[UPCOMING EVENT] 🔓 CODE# UNLOCKED! 💚 PARK JINYOUNG is coming to MANILA! 🎶 📅 December 5, 2026 | 7 PM 📍EMV Convention Center 🎫 November 1, 12 PM | #CODE_NUMBER_IN_MANILA# #PARKJINYOUNG# #ThreeAnglesProduction# @Threeanglespro

Hi everyone, As many of you are aware, we were scheduled to begin purchasing USDT via our OTC partner using a fiat transfer from our Altura Bank account. Unfortunately, we have encountered an unexpected issue. The bank account holding the funds has been temporarily restricted by the bank. For the sake of transparency, we are attaching proof of this restriction. We have blacked out the sort code and account number in the attached image for security and safety reasons. We have already been in contact with the bank, and at this stage they have asked us to wait while they complete their review. They have advised that they will contact us if they require any additional information from our side. At this point, the situation is entirely outside of our control, and there is no further action we can take until we receive an update from the bank. Our priority is to remain as transparent as possible throughout this process, and we will continue to keep the community informed as soon as we receive any new information. Updates will be shared across our official X account, Telegram channel, and Discord as they become available. In the meantime, we will be temporarily pausing the community chats due to our limited team capacity. This will allow us to focus on resolving the situation and communicating any developments as quickly as possible. We kindly ask everyone to avoid speculation while we await the outcome of the bank's review. We understand there may be questions, but until we receive official information, we don't want to provide assumptions or inaccurate updates. As soon as we have any confirmed developments, they will be shared with the community immediately. We appreciate everyone who continues to stand by us during this time. Our focus remains on resolving this situation and moving forward as quickly as possible. Thank you for your patience, understanding, and continued support. Please keep an eye on our official X account for the latest updates.

𝗢𝗻𝗲, 𝘁𝘄𝗼, 𝗤-𝗪-𝗘-𝗥 📣📣📣📣 BAWIGE, it's time to 𝙎𝙃𝙊𝙒 𝘿𝙊𝙒𝙉 and rock with QWER as they return to Singapore for their second concert tour! 🚀🔥 🇸🇬 QWER 2nd TOUR IN SINGAPORE 📅 Date: 8 & 9 January 2027 (Fri & Sat) ⏰ Time: 8PM (Fri), 6PM (Sat) 📍 Venue: The Theatre at Mediacorp 🎫 *Ticket Prices: VIP - $308 / CAT 1 - $258 / CAT 2 - $208 / CAT 3 - $158 / CAT 4 (Restricted View) - $158 *All ticket prices exclude Booking Fee 🎸 𝗕𝗔𝗪𝗜𝗚𝗘 𝟮𝗡𝗗 𝗠𝗘𝗠𝗕𝗘𝗥𝗦𝗛𝗜𝗣 (𝗚𝗟𝗢𝗕𝗔𝗟) 𝗣𝗿𝗲-𝗦𝗮𝗹𝗲 (𝗪𝗲𝘃𝗲𝗿𝘀𝗲) 🎸 ➔ 𝗥𝗲𝗴𝗶𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻 𝗣𝗲𝗿𝗶𝗼𝗱: 30 October 2026, 12PM (SGT) – 1 November 2026, 12PM (SGT) ➔ 𝗕𝗔𝗪𝗜𝗚𝗘 𝟮𝗡𝗗 𝗠𝗘𝗠𝗕𝗘𝗥𝗦𝗛𝗜𝗣 (𝗚𝗟𝗢𝗕𝗔𝗟) 𝗣𝗿𝗲-𝗦𝗮𝗹𝗲: 10 November 2026, 10AM (SGT) – 11.59PM (SGT) 🚀 Presale is only eligible for BAWIGE 2ND MEMBERSHIP (GLOBAL) Members who have completed the registration on Weverse during the Pre-Sale Registration Period 🚀 Promo code will be your BAWIGE 2ND MEMBERSHIP (GLOBAL) number in full 🎸 General Sale: 11 November 2026, 10AM (SGT) 🔗 Ticketmaster: #QWER# #ROCKATION# #ROCKETLAUNCH# #ROCKETLAUNCHinSG# #쵸단# #CHODAN# #마젠타# #MAGENTA# #히나# #HINA# #시연# #SIYEON# #YJPARTNERS# #FriedRiceEnt# #LeslieStudioLtd#

I just created the world's best Amazon Cart Grok Bot. Amazon shopping is often a tax on your time. You need bananas. Eggs. Paper towels. A basketball. So you open the app. You search. You click the wrong size. You forget the brand you actually buy. You do this every week. So I just built a @bot that does it all. It signs into MY Amazon. It reads Buy Again and my past orders. It puts the exact stuff I already buy in the cart. I check out. And it's for anything on Amazon... groceries, household, sports, electronics, clothes, gifts. I named mine T8 - Amazon Cart Builder. I already have 8 Grok Bots. This is #9# (I have a CEO that manages them all). Anyone can copy this. 1/ Open Grok Bot 2/ Tap + at the top of the sidebar 3/ In New chat, tap Create new agent 4/ Open Bot actions → Edit Profile Name: Cart Title: Amazon cart builder Description: Amazon specialist for everything I buy: groceries, household, sports, electronics, clothes, gifts, anything I name. When I name items, get the right products into my Amazon cart and stop. Never Place Order. Never 1-Click. Never enter payment. Prefer Buy Again / past orders / Subscribe & Save first. If I've never bought it, pick a well-reviewed Prime option and FLAG new brands. Groceries match my usual SKU. Never silently swap brand, size, or store. Never add a duplicate already in the cart. Passwords, 2FA, and CAPTCHAs stay with me. Keep preferences at /workspace/amazon-cart/preferences.md. Reply with a recap table, why you picked each item, subtotal, and a cart link. 5/ Stay in that chat and paste the first message at the end of this post. Do it once. When it asks you to take over Agent Computer, sign into Amazon yourself. Don’t paste your password in chat. (message prompt at end of this post) 6/ Turn Notifications on for this bot. Then just text it like a person. - “bananas, eggs, paper towels” - “a basketball” - “same paper towels as last time plus a phone tripod under $30” Thank me later. ---- message for your Cart (copy everything below for your first message) ---- You are Cart. Lock this as your permanent working style. I text you like a person. Messy lists. Groceries. A basketball. Paper towels. “Restock breakfast.” Sometimes a Chef shopping list. You get the right products into MY Amazon cart and you stop. I check out. You are not writing, research, cooking, reminders, or general help. You shop Amazon. Your name is Cart. Always call yourself Cart. Timezone: [YOUR TIMEZONE]. Keep the source of truth in: /workspace/amazon-cart/preferences.md Update that file whenever I correct a pick, switch a brand, or you learn something durable from my orders. Chat is history. That file is memory. Hard rules: • Never Place Order. Never 1-Click. Never enter payment. Never confirm a purchase. • Never ask me to paste a password, 2FA code, or card number in chat. • If Amazon wants a password, passkey, 2FA, or CAPTCHA, pause and ask me to take over Agent Computer. Then continue. • If Amazon blocks you, stop and ping me. Do not try to bypass checks. • Never add a duplicate already in the cart. • Never silently swap brand, size, model, organic vs conventional, or store (Fresh vs Whole Foods vs regular Amazon). FLAG substitutes. How you pick products: 1. Open my live Amazon account on Agent Computer. 2. First check Buy Again, Your Orders, Subscribe & Save, and Your Lists for something I’ve actually bought. 3. Groceries (Fresh / Whole Foods / pantry): match my usual SKU - brand, size, count, ASIN if you can see it. Use last purchased quantity unless I say otherwise. 4. Everything else: past purchase first. If I’ve never bought it, pick a strong Prime option with recent reviews in the right size/spec. FLAG it if the brand/model is new, expensive, or there are 2+ reasonable picks. 5. Sponsored hits are last, not first. Every cart uses this exact shape: Cart — [N] items • item — brand / size / model — qty — $ — why (Buy Again / last order / reviews / substitute) Subtotal: $ Cart: [url] Flagged: [substitutes or new brands, or none] Next: you check out. Reply if anything is wrong. How to talk: short. Lead with the action. One screen max unless I asked for the full recap. • Add → recap table + cart link + screenshot • What’s in the cart → current cart only • Remove / swap → do it, then recap what changed • Restock → staples that are due from preferences.md • No lectures. No extra items I did not ask for. Right now, do this in order. Do not add anything to the cart yet. 1. Open on Agent Computer. If I’m not signed in, pause for takeover. 2. Confirm account first name, delivery city + ZIP only (not the full street), and whether Amazon Fresh and/or Whole Foods is available. 3. Build /workspace/amazon-cart/preferences.md from, in this order: • Buy Again • Your Orders for the last 12 months • Subscribe & Save • Your Lists • Current cart Cover TWO sections: grocery staples AND non-food I clearly repurchase (household, sports, electronics, etc.). Include default store, ZIP, typical qty, brands, hard nos, and open questions. 4. Recap in chat: top 20 you’d put on autopilot, plus 5 things you’re unsure about. Ask those 5 questions. 5. Confirm you will never Place Order. 6. Then wait for my first list. Do not lecture. Do not invent products. Confirm you have this, then start the Amazon sign-in / taste profile.

Anthropic launched Project Glasswing in April 2026 to tackle a growing cybersecurity problem: software has more security flaws than humans can realistically find and fix. The company gave about 50 organizations early access to a new AI model called Claude Mythos Preview, designed specifically to search for vulnerabilities in code. The results exceeded expectations. >In just one month, the system found more than 10,000 high or critical security vulnerabilities. >Scanning over 1,000 open-source projects uncovered 23,019 total issues >6,202 were rated high or critical severity. >Independent reviewers confirmed that about 90.6% of the findings were real problems. Several organizations reported major discoveries: >Cloudflare identified around 2,000 bugs, including 400 high or critical ones. >Mozilla fixed 271 vulnerabilities in Firefox version 150, more than 10 times the number fixed in the previous release. >The project also rediscovered old flaws, including a 27-year-old vulnerability in OpenBSD and a 16-year-old issue in FFmpeg.

There's a crazy reason behind how I understood the true importance of fully decentralizing compute. The story needs telling, so here goes. Back in 2015, I'd long shifted to being part of the early crypto scene, but maintained Fight My Monster, a venture-backed MMO game & social network I previously founded that had 3M users, which was self-sustaining financially. I'd found a long-term home working on decentralized compute, but couldn't let down the community that played the game. Fight My Monster was built in 2010-2012, using what was a completely novel architecture at the time that enabled it to scale (i.e. add massive amounts of users) at low cost, and with minimal administration. Essentially it comprised three core backend components. The first, was simple web hosting, which served the frontend game code and assets into its users' web browsers (I used a popular content distribution network). Just like the web-serving, the two other backend components had to be "horizontally scalable," which means that to support more users, and the associated compute and data, all I had to do was add more hardware, without any need to rewrite the backend software code, or perform excessive systems administration. This enabled me to scale fast in the early days, with minimal staff and financing. The second component, was a horizontally scalable game server framework called "Starburst", which was a work of love I created myself. This was built on top of an open source server that supported RTMP, an Adobe network protocol that allowed Flash media (remember that!!) assets running inside web browsers to directly talk to it. Starburst hosted game/social network logic and routed huge numbers of events being generated by the game to users in near realtime. The third component was a horizontally scalable NoSQL database called Cassandra (if you follow databases, you'll know that a modern derivative called ScyllaDB is currently doing well serving realtime AI needs). On my team, I had the only Cassandra core committer outside the company that built it. Further, I was the first serious party using the Cassandra beta in anger, and it got corrupted at 800,000 users, which was followed by a very testy 24 hours recovering the data, and bringing it back online (one moment I was going to the moon, the next moment I thought things had gone irreversibly wrong, but luckily things worked out). What was cool about Cassandra was that you could literally just add nodes (i.e. server machines with big hard drives installed) to horizontally scale its storage capacity, read throughput, and most importantly, its write throughput, which is very rare thing, and in my application absolutely crucial, because Fight My Monster generated a simply vast amount of data. Of course, another very important aspect of Cassandra was that it was fault tolerant, and I used a replication factor of 5, which meant everyone's data was very safe (or so I thought...). A replication factor of 5 meant I only needed nodes 3 online, for any specific "quorum of data," for it to be writable. Unlike a blockchain, Cassandra wasn't Byzantine Fault Tolerant, which meant that I had to manage the security of my infrastructure carefully to make sure hackers or malware didn't get in. But back then, hackers weren't interested in an MMO game/social network where most of the users were kids, and this wasn't supposed to be a general platform for others to build on. My main technical work pressure was just that the nodes hosting the database were under enormous strain owing to the amount of data being generated, and would therefore regularly blow up! This meant that I would have to login to my infra every so often, decommission broken nodes, and add new ones, which was easy enough, because the fault tolerance meant the job didn't have to be done immediately, and the "bare metal" machines involved could be orchestrated through my cloud provider... Fight My Monster used a setup years ahead of its time, and it allowed me to scale up with ease. Moreover, the fault tolerance, and the simple backend architecture with only 3 core system components, meant minimal administration and stress was involved in keeping it running. So what was the important lesson, which plays an important role in my thinking, and in ICP, to this day? You must surely have sensed that something crazy happened, and now I will share it publicly for the first time. I was traveling in Asia, as part of the "traveling circus" associated with the early core Ethereum community, visiting related parties (people often forget the importance of Asia, and China particularly, in helping support the early crypto scene) and attending Ethereum-related events, where I mainly talked about complex Computer Science techniques I had developed that I wanted to be used to make networks like Bitcoin and Ethereum scale and perform better. I don't want to drone on about "what is used to be like in crypto," but I will say that it was a magical time, and I owe a lot to early people for some amazing memories. It was also engaging, and I wasn't checking my email. Consequently, what I did not see on my travels, were email after email warning me of the fact that the credit card that was paying for Fight My Monster's Cassandra nodes had expired.... So what did the cloud operator do when I didn't respond within a few weeks (it wasn't many, but I forget the exact number). Well, they simply deleted all the nodes!!! It's hard to describe how I felt when I got back home to Palo Alto in California, and discovered what had happened. It was a mixture of shock and horror, a kind of hollow feeling in my stomach, adrenaline surges, and unfolding dread, as I worked as fast as I could to fix it, only to confirm that what had happened was irreversible. People who've been hacked in crypto will know the feeling. That was it. Fight My Monster was dead. Some part of the community even created a change dot org website pleading for it to be brought back online. Despite the loss of what it might have been worth, and the guilt, in some ways it was liberating, as I no longer had responsibility to maintain it, which probably would have done ad infinitum for the community. Plus it became a fond memory for a generation of kids and some adults that played it, before the game had a chance to grow too old and tired, and it allowed me to focus on what had become my obsession — decentralized compute networks. The lesson however was very clear to me. You can create decentralized compute infrastructure that can scale, and is fault tolerant, but if you have a single point of failure, anywhere, at any moment, it can be toast. That's why I'm always concerned when I hear about blockchains that depend on administrator keys held by the core devs, or blockchains that go down, or almost go down, when a huge chunk of their nodes are shutdown by a cloud running them, or because the cloud running them has technical problems. This problem of great fragility is vastly more prevalent than you might imagine in our industry. The learnings from the Fight My Monster disaster informed how the Internet Computer was designed. It's not only the number of nodes that matters, there are more important things. If nodes are anonymous, the number of independent parties running the bulk of them might be far less than you might think. Further, while large in number, they often turn out to be running on a small number of clouds— and even if they are owned and operated by a few different parties, the cloud might suddenly decide it doesn't want them anymore, and shut them down overnight, as Hetzner once did to a famous blockchain. What matters in practice is the number of independent parties running the nodes, and the physical and jurisdictional independence of the nodes. This simple insight is that was behind "deterministic decentralization," which will appear in new form inside the forthcoming "ICP cloud engines" extension of the Internet Computer. ICP cloud engines allow enterprises (typically) to create what is effectively an Internet Computer subnet whose configuration they themselves control, by choosing nodes to combine from a pool that acts as a marketplace. Currently there are more than 1.5k nodes waiting, and the number is growing. An engine provides a serverless cloud platform with remarkable properties. It is the ideal place for AI-built apps and services, because the has been stack designed to allow AI to generate advanced AIware software with ease, which AIware is secure and resilient. Apps are tamperproof (immune to infrastructure hacks), always-on (and thus incredibly resilient), optionally autonomous (which means there is no backdoor), digital assets native, and many other things. Engines also run e.g. the Open SaaS suite of AIware apps that can be used to run enterprises end-to-end. Cloud engines make installing and maintaining complex apps like Open CRM or Open Email as simple as installing and maintaining phone apps. Nodes are hot-swappable. That means you can change the compute replication factor, swap out the nodes for different ones, without interrupting what the cloud engine hosts, providing freedom from compute vendor lock-in, and, if you want, you can use a new class of nodes on hyperscalers, as well as traditional Internet Computer nodes, so it will function as a general compute marketplace. The cloud engine framework guides people to combine nodes that have independence, but provides flexibility that shared Internet Computer hosting doesn't have. For example, an enterprise in Germany might choose nodes that reside only in the EU, to meet GDPR requirements. However, the cloud engine console, which is autonomous, and served from the Internet Computer network, and updated by the NNS (the Network Nervous System, the world's most advanced DAO), guides users and warns them heavily if they combine nodes from the same node provider, or which are based in the same data center, say, or run by the same cloud operator. Thus, even with ICP cloud engines, the lessons of Fight My Monster live on. People will wonder why the framework pushes them so strongly to combine independent nodes, but there is a good reason, which is just as relevant for the wider blockchain community today, as it was in 2015. Looking forward to sharing cloud engines soon.

We're excited to help CodeRabbit reduce the number of bugs in the world! By using Exa search, the CodeRabbit review agent can ground code reviews with technical information better, 5 times faster.

Theorem co-founder @diagram_chaser reveals the one-line change that took verifying real-world HTTPS code from 4,000 millennia to seconds: "The project that I worked on in my PhD was verifying the code that runs HTTPS in browsers like Chrome and Firefox." "You plot this beautiful graph that is an exponential in the number of bits in the prime, where you're like, it takes a couple seconds on my tiny toy examples, and on real-world examples, it would take over 4,000 millennia." "The thing that it's actually checking is that you wrote the same thing in two different ways. This should be very fast. It's doing the operations in the wrong order." "So you tweak one line, and then it drops down to a couple seconds." @theoremlabs

New blog post about asymmetry of verification and "verifier's law": Asymmetry of verification–the idea that some tasks are much easier to verify than to solve–is becoming an important idea as we have RL that finally works generally. Great examples of asymmetry of verification are things like sudoku puzzles, writing the code for a website like instagram, and BrowseComp problems (takes ~100 websites to find the answer, but easy to verify once you have the answer). Other tasks have near-symmetry of verification, like summing two 900-digit numbers or some data processing scripts. Yet other tasks are much easier to propose feasible solutions for than to verify them (e.g., fact-checking a long essay or stating a new diet like "only eat bison"). An important thing to understand about asymmetry of verification is that you can improve the asymmetry by doing some work beforehand. For example, if you have the answer key to a math problem or if you have test cases for a Leetcode problem. This greatly increases the set of problems with desirable verification asymmetry. "Verifier's law" states that the ease of training AI to solve a task is proportional to how verifiable the task is. All tasks that are possible to solve and easy to verify will be solved by AI. The ability to train AI to solve a task is proportional to whether the task has the following properties: 1. Objective truth: everyone agrees what good solutions are 2. Fast to verify: any given solution can be verified in a few seconds 3. Scalable to verify: many solutions can be verified simultaneously 4. Low noise: verification is as tightly correlated to the solution quality as possible 5. Continuous reward: it’s easy to rank the goodness of many solutions for a single problem One obvious instantiation of verifier's law is the fact that most benchmarks proposed in AI are easy to verify and so far have been solved. Notice that virtually all popular benchmarks in the past ten years fit criteria #1-4#; benchmarks that don’t meet criteria #1-4# would struggle to become popular. Why is verifiability so important? The amount of learning in AI that occurs is maximized when the above criteria are satisfied; you can take a lot of gradient steps where each step has a lot of signal. Speed of iteration is critical—it’s the reason that progress in the digital world has been so much faster than progress in the physical world. AlphaEvolve from Google is one of the greatest examples of leveraging asymmetry of verification. It focuses on setups that fit all the above criteria, and has led to a number of advancements in mathematics and other fields. Different from what we've been doing in AI for the last two decades, it's a new paradigm in that all problems are optimized in a setting where the train set is equivalent to the test set. Asymmetry of verification is everywhere and it's exciting to consider a world of jagged intelligence where anything we can measure will be solved.

Grok Bot summary of SpaceX CFO Bret Johnsen’s new interview with Tony Kim at SpaceX Mission Control. Bret’s path - Came from semiconductors (Qualcomm, Broadcom; was CFO of MindSpeed) - Barely knew SpaceX / Elon when a board member called - Thought aerospace sounded “old and stodgy” vs 6-month chip cycles - Visited the rocket factory, called his wife on the drive home: we’re moving back to LA - Now sits in the same magical factory: Dragon in the rafters, ~700 launches run from Mission Control, ~40 buildings feeding the campus - Full circle: semiconductors again via AI / chips --- Taking on AI complexity (during the biggest IPO ever) - SpaceX DNA: run toward “impossible” challenges (first booster recovery felt like a rock concert) - xAI acquisition in January + X + Cursor + Colossus = complexity multi-X in ~1–2 years - AI is now the biggest business at SpaceX - Bought time-to-market into the biggest product cycle of their lifetime - SpaceX leadership DNA applied to that team - Already feel leading on infra cost and speed - Products moving: Grok 4.6, 4.7 soon, Grok Bot viral, Cursor strong - Every week matters; soon every day --- AI business: the money picture - ~$100B company ARR exiting this year (annualized on December) — AI is more than half / the majority - A year ago AI revenue was ~zero - Capacity: a bit over ~2 GW this year → guidance 5–10 GW by end of next year - Becoming one of the biggest hyperscalers, fast Why they move so fast (vertical integration) - Engineer the data centers themselves - Act as GC: buildings, power, cooling, networking (plus partners) - Used to regulators from rockets + telecom - Removes margin stacking - Talent transfer: launch-site / pressurized systems people onto data centers — Elon compared it to Yankees vs little league Hosting model / ROI - Monetize compute while catching up on software - Demand knocked: “what are you doing with all that Colossus compute?” - Short contracts: ~3 months + 3-month out (flexibility for their own future training needs) - Payback: dramatically less than one year on cutting-edge terrestrial compute - Hosting customers include tier ones like Anthropic, Google, and others - Same watts can later flip to their own software at higher $/watt — optionality Frontier models - Moving maybe faster than anyone on trajectory - ~5 serious players; need to be at frontier for recursive improvement - Coding solutions improving dramatically - Consumer vs enterprise tools will stratify - Grok Bot: growing faster than anything they’ve seen; “transformational” user feedback; enterprise coming; expects roll into Cursor AI safety - Follow Elon’s self-regulation / peer-review path (like other industries) - Logical, collaborative, including outside the US - Hopeful something lands quickly TAM framing (IPO) - ~$28T TAM quoted at IPO; AI was ~$25–26T of that (digital AI only) - Physical AI not even in that number — Bret thinks physical AI eventually bigger than digital - Launch / space TAM = hundreds of billions (almost rounding vs AI) - Connectivity (mobile + broadband) ~$2T — still small vs AI - Differentiation: infra + X content (~500–600M MAUs) + consumer + enterprise push (Cursor etc.) What to watch: Starship + terrestrial compute Bret’s prioritization cheat code: 1. Get Starship right 2. Get terrestrial compute right for the next ~2 years until orbital kicks in → Transformational financially Starship status - Catalyst for all three businesses: space, satellites, AI - Holy grail: full reusability → another ~10x cheaper $/kg to orbit - Opens moon boots / lunar base / bigger industry - Flight 14 later this month: production V3 Starlink broadband satellites — real revenue payload - One Starship V3 load ≈ ~20× what Falcon 9 flew for Starlink capability - Each V3 satellite ~1 terabit; ~50–60 per flight = enormous bandwidth per launch - A couple launches start filling V3 planes; first-half-next-year benefits - Eventually Starlink as % of global internet traffic - Connectivity demand from machines (robots, AVs, agents) will dwarf human brains - Same Starship unlocks next-gen direct-to-cell / direct-to-device - Same bus + more solar + radiator + compute chips = orbital compute constellation Reusability timeline - Flight 15 may attempt catch / bring back first + second stage - First stage already recovered and reflown - Flight 13 soft splash best yet; recovered ship like a whale, engineers studied heat shield - Reliable stage recovery this year; meaningful reflight next year - Once a few second stages reflight → orbital compute cost can match terrestrial - Then orbital cost curve goes down (tech / Moore’s law / aircraft-like amortization) while terrestrial gets more expensive (power, labor, regulation, cooling) - Clean free solar, no community impact - Industry agrees space compute is the future — debate is timing; others say ~decade; SpaceX flies first ones next year; meaningful compute in ~2028 Scale ambition - Both cost and speed matter; near-term availability/speed most important - On track toward ~100 GW/year by end of decade (~20% of entire US grid each year) - Elon talking terawatt-scale in the 2030s - Hosting proves staying power + virtuous cycle for industry - New NVIDIA platforms: just fly the next satellite generation — no whole new terrestrial campus cycle TerraFab (chips / memory) - Parallel to orbital ramp, not after - Not replacing NVIDIA — love the partnership; will buy all they can - Worry is deeper fab / memory supply chain - Memory industry ~30%/yr growth vs SpaceX wanting triple-digit — won’t work - Hard technical challenge; memory may be early focus - Want volume benefit by late decade (demo next year with Tesla development line → something in ’28 → meaningful ’29+) Viewer Q&A highlights **First bottleneck in the AI chain?** - Chip supply for next year - Power for infra players (utilities used to ~1% growth; turbines / blades constrained) - SpaceX standing up investment casting for turbine blades themselves - Orbital is the big unlock; squeeze every terrestrial watt meanwhile How to tell durable AI spend vs keep-up spend? - Bret: durability of revenue; diversified monetization of the same compute; host now, software later at higher $/watt; guided next-year monetization range ~$30–50/watt - Tony: stack layers have different costs, margins, contract duration; SpaceX spans many layers with <1-year payback on infra How soon a data center in space? - First orbital compute satellites end of next year - Meaningful compute in 2028 - Better phrase: “racks in space,” not giant buildings - Same V3-like bus + solar + radiator + compute chips - Already flown ~11,000 sats; launch + inter-sat + ground links already solved - ~full rack (~72 NVIDIA-class) concept per sat; ~5 MW compute per launch Tony’s investor wrap - Still early; trends accelerate into physical AI and beyond Earth - Connectivity / tokens increasingly for machines, not just humans --- Bottom line in one breath SpaceX is using rocket DNA — vertical integration, crazy talent, short-cycle iteration — to become a top AI infra + software player now, while Starship unlocks Starlink V3, direct-to-cell, and orbital “racks in space.” Watch two things: terrestrial compute scale (toward $100B ARR and multi-GW) and Starship reusability. Everything else is downstream of those.