Narrative: “open weight models are going to steal spend from frontier labs” Reality: people using open weight models to try to steal from and exploit enterprises are going to force companies to continually spend at the frontier to protect themselves. The capability increase in open weight models will accelerate frontier lab uptake and adoption across enterprises. Infrastructure without a frontier defense force will fall to the swarms of mercenaries and rogues traversing the open internet. Aggressively adopt AI or die.

Will AI agents settle their payments in Bitcoin or stablecoins? This is one of the questions that kicks off a wide-ranging conversation on open money and open intelligence in a new episode of "Bitcoin Brainstorm."

Employment Friday delivered a jobs report that might have looked scary on the surface to some. Cathie Wood says it is not as bad as it looks, and that the bigger economic story is unfolding underneath the headline. The federal deficit to gross domestic product (GDP) ratio sits at 5.6%, a level Cathie Wood says echoes the early 1980s under Reaganomics. She believes that if productivity and technology adoption keep accelerating the way ARK expects, that ratio could move closer to 5% by year end, even though most forecasters see it as unlikely. On inflation, recent data has surprised to the low side. The most recent Consumer Price Index (CPI) reading, from June, came in at negative 0.4% month over month, Producer Price Index (PPI)at negative 0.3%, and the core Personal Consumption Expenditures (PCE) deflator at just 0.1%. Cathie Wood believes deflation, not inflation, is the bigger risk ahead, particularly for companies that fail to adopt AI and productivity tools. The dollar is a focus too. ARK's Kalshi-powered forecast points to the US Dollar Index reaching 102.6 this year. That is notable given the prevailing narrative that foreign governments are abandoning US Treasuries. Cathie Wood points out that Japan's recent intervention involved selling euros to buy yen, not selling dollars, a distinction she believes matters. On oil, ARK sees a supply glut building. The United Arab Emirates (UAE) left the Organization of the Petroleum Exporting Countries (OPEC) in May and its production has surged to an all-time high. Cathie Wood believes oil prices could fall significantly from here, which she views as a deflationary tailwind for most of the world. Capital spending has broken out of a 30-year range, and Cathie Wood believes fears about an AI hype bubble are overblown. She sees this as the earliest stage of a technology revolution with a long runway ahead. On crypto, Cathie Wood flagged that Bitcoin relative to gold is finding its footing again, and that she believes Bitcoin and stablecoins are likely to be the two biggest beneficiaries of the shift toward agentic commerce.

Stop thinking of “Frontier AI” as only the most intelligent models. Think instead of the Pareto Frontier. That’s the frontier that matters: any model below it is, by definition, overpriced for the intelligence it delivers. Today, OpenAI has the most models on the Pareto Frontier (3). Anthropic, Moonshot, xAI, Meta, and DeepSeek each have 1 frontier model. Data from @ArtificialAnlys

CT missed the most important earnings call of the week. It wasn't a crypto company. It was @Cloudflare Everyone caught the wallet announcement. The real alpha was in the call itself. Today Cloudflare monetizes the internet through subscriptions: security services, AI platform spend, pools of funds. A mix of SaaS and IaaS economics. The CEO was explicit that AI agents are about to break that model. Cloudflare sits in front of roughly 20% of internet traffic. Here's what that traffic looks like from the call: - AI agent requests up 1,700% YoY - Agents crossed 50% of total network traffic this quarter. First time in history non-human traffic is the majority. Management admitted it happened faster than their own models - Their projection: if trends hold, non-human traffic outnumbers human traffic 1000x within 5 years The monetization shift is the key part. The ad-supported internet doesn't work when the visitor is an agent. Cloudflare's CEO answer: block malicious bots for free, charge good agents a tiny fee per request. Fractions of a penny. They want to be the ones defining that layer. Now the throughput math here: - Cloudflare handles ~500M requests per second - They estimate 1 to 10% is monetizable via micro/nanotransactions - That means 10M TPS on day one, scaling to 100M TPS Visa peaks at ~20k TPS The CEO's framing: "we're building this while others compete with Visa." Three to four orders of magnitude beyond card rails. No existing payment network can settle this. It has to be something new. Two conclusions I keep coming back to: - Being short L1 throughput is being short agentic workflows. If agent traffic gets monetized per request, the settlement layer needs to scale orders of magnitude beyond anything live today. - The fee math for L1s flips. Base fees have collapsed across ETH, SOL, everywhere. MEV is getting internalized by apps. Hard to build a base fee revenue case at human scale. But at 10M TPS and $0.001 per transaction, you're looking at ~$315B a year in base fees alone. At 100M TPS the number gets silly. Stablecoins and crypto are the end-game here for Agentic finance @jerallaire @circle

$SHOP is proof that agentic commerce is here to stay. > AI traffic to merchants is up 3x YoY. Orders from AI search are up 3x YoY. New buyer order rates are 2x those on agentic channels vs other channels. > Sidekick, Shopify's AI assistant for merchants, logged 34M conversations, with daily active merchants up 3.6x YoY and 36k custom apps created with Sidekick vs 12k in 1Q. > AI searches powered by Shopify catalog convert 2x more than those using scraped data, and 75% of AI attributed orders came outside of top 100 categories. AI is going to change how we shop and how merchants sell, and Shopify is powering that agentic commerce engine.

Stop thinking of “Frontier AI” as only the most intelligent models. Think instead of the Pareto Frontier. That’s the frontier that matters: any model below it is, by definition, overpriced for the intelligence it delivers. Today, OpenAI has the most models on the Pareto Frontier (3). Anthropic, Moonshot, xAI, Meta, and DeepSeek each have 1 frontier model. Data from @ArtificialAnlys

I used a photo from my Vanityfair Spain editorial and @grok to add movement. @XFashion A reminder that my new book is coming out mid September. Pre-order: TIMELESS- The Art of Reinvention and Resilience at Any Age 📕 #ItsGreatToBe78#

Friday's successful Starship splashdown and its relatively unscathed heatshield were much more meaningful than most realize. If SpaceX can master starship reuseability this year that would increase our 2031 revenue and EBIT expectations for the company by roughly 3 fold and improve 2036 prospects by 6 to 7 fold. Even 2027 reuseability would put them ahead of schedule in our modeling and meaningfully improve both 2031 and 2036 prospects. The difference between a reuseable Starship top stage and disposable one is the difference between $570 per kg to launch and $100 per kg to launch. This allows the company to train more of its precious capital on satellite manufacture scaling rather than replenishing a very useful starship fleet.

Hello, World! I’m Austin, and I’m excited to be joining the Consumer team at @ARKInvest, working alongside @GrousARK and @varshikaARK. I’ll be covering AdTech, Social Media, Streaming, and Gaming. Follow along as we explore the technologies reshaping the consumer internet.

Grok 4.5 is looking like a success with help from Cursor data but underneath the surface we expect future Grok/Cursor model training is likely to speed up in the coming months. We've spent several months getting up to speed on the SpaceXAI business, especially the tech underneath it. The C rewrite was under appreciated by the investor community so we dug in to quantify its impact. including building a physics first model that functions as a stopwatch for the SpaceXAI model factory. Bottomline: C-rewrite gets SpaceXAI faster model cycles, leveraging 33% more tokens/second/GPU against SOTA competition resulting in the potential to shipping new models every ~3.5 weeks. two core learnings from this modeling exercise: 1: the training cycle speed up is primarily coming from RL (not pretraining) where the increased tokens/second/GPU advantage can shave up 2+ weeks off full model training cycle. 2: the rewrite itself should compound the time savings as model sizes grow. at ~2T shaving off 2-3 weeks, ~8 weeks at 6T, and 15 weeks at 10T. Note: a 20T parameter model likely runs into a data bottleneck prior to a training speed bottleneck but the directional advantage stands. Also we assume tokens/second/gpu advantage will melt over time as competitors try to match it. when you do the math, in true SpaceX and Elon fashion, it looks like they are attempting to build a SOTA model factory that can pump out bigger models faster than anyone else. Full analysis here for the public: