WSJ saw Anthropic’s draft IPO papers: $16.3B revenue in 6 months, $80.7B loss on paper, and about $450B promised for compute > Anthropic spent nearly $9B on compute and infrastructure in the first half, more than its $7.3B total for 2025, and its new draft prospectus shows the bill climbing toward $750B. > The 3 big clouds account for $252B of fixed payments over the next 7 to 10 years: $111B to Google, $110B to Amazon Web Services and $31B to Microsoft, all due whether or not Anthropic uses the capacity. > Chips form the next layer. Anthropic has more than $160B in Google chip leases through Broadcom and up to $86B for Nvidia GPUs rented from Nscale and Lambda, with those contracts ending in 2033 and 2034. Another $150B in deals is still under negotiation. > Money isn’t the only limit. Anthropic says Amazon, Google and Microsoft wouldn’t supply all the capacity it wanted, and they now treat it as a rival. It is also leaning on newer data center operators, which it warns have short histories and possibly less reliable sites. > One deal goes unnamed: the September draft’s up-to-$84.5B xAI capacity agreement, which runs through 2029 and can largely be cancelled on 90 days’ notice. It may sit inside the roughly $112B of the mostly noncancelable $450B cloud total that isn’t itemized, alongside AMD capacity expected to top $20B, or it may have been dropped. > In its risk factors, Anthropic says it is liable for some damage its models cause, such as data theft or security breaches, and that a catastrophic AI incident could exceed any insurance it could buy. > Revenue is concentrated too: 2 unnamed customers delivered about 20% of first-half sales, and Anthropic expects a small group of customers to keep supplying a large share for the foreseeable future. > It also lists increasingly organized opposition to data centers as a business risk and says its high profile could bring extra scrutiny to its projects.

HOLY WTF. Reuters is reporting that a Chinese-speaking hacker used AI agents to compromise websites belonging to more than 100 companies, including Deere, SharkNinja and Cummins, after attempting intrusions against over 1,200 sites. The estimated cost: roughly $5,000 in AI tokens, or $50 per successful breach. Successful intrusions took around 23 hours on average. The attackers used the open-source Cairn agent framework, apparently combining DeepSeek models with a Claude Code harness. One campaign gave its agents instructions in fewer than 100 Chinese characters. NorForge researcher Snorre Fagerland: 'It's a new world now.'

We're expanding access to GPT-Rosalind for academic research. The new Rosalind Academic Plan helps scientific labs bring AI into the work of understanding biology, exploring ideas, and deciding what to test next—with sponsored access for eligible labs at U.S. universities. Academic research starts with questions whose impact may take years to impact human health.

"DepthART: Scaling Foundation Monocular Depth to Tiny Models" TLDR: brings foundation-model mono depth estimation to tiny networks through bias-resistant data sampling and camera-conditioned fine-tuning; robust relative+metric depth with realtime inference on small hardware.

Sometimes I feel we're so caught up in the next market move that we forget how much is actually changing in crypto. New technology, new financial use cases, and things that could make a real difference over time. Take what's happening with @SuiNetwork recently, for example. 🔹Later this month, Hashi is expected to launch with $500M in capital commitments from 20+ partners. It would allow institutions to borrow against their Bitcoin without moving it off the Bitcoin network. That's an interesting use case for BTC holders. 🔹@EmanAbio also mentioned that post-quantum accounts will go live on Sui devnet next Monday. Basically, work is already underway to prepare user accounts for potential security threats from quantum computers. 🔹Meanwhile, CCTP V2 is now live, allowing native USDC to move between SUI and 29 other supported blockchains. A useful step for anyone moving stablecoins between networks. Three different developments, three different areas of crypto, and all happening around the same ecosystem. What interests me most is how crypto is slowly moving beyond just trading, with more focus on real financial use cases, security, and moving money between networks. There's still a big gap between building something and getting people to use it, and that's where things will get interesting.

Pi Network clarified its stablecoin strategy. OUSD will not replace Pi. Here is the split: ✸ Pi | native ecosystem utility and identity ✸ OUSD | predictable pricing and merchant settlement Most stablecoin issuers keep reserve yields. OUSD shares that revenue with partners. Stablecoins solve price volatility. Pi stays the primary currency.

Finally, rewrote the lighting by importing @takram's atmosphere library and adding lightroom-style settings, support both photorealistic and abstract modes. Also did some performance improvements :) Now I can fully focus on fixing the roads rendering. Live demo: jeantimex.github.io/tokyo/ (not recommended on mobile browser :))

Today we’re previewing our work to build an open standard @Meta (@muse and Meta Business Agent) and Sierra are developing along with industry partners at Genesys, NiCE, Decagon, Rocket, Shopify, Stripe, and Walmart that defines how personal agents interact with businesses. What motivates this work is making things even better. More frictionless for people using Muse, and better for the businesses serving our shared customers. Excited to build this together.

Introducing Empryo for iPhone, iPad and Apple Watch. Your coding agent keeps working on your computer, and you steer it from your pocket: on the same Wi-Fi, over Tailscale or from anywhere, end-to-end encrypted. Also in Empryo 3.9: → The engine outlives the window: quit the desktop app, the turn keeps going → Worktrees in tabs: every tab on its own branch, side by side → Live voice: words appear as you speak, hands-free on desktop, terminal and phone → /tier ultrafast for GPT-6 Astra, and Cursor and Devin with their own prices → Windows: open WSL repos and run the agent inside Linux → Terminal lessons on empryo.com, one feature at a time → Faster everywhere, and an immune system: agents hunt bugs, and every fix ships a check so the bug can't come back

There are two fundamental misapprehensions behind the belief that independent, general-purpose agents will broadly disrupt the eCommerce landscape: 1. That merchants would prefer organic agent recommendations and transaction fees to auction-based advertising because branded search advertising is a zero-value "tax" or "toll." This isn't the case; branded search advertising has been observed to deliver incremental value. In “Competition and Crowd-Out for Brand Keywords in Sponsored Search," Simonov et al. found that own-brand ads on Bing increased traffic even without competing ads, while competitors captured 18–42% of clicks when brands facing competition did not advertise. Even in the specific advertising context of Amazon, the available research supports incrementality. In “Data-driven budget allocation of retail media by ad product, funnel metric, and brand size,” Koen Pauwels (friend of the podcast!) and coauthors estimate positive contributions from Sponsored Products to awareness, consideration, and revenue across more than 122,000 brand/category combinations. Characterizing branded search advertising as a tax doesn't really address the core economic issue, which is whether it can deliver incremental value. Obviously, like all advertising, branded search can be wasteful. But the research supports the idea that branded search can generate incremental returns. So the question is why a retailer would prefer to integrate with an agent and pay a fixed, percentage fee on each transaction rather than bid for distribution according to what reaching a particular customer is worth to its business. I don't think it would. Even if a retailer can accommodate an agent's fixed transaction fee (OpenAI's was 4% on Instant Checkout), wouldn't it sometimes prefer to pay more for the opportunity to reach a relevant customer? A bid allows a retailer to express its own derived willingness to pay. So if agents become important sources of demand, why wouldn't merchants want to compete for exposure on those surfaces, too? 2. That consumers will demand a single general-purpose agent that handles all of their needs, crowding out platform-owned agents while also retaining all of the functionality that the platforms provide. Suppose the mono-agent succeeds. Amazon loses its consumer traffic and becomes middleware that fulfills agentic purchases. Why assume that Amazon continues operating at its current scale if the agent has subsumed the customer relationship and much of the revenue opportunity ($69BN in advertising revenue in 2025)? Fulfillment, customer service, returns, fraud prevention: who manages these if the revenue opportunity from the consumer relationship is undercut? The mono-agent thesis assumes that an agent can dismantle the platform's business model yet benefit from the infrastructure upon which those businesses were built. But the more totally any agent displaces the platform, the more of the platform it needs to replicate. I think the most concrete and immediate example is reviews. If people stop buying on retail platforms, where are the reviews submitted? In the agent? That defeats the purpose. But if product reviews evaporate, how does the agent rank products? This is obviously why Amazon has been reluctant to integrate with third-party agents. And no, Walmart's cooperation with various agents isn't instructive here; the two companies operate against fundamentally different incentives. Walmart has a much smaller eCommerce footprint and advertising business: it has more demand to acquire and less existing demand to defend. And it also operates its own agent.

Last night I gave a guest lecture at Sogang University's executive program on AX transformation and digital assets: AI and Stablecoins, Financial Infrastructure for the Agent Economy. The argument I brought to a room of executives was simple. Agents are starting to transact on our behalf, and they need money that works the way they do, settling the moment it moves and following rules written in code. Stablecoins are that money, and the infrastructure underneath them is being built now, much of it in Asia. A closed classroom turned out to be the right room for it, because the questions get to be about your own business rather than the industry. The public version comes today: I take the same argument to the Blockfesta keynote stage, in front of Korea's financial institutions. Thank you to Sogang and the S-Meta cohort for having me. [국어 아래]