Jon Rotbard

Today we're releasing Agora: the first ever pretraining stack that allows non-collocated consumer GPUs to be competitive with centralized clusters Agora is 15x faster than Megatron-LM in this setting and is only 1.5x less efficient in terms of tokens per unit compute than TorchTitan on H100s, despite running on devices that have no NVLink or InfiniBand support.

@IamSean90 Need an analysis from @nikitabier on what % is Chinese bots

You legit can just blame data centers for anything & get infinite likes on social media, no matter how stupid & made up the claim is

@JasonYanowitz @sethginns For which tokens? All of them, 20 of them, etc.

Calling it: Beginning of a new bull market Small signs here and there of life. Will become more obvious as the year progresses. Giddy up.

@AshleyLLouise @garrytan Sure but what's also net negative for communities is being squashed by China because they zoom ahead of us, because we put a moratorium on data centers and fall behind on compute

@garrytan As of right now most research shows data centers are largely net-negative for local communities and are causing real resource strains Rising electricity costs, water, noise and pollution … Data centers are necessary but these are all very real issues that need to be solved

Sanders and AOC introduced a bill to pause ALL AI data center construction. 300+ local bills filed. Half of planned 2026 data centers facing delays or cancellation. Each one brings billions to local economies. The people who say they want American jobs are trying to block the biggest job creation engine since the interstate highway system.

@dadiomov Agents are going to be doing a lot more than e-commerce shopping. Specifically for that, agree existing cards are fine

I don't understand the premise that "agents must use stablecoins." Why can't an agent remember the 16 digits of a credit card? Sorry maybe I'm a payments n00b. Stablecoins have a lot of great use cases, but ecommerce shopping is pretty well optimized already for cards.

It depends on the project but I broadly disagree here I think any non-speculative tokenholders are also expecting to accrue future profits, and also have trust in management. Obviously that trust on both fronts is often misplaced, hence the dark view - but while buybacks prove some goodwill, if you're buying most tokens purely for buybacks and not future turbogrowth, I don't think that's a great idea. And to Guy's point, that turbogrowth could be much more easily achieved if they redirected buybacks into growth. At EOD these are startups and don't defy the laws of physics or follow dramatically different rules to those in web2, despite token-related differences Re: tokens --> faster growth, I don't think this is true overly often anymore, look over at the AI party and at the struggles many DeFi companies are going through (and often flattening traction post TGE) despite token incentives The sentiment though that we need more mechanisms for trust in token projects' good faith in both growing and accruing value to the token is one I strongly share, though, and will go to converging these two views

Disagree with a16z completely. This is such a VC "everything will just workout" mindset and completely misses the structural flaws. x.com/jdorman81/stat… The biggest (and really only) difference between stocks and tokens is that equity investors trust equity issuers (partially for legal reasons, partially due to how long equities have been around). Now, that is often misguided, as management can still make terrible decisions with capital allocation (pay employees too much, make bad acquisitions, bad R&D expenses, etc)... but for the most part, paying dividends or doing buybacks is not necessary right away for equity investors because they either blindly trust management, or because they know eventually the profits of the entity will go to shareholders. Thus, dividends/buybacks can wait. But token issuers grow much faster than traditional companies, largely because of the token incentives that help these entities grow. Therefore, they should be paid back quicker. And buybacks are the ONLY viable path towards paying back tokenholders. The only thing that would change my mind would be if token issuers did a better job of communicating their visions, and built trust with their token holders. In doing so, then token holders might be willing to wait longer to be repaid.

@richardchen39 Taxes and the friction and/or inability to move offchain

What's the best explanation why there's tens of billions of dollars in DeFi earning 4-5% APY when the risks are way higher than treasuries?

@bwein_ @SynchronicityHQ would love to test drive!

Modern markets are fundamentally extractive We've spent three years rewiring them from scratch One core feature is everyone deploying their own bots without code or infra Here's a first look @SynchronicityHQ We're looking for strategists to test drive. Reply for early access

@guywuolletjr The only one this makes any sense for is Hyperliquid, where they make too much money to know what to do with + are late stage atp

it's insane to me that buy and burn has become the default for profitable crypto protocols why would a high growth startup would ever take its profits and distribute them to shareholders instead of re-investing for future growth, or at least holding it for runway

Rounds are a social construct

@0xJeff This may be true today but would be surprised if it holds. M2M is way too big of an opportunity for Stripe to lay down and let someone else take agreed though that Stripe's advantage is more so in M2V, so they def start there

x402 vs MPP x402 is designed for agents (M2M) — it's designed to be open, permissionless, self-propagating, great for experimentations MPP is designed for vendors (M2V) — it has built-in enterprise-ready features (session-based payments, shared payment tokens), budgeting policy, full stack that automates tax/accounting/finance MPP likely grow faster than x402 in short-medium term thanks to Stripe's distribution BUT long term x402 might just win thanks to its "open" nature

@KyleSamani It morphs into a game of chance when you systematically ban all the sharps, which is the strongest argument for the difference between sports PMs and sportsbetting I've seen

It is important to delineate between games of skill and games of chance It is universally agreed that trading in public markets - commodities, equities, fx, etc - is a game of skill. Many firms have structural alpha. You can play the game in an EV positive way It is also universally recognized that pulling the lever on a slot machine, craps, blackjack, etc is not a game of skill. It is strictly a game of chance, and it is known to be EV negative Wagering on sports is far more like trading public markets than gambling at the casino. There is immense and heterogenous publicly available data that can be analyzed and used to develop systematic trading strategies Wagering on sports is a game of skill, not a game of chance

There's a lot of novel questions about what will accrue value, have moats, or be attractive if the target you're selling to is an agent vs. a human, which I (maybe weirdly) find fun to debate. There's also a lot of paths it can go, vs. other fields where the winning vectors are more straightforward There's also a credible chance (nowhere near a given) that agents prefer crypto rails en masse, which would make this one of a select number of areas crypto could break out into the mainstream in the next couple of years - main reason I'm in the space is to find those. Though that's more idealism vs intellectual Agents generally are also just a fast moving and exciting space. I found using Claude Code for the first time more mindblowing than anything else I'd seen in tech, incl. seeing ChatGPT come out. Forming an essential part of the stack, even if it's invisible to everyone but the agent, is cool

@jonrotbard Why? Maybe I don’t get it. Like giving api keys to your agent is interesting and probably useful but why is this so intellectually stimulating? Idk that 99% of people will even know that we created a standard for internet payments in 5 years assuming it all works

The hardest part of being in crypto is pretending to care about agentic payments

Not to mention the bottom now essentially has a Buzzfeed article. Cancelled sub "Want to know the top 3 things people get wrong when they're in your exact position?"

I've ran the same prompt for deep research through GPT 5.4, Opus 5.6 and Gemini Deep Research (I assume Gemini 3.0) most of them ran for ~30 mins GPT 5.4 is *REALLY* annoying! it's "reflexively contrarian", it prioritizes showing you what's wrong with your thinking, NOT actually helping you solve the problem ME: my house is on fire! GPT 5.4: While it's true that combustion is occurring, it's important to note that not all of your house is on fire. The garage, for instance, appears structurally intact. (this is a pattern with it, btw, many such examples) I'm not sure if this is because these are health related questions, but this has been an incredibly annoying model for this specific task

There's a lot of really exciting developments happening in decentralized AI training this year. Here's my take on why decentralized training is moving from "impossible" to "investable". 🧵👇

@PredictParity Now show performance by badge 😈

Overview of the prediction market badge economy

@spermracing Can't beat US microplastic-powered

Introducing the Sperm Racing World Cup. 128 Countries. And a $100k grand prize. Apply below to represent your country:

NEWS: Kraken DeFi Earn has surpassed $100 million in total deposits. It’s never been easier to earn on idle assets. Powered by 3 Veda vaults: Balanced + Boosted → curated by @chaoslabs Advanced → curated by @sentoraHQ Available on @krakenfx, @krakenpro, and @krak

@dotslashnaz @techdollarhq @grok they're AIs

@techdollarhq @grok who are the people behind techdollar?

Introducing: Techdollar Until now deep tech equity has been sitting locked up. We are here to change that. Borrow against your equity in Anthropic, SpaceX, OpenAI and more without selling a single share. Techdollar.com