structured upfront with deeply discounted prices locked in across multiple hardware generations and consumption models, so they are not exposed to the same list-price fluctuations that regular enterprise customers face. 3. The expert also highlights three contractual protections that matter most for AI labs. Exit clauses tied to performance guarantees give labs a way out if capacity fails to meet agreed targets. Portability provisions protect against vendor lock-in, ensuring spending can be paused or redirected if platform constraints arise. Power guarantees are the most critical of the three. Committed GPU capacity is effectively worthless without sufficient power to run it, making power delivery commitments the single most important negotiating lever in any large infrastructure deal. 4. The expert explains that one-year committed use typically yields around 20% off list price, three-year commitments get to 30-40%, and Spot instances can reach 70%, though these are not suitable for production workloads. For AI labs signing large strategic contracts, discounts can reach 70-80% off retail pricing. The expert notes that hyperscalers are willing to offer capacity to strategic partners like Anthropic at near bare minimum margins, as the relationship itself carries strategic value that goes beyond the economics of any single deal. found on @AlphaSenseInc

@TeslaJoy 😄 I hope Elon is right on this. It’s the best outcome for $tsla That said, eventually AI5 could make their way to cars too, and add additional features. I have no doubt.

@Wildreamz You just wait.

AI4 owners, your days are numbered. 🤣

People will dunk on Jensen for this, but this is the exact mindset that it takes to be the best in the world. America will win.

@lialiascg 好多年前去过了，现在应该变化挺大。 东西便宜，美食不错。 逛吃、看风景、体验当地氛围，应该还是挺值得去的。

@Wildreamz 去过吗 好玩吗

Kind want to visit Vietnam and see what it’s like over there 👀

Let's talk about the AI5 picture. Memory From what I can gather from the grainy low res photo, those RAM chips appear to be SK Hynix H58G66DK9QX170N 8GB LPDDR5X with 9600Mbps bandwidth. 12 modules = 96GB @ 1.15TB/s The Die Size appears to be half-reticle (~430mm2). This gives it a yield and cost advantage over full reticle dies such as Nvidia's H100 (<800mm2). If we assume Tesla is using TSMC 3nm process, this gives the chip 108-125B transistors. Performance With that many transistors and memory performance, when constrained to ~150W such as a car or Optimus, we're looking at 2000-2500 TOPS which matches up nicely with a H100. Unconstrained, such as in a datacentre it could be much more. The Package This packaging is pretty cool. It includes memory on package which gives huge memory latency advantages than the typical memory on board config. To me, this amount of RAM packaged in this way is way overkill for a car. I believe we're looking at the datacentre version in this image. I believe for the cars or Optimus, we will see a traditional memory on board config (with less of it - e.g 32GB).

Elon has reiterated that there are no plans to use AI5 chips in customer vehicles for the foreseeable future. The chip will be focused on @Tesla_Optimus and supercomputers. Tesla remains on track for releasing Full Self-Driving Unsupervised on the latest FSD computer (AI4).