Strategy has acquired 1,142 BTC for ~$90.0 million at ~$78,815 per bitcoin. As of 2/8/2026, we hodl 714,644 $BTC acquired for ~$54.35 billion at ~$76,056 per bitcoin. $MSTR $STRC strategy.com/press/strategy-…

First, the good part of the Anthropic ads: they are funny, and I laughed. But I wonder why Anthropic would go for something so clearly dishonest. Our most important principle for ads says that we won’t do exactly this; we would obviously never run ads in the way Anthropic depicts them. We are not stupid and we know our users would reject that. I guess it’s on brand for Anthropic doublespeak to use a deceptive ad to critique theoretical deceptive ads that aren’t real, but a Super Bowl ad is not where I would expect it. More importantly, we believe everyone deserves to use AI and are committed to free access, because we believe access creates agency. More Texans use ChatGPT for free than total people use Claude in the US, so we have a differently-shaped problem than they do. (If you want to pay for ChatGPT Plus or Pro, we don't show you ads.) Anthropic serves an expensive product to rich people. We are glad they do that and we are doing that too, but we also feel strongly that we need to bring AI to billions of people who can’t pay for subscriptions. Maybe even more importantly: Anthropic wants to control what people do with AI—they block companies they don't like from using their coding product (including us), they want to write the rules themselves for what people can and can't use AI for, and now they also want to tell other companies what their business models can be. We are committed to broad, democratic decision making in addition to access. We are also committed to building the most resilient ecosystem for advanced AI. We care a great deal about safe, broadly beneficial AGI, and we know the only way to get there is to work with the world to prepare. One authoritarian company won't get us there on their own, to say nothing of the other obvious risks. It is a dark path. As for our Super Bowl ad: it’s about builders, and how anyone can now build anything. We are enjoying watching so many people switch to Codex. There have now been 500,000 app downloads since launch on Monday, and we think builders are really going to love what’s coming in the next few weeks. I believe Codex is going to win. We will continue to work hard to make even more intelligence available for lower and lower prices to our users. This time belongs to the builders, not the people who want to control them.

The company Dustin Moskowitz founded (Asana, $2b mkt cap) is now worth less than his stake in Anthropic ($4b).

Guardrails: - enable sandbox - enable white-list if you want to run commands out of it - read security doc - use model that has best-what-we-have prompt inject defense - run `clawdbot security audit` - don't add it to group chats if it is your personal bot docs.clawd.bot/gateway/sandb…

Compute & storage demand would go parabolic once proper long video generation comes online (15min-1hr) Lots of good memory (SSD & HDD) stocks can be picked up rn

Bought Sandisk around this time and never sold and I am still not selling anytime soon. My thesis has always been the same… In modern LLM’s storing & referencing KV Cache is the largest bottleneck. For models like ChatGPT & Gemini it has got so big that you can’t fit it on a single node NVL72 node you need multiple nodes with networking in between. KV Cache also scales linearly with model size & context so the chase for larger and larger MoE models makes this a behemoth. So where do you store the KV Cache? To answer this we need to understand the basics of computer engineering. There are a few ways you can tackle it, you could add more HBM memory to hold it, scale to more servers and use networking to share it between them, or keep it on “disk” which is muuuuch slower than HBM or even DDR memory. So for the longest we just decided to scale out to more servers and this makes sense at first if you remove yourself from the inference mindset. If you are doing training or sharing hardware among researchers this is actually a more ideal setup no big KV Cache to carry around training small experiments and focusing on bandwidth over latency… But inference is a latency game. Adding network hops and switches in between to read memory adds latency and all the networking hardware Nvidia makes is optimized for bandwidth not latency. So when you are serving models the literal physical location of a specific part of your cache ends up dramatically increasing your tail latency. If you think LLM serving was going to keep growing this was not going to be a sustainable solution and could leave the door open for a competitor to come up with an inference only server solution which I’m sure big labs and big tech were researching. So you might think just add more HBM? Adding more HBM increases your costs lowers your yield and negligibly improves your memory bandwidth compared to just increasing the pin speed which gives way more bang for buck. Even if you did stack a terabyte of HBM on a GPU the bandwidth won’t scale to make it usable and having more HBM will increase the time you need to refresh it to stop data corruption which will increase latency. So with these constraint on HBM & Networking killing inference latency what do you do? Contrary to what you would get taught in your EECS class you move up the hierarchy to “disk” and the fastest memory there is flash. Instead of having more and more nodes of GPUs with expensive networking equipment you can just do the brain dead simple thing and pack a bunch of flash on each node. Flash access latency is still lower than networking and if you DMA into your GPU ‘s memory cleverly you can hide that latency by overlapping writing to HBM from disk with consuming it in logic. The simple dumb easy solution wins you just gotta write better hardware drivers lmao. This is why Sandisk & Western Digital were always going to win but Sandisk was a more pure play flash company so their top line exploded even more. I don’t think the prevalence of flash in AI is fully appreciated yet either. As I said longer context lengths & larger models = larger KV cache = more flash. Additionally with video models you need even longer context windows fast access to video’s you generated for users videos for training. The dance between HBM, Networking, & Flash will most likely lean towards flash every-time just cause of scale & cost. I fully expect Sandisk to be worth $150B by the end of this year b/c of this.

*OKLO RISES 12% AS THE COMPANY RUMORED TO BEGIN SOIL SAMPLES FOR 2045 SITE $OKLO wow