Muse Spark 1.3 is rolling out today with frontier performance almost too cheap to meter. This is the biggest jump we've made so far on coding and agentic work. Try it in Muse Code and our API. Next up 🍉 and Muse Spark open weights releases coming soon.

Today marks the start of the Seaglider era - WE FLEW!! Our 15,000 lb, human-crewed Viceroy Seaglider prototype took flight today. Years of hard work culminating in this moment. Flight was smooth and exactly as planned. So proud of this team. More details to come!

Dear short shorts, A wise man who knows Elon well once told me: 1. Never short a pipe dream. 2. Never bet against Elon Musk. The obvious trade: short $SPCX into the first lockup expiration. I bet on that pipe dream 9 years ago. I own several million shares and will sell zero. This week, you may learn both lessons at once. I’d gladly lock up my shares for 9 more years, and I’m far from alone. With such a small float, it hasn’t taken much selling pressure to move the stock down. But obvious trades don’t stay easy for long. If the company keeps executing and far fewer insiders sell than expected, that dynamic can reverse very quickly. You’re betting against a founder who gets immense pleasure from proving doubters wrong and has a history of turning pipe dreams into reality. Good luck. You may need it sooner than you think.

The market is being stupid on META and AMZN again. Buy both, and sleep well. Read this for confirmation… META should not be down 8% after hours. They grew revenue 28% at enormous scale. The headline EPS miss included $3.6B of legal and severance charges. On a normalized basis, EPS was approximately $7.35, well above consensus. The market keeps treating higher capex as a negative. It has this completely backwards. AI demand is supply constrained, so productive capex is future revenue capacity. META will increase earnings drastically and use it internally, or lease excess supply and make a large arbitrage either short term or long term. In all cases, it’s a massive win. In Amazon’s case it is spending today to unlock years of high margin (40%+) AWS revenue tomorrow. Capex ROI is likely around 30%. AMZN has a clean setup going into earnings tomorrow: AWS grew 28% last quarter, its fastest growth in 15 quarters. Amazon has a custom chip business (to bypass NVDA and increase margins) that passed $20B ARR and is growing triple digits. AWS generated nearly 60% of Amazon’s operating income last quarter. This percentage will continue to increase, even if we get an increase in prime membership pricing which happens every 4 years, and has not happened although we are in year 4 and due for an increase if you believe in patterns. Perhaps tomorrow? There are many data points that support cloud AI demand is accelerating. My bet is AWS reaccelerates sharply to 40%+ the second half of 26 as new capacity comes online. 45–50% in Q4 is very possible if the capacity ramp hits, which means earnings estimates are far too low. AMZN average EPS for 2027 is $10.09, but I would not be surprised if they do $14+ EPS. AWS is worth more than the current market cap of AMZN. Full disclosure I have meaningful positions in both companies and believe both META and AMZN will at least double to 1,000+ and $450+ respectively by 2028.

The market is being stupid on META and AMZN again. Buy both, and sleep well. Read this for confirmation… META should not be down 8% after hours. They grew revenue 28% at enormous scale. The headline EPS miss included $3.6B of legal and severance charges. On a normalized basis, EPS was approximately $7.35, well above consensus. The market keeps treating higher capex as a negative. It has this completely backwards. AI demand is supply constrained, so productive capex is future revenue capacity. META will increase earnings drastically and use it internally, or lease excess supply and make a large arbitrage either short term or long term. In all cases, it’s a massive win. In Amazon’s case it is spending today to unlock years of high margin (40%+) AWS revenue tomorrow. Capex ROI is likely around 30%. AMZN has a clean setup going into earnings tomorrow: AWS grew 28% last quarter, its fastest growth in 15 quarters. Amazon has a custom chip business (to bypass NVDA and increase margins) that passed $20B ARR and is growing triple digits. AWS generated nearly 60% of Amazon’s operating income last quarter. This percentage will continue to increase, even if we get an increase in prime membership pricing which happens every 4 years, and has not happened although we are in year 4 and due for an increase if you believe in patterns. Perhaps tomorrow? There are many data points that support cloud AI demand is accelerating. My bet is AWS reaccelerates sharply to 40%+ the second half of 26 as new capacity comes online. 45–50% in Q4 is very possible if the capacity ramp hits, which means earnings estimates are far too low. AMZN average EPS for 2027 is $10.09, but I would not be surprised if they do $14+ EPS. AWS is worth more than the current market cap of AMZN. Full disclosure I have meaningful positions in both companies and believe both META and AMZN will at least double to 1,000+ and $450+ respectively by 2028.

Whop Co-Founder & CEO @cultured’s advice to young people: - Experiment as much as possible - Don't feel like you're wasting time "The only real risk is not doing anything."

Today, we're batch shipping 80% of the mobile code we’ve written this past quarter. First, a breakdown of the other 20%: across 42 releases, our 5-person squad launched the first perpetual futures in the US and trading experiences (explore overhaul, 15-minute charts, live chat, World-Cup hub, inner-circle, etc.) that have led Kalshi to top the App Store rankings for most of the past month. But all of this was just a fraction of the team’s firepower. At night, we were quietly working on a big project: “V2”, a complete rewrite of our app and the biggest technical leap I’ve seen since I joined the company 6 years ago. 4000 commits and 1 million lines later, the numbers are in and they justify the crazy sprint we just went through: - CPU usage: 369% peak -> 15% peak - Memory use: 350M -> 70MB - Install size: 315MB -> 95MB (for reference: RH @ 934MB, Coinbase @ 310MB) - External dependencies: 94 -> 11 - Test coverage: 17% -> 100% When you get the update, Kalshi should immediately feel significantly faster, smoother, sleeker, and more responsive. For our product velocity, V2 represents a massive shift as well: we now have a single codebase from which the team can push changes to our backend, iOS, and Android builds. The only code that runs on the client is the code that the team wants to run on the client. We have end-to-end type safety, CI with version-tagged deploys, and client<->server integration tests. We’d already gotten in the habit of daily releases, but with our latest tooling, it’s now a breeze. We’re keeping the team lean, but if you are incredibly talented, work hard, and want to build the next generation financial market,. DM me or apply on our careers website.

Kalshi CEO @mansourtarek_ thinks compute could become the largest commodity on the planet, creating the largest derivatives market alongside it. He says companies already spend roughly $1 trillion a year on compute, with that figure potentially growing 10x by 2030. As prices become more volatile and compute becomes a larger corporate expense, producers and buyers will increasingly want to hedge their exposure. “Historically, if that market has any degree of success, it ends up being at least 10 to 15 times the underlying market.”

I predicted the same outcome on Kalshi with 15 mins left when Argentina was down 2-0. Also went long Argentina to win the game in regulation at 4% odds and to win the World Cup at 1.4% odds (now 18%). Outliers are fun.

@AppenResearch independently evaluated @subquadratic's SSA kernel - a learned sparse attention mechanism designed to reduce the quadratic scaling limitations of full attention. Results at 1M-token context lengths: - 56.2× wall clock speedup vs. FA2 - 62.8× FLOP reduction (validated via torch.profiler, <4% variance from theoretical) - 95.6% average score across RULER tasks at 128K - 86.2% average score on the hardest MRCR 8-needle bucket (512K–1M contexts) - 81.8% SWE-Bench Verified resolved rate Full report: