Compute is a fundamental resource for the AI industry, powering training, post-training, inference and more in the AI pipeline. Yet it's volatility, with H100 spot going from $8 to a range between $2 - $4 in just two years, has hardly been hedgeable. This is changing now. GPU-hours are becoming benchmarked, financialized, and traded, and this article explores the rise of the compute capital markets: who are the players, what's required for compute standardization and who will eventually dominate the market.

finding the right API is as easy as that with @orthogonal_sh

think about your Muse / Instinct having access to stablecoins, which can > be used directly to pay for services / products: see Muse's integration with Instacart, Expedia etc. > used to earn yields on-chain

it's all just a dark forest innit

AI-enabled drug discovery is seeing good progress, with Moderna's AI-driven cancer vaccine being one that caught people's attention recently - equities in this sector have seen impressive growth $SDGR $RXRX $MRNA ultimately we will still have to see whether this improved efficiency in drug discovery can be brought to the last mile: regulatory approval if the reduced OpEx cost is transferred to patients, we could see increased drug sales with higher patient accessibility. TAM could increase with discovery of new drugs it will be interesting to monitor the balance sheets of companies within the AI drug discovery and drug distribution sector

x402 is the best when trying out new API services or completing one-off tasks via API endpoints! Monad API Hub launched a couple of days ago, it has a pretty great set of services that can be used in pretty creative ways to automate a lot of your work! The video below is a quick intro. Monad API hub: app.monad.xyz/agents/api-hub 1.25x speed recommended.

I wrote an essay about why the next frontier for AI is the world outside the data center Nature is a system more complex than anything we've ever created AI can learn how the system fits together, so we can finally understand how to shape it How? naturalgeneralintelligence.a…

at the end of the day it's about making it simple for the user, nobody wants to be talking to 5 agents with siloed context on toilet paper preferences, flight seat preferences (folks who prefer middle seats r yall ok) etc. more and more tasks will be delegated to agents, the interesting thing will be to see whether the agents choose to go direct to providers / route through aggregators - we will likely see a shift in economics where discounts are used as a way to draw more agent demand

interesting, keen to see what catalysts will bump that % up

instinct acted on instinct

We sat down with Rich Jaycobs, the former president of Cantor Futures Exchange & legendary creator of 7 exchanges & 6 clearing houses, and 1/5 of all approved DCMs since 2000. Here's what it really takes to build a compute market. sfcompute.com/news/commodity…

It's time to focus on the next compute bottleneck: credit

I don't think the tokenization story is well understood - so let's break it down Basically - the US Government is in vast, unprecedented amount of debt. And its long term debt is selling off a lot (40%+ over 5 years) - while Gold is up 140%. Gold has crossed US treasuries as the #1 asset held by other Central Banks. The buyers for US debt - China, Japan and Europe are stepping away for different reasons. Geopolitical conflict. Currency related. Political difference. The government is afraid of running out of buyers for its debt. And it's worried about the liquidity of markets. Historically, this concern is much worse when the Fed is hiking and there's high inflation. As typically the response to unstable debt markets is easing (buying debt). Therefore, the US government is very incentivized to allow Stablecoins. Stablecoins hoover up US debt. Tether has different balance sheet behavior than banks and works closely with the US government. Stablecoins used to be assumed to be criminal operations and were prosecuted by the USG. Now they're welcomed to Washington DC. When you are worried about your reserve currency status, you also want to ensure the dollar is accepted in many places. So stablecoins not only serve as a buyer of US debt, but a promoter of the US dollar globally. Enter CBDCs. Europeans see the US government promoting US dollar stablecoins as a matter of foreign policy. Visa and Mastercard going heavy in the space. And see Tether freezing Balances along with the DOJ and say, "We cannot have this. We don't want Trump to be able to use the US dollar in negotiations with us. Therefore we have to digitize our currency" Europe and the UK also have fiscal and political problems. So the perverse incentive to digitize their currencies is not just to ensure 'monetary sovereignty' but also to potentially implement wealth taxes, or balance based transaction taxes. This is also the flip side of wide adoption of US dollar stables. We don't call it a CDBC, but its basically an extension of the government. Balances are frequently frozen. Other policies could be implemented under the left. So rising global debt -> US support of stables -> pressure on Europe to do CBDCs to respond. The US is then incentivized to grow the stablecoin market as fast as possible. Stablecoin's usage is primarily driven by speculation. Holding it as collateral for perpetual swaps. Keeping it on exchanges. The problem: crypto doesn't have lots of good assets to trade. Bitcoin has been very volatile. Most altcoins collapsed. This slowed the growth of the Stablecoin market, which has basically flatlined year to date. Scott Bessent wanted us to be going at 40-50% CAGR not 0. The question then, is how to create appealing speculative markets. The answer: on chain stocks, and prediction markets. Big picture, if you want a lot of stablecoin balances. You need to have good things to trade. Over the past 6 months the 2x leveraged Micron ETF traded more than Bitcoin in dollar terms. Trade XYZ launched commodities, and stocks and commands a large % of volume on hyperliquid. Prediction markets are growing fast. The losses incurred by retail investors in these markets are substantial. They are allowed for the same reason Casinos are allowed on Native American reservations. Necessary evil due to funding pressure and the geopolitical factors I described above. This is, at a high level, why we are talking about crypto in the middle of a war with Iran. And why its' consuming the executive department's attention. The US Dollar is ultimately a matter of national security. This is true regardless of whether Democrats or Republicans are in power. Both have debt and spending addictions. Both are reliant on the US Dollar. But now zoom out: what do you have? You have a bunch of governments adopting digital ledger technology in the middle of a war. Freezing peoples balances. People don't want their balances frozen. They don't want it to be subjected to a wealth tax. Enter Z Cash. Over the past 5 years, ZK technology has improved substantially allowing fast, private transactions. Orchard and Halo since 2022 evolved from something Niche to something usable. The EU tried to ban Monero. Exchanges delisted it. Its price increased. Z Cash is open source. Vitalik is also prioritizing adoption of ZK technology on ETH. You can move ETH relatively privately on Railgun already - but real private transactions are on the roadmap. ZK proofs can be accelerated by GPUs. Due to the AI boom, the number of people with access to GPUs and agents capable of accelerating proofs has skyrocketed. Not only can this technology facilitate private transactions but also proofs of reserves that allow trustless accounting for portfolios. This tech is commonly used at crypto exchanges to prove reserves, and algo stables such as Ethena. Thus - it's not just that we are set to own a bunch of shielded Z cash, as a base asset. It's that you can swap anything on ETH privately. Including stables. Relatively soon. And you can do so for provable reserves. Privacy is incredibly important for large institutions doing financial transactions. They're not concerned with the tax authorities, but rather front running. Slippage. And people hunting their positions or slow moving exits. Many big institutions own multiple days of volume in equities that could drop the price 30%+ if the market sniffed out they were selling. And intervene in FX markets, routinely interacting with banks fined billions of dollars for illicitly trading ahead of their flows. So a meaningful increase in tokenized trading volume inherently will require privacy. As real institutional counterparties require it. Retail is a great market and is less privacy sensitive, allowing initial growth. But for the big players to enter -- you need that. It's not just trading desks that care about privacy. You cannot run corporate treasury functions publicly. People could sniff out that you're doing M&A, or figure out what you're doing -- giving up competitive positioning unnecessarily. At the same time blockchains can offer corporations major cost savings for finance functions. It's estimated - for example, that a properly implemented CBDC could save companies in Germany billions of dollars a year in corporate finance banking fees. There are three basic approaches to Privacy. Z cash style. Private, permissionless neutral chain. Canton style: basically segregated databases that interact with each other on chain as little as possible with a permissioned validator set. And private blockchains - which simply deal with the issue by not having a public ledger or bolted on cryptocurrency people are tracking The argument for a permissionless neutral chain (aka a cryptocurrency) winning is that 1. there's no real reason to trust a banking consortium (i.e. what Canton does) 2. governments hate each other and are only doing CBDCs out of geopolitical pressure in a realpolitik environment. we wouldn't be here if Trump wasn't antagonizing Europe and Canada wasn't talking about joining the EU 3. the tech exists, so why would you want a counterparty in between things if you don't have to The world in which crypto loses would be that governments come to an accord about how to do this. China and the US maybe resolve differences. Nationalism subsides. And people say, "You know what, the externalities of all this gambling and absurd crypto shit are not really worth it - we should just have a consortium of nations and corporations for a global CBDC" There is another world where crypto loses. The debt problems go away bc we enter an age of productivity and abundance. In my opinion, the reason AI people hate crypto so much natively is that crypto is a bet AGI isn't the economic Hail Mary it's marketed as. regardless, we are in neither of those worlds right now. But crypto hasn't done great either. Why? The big problem with crypto, from a valuation standpoint, is that it has never been clear how you pay for all the validators or miners without a block subsidy. And the only coherent way that happens is that on-chain volume, swaps, and trading 5-10xes. The way that you get there is that high quality assets get tokenized and traded. But probably less understood is that an entire swathe of new assets hit the blockchain, and defi functions like borrowing and looping create carry trade and FX trading opportunities. So rather than pendle looping Ethena, you have looping with RWAs. This already exists to some extent in niche markets like Brazilian credit card debt (looped 28%+ APY) but is relatively tiny. Just on this example, you might immediately say "that has huge FX risk", and you'd be right. Which brings a natural demand for FX hedging. Which will hit after you get internationalized RWAs on chain. Which will occur naturally after corporate finance functions hit CBDCs. So you have a sort of promethean progression: 0. bitmex and native perp yield 1. weird crypto yield / credit risk [low quality pre FTX era] 2. ethena (systematized perp yield) 3. defi/aave/ pendle etc (levered yield) 4. on-chain stocks 5. on-chain perps 6. leveraged stock vs perp yield < we are here > 7. private stock trading 8. institutional lending 9. direct corporate bond or equity issuance (USD) 10. CBDC facilitated corporate finance (EUR, GBP, NOK) 11. permissionless private fx swaps <the promethean explosion> 12. looped international corporate fixed income Note that the entire time I've talked so far, AI hasn't really been mentioned. AI makes all of this easier. The most concrete example is that the Norgesbank is vibe coding their CBDC with Claude. But more profoundly - eventually agents will be able to trade assets directly on chain. investment management is becoming increasingly agentic already, with every major lab launching finance products. And banks rapidly adopting AI across workflows. By the time you get to step 9 on the table above, there will likely be investable AI agents. Perhaps in gated jurisdiction. But there will be a new primitive of an agent with a verified balance, and business model that you can buy. The same way you'd buy a stock or subscribe to a vault, or hedge fund. I think Step 9 (companies choosing to issue straight on chain) is therefore the most important thing to monitor given the sh1t show with Robinhood's CEO and AMC. Tokenized representations of stocks have major legal risk, and ADRs (American Depository Receipts) as an asset class have quite an ugly history of being banned or depegging (most recently YNDX just straight up went to 0 when Russia went into Ukraine). For you to get really clean on chain stocks, and fixed income that can be effectively looped for yield, you need the credit risk of the actual underlying legal structure to be very low. You'll also want to see a proliferation of privacy and ZK accounting products take hold that interact with OTC trades, portfolio swaps, and FX as CBDCs come online. That's the boring bridge to the eventual wild future everyone was envisioning in 2023 where we have a bunch of Accelerando esque corporations existing entirely on chain, compounding capital and investing in their own training The Track to Financial RSI is paved with Fomo, hyperliquid, gambling on Robinhood and seemingly irresponsible regulatory actions. At least, it seems mad until you consider the alternative. Illiquidity. And currency failure. You can complain about Trump's ethics. You can stop Clarity. But the crazy train has already left the station. All aboard! It's not like you have a choice.

amaze amaze amaze 🤯

does seem like a better form factor for talking to it rather then having to whip out the phone but I feel that notifications would still be clearer on a screen

gotta say, @bot is amazing at finding restaurants based on your requirements

the most comprehensive report on agentic finance it has everything you need to know from the tech to the use cases thank you @Jun__Yoo for this!

went for a family dinner recently and an older family member mentioned 'ah yes I'm learning AI now through a course' a cousin of mine then asked 'what is learning AI, what does that entail exactly' to which the fam member replied 'I'm learning AI' and my cousin prompted further 'are you building a LLM, are you building a harness, like what particular aspect of AI are you working on' to which the fam member replied 'I'm learning AI' and the cousin crashed out shortly after based on a true story and the fam member replying was not an AI agent he just didn't know what else to say --- for tech-savvy folks on the internet everyday we rejoice when there's a new model with better math abilities than the asian kid at math olympiad, model sizes so big i can't figure out how to call it (what-lion? gazillions?), improved coding abilities that will take away CS courses in schools (that's a joke... I think) but to get widespread AI adoption, we will still have to discover more outcome-based apps that will appeal to the broader population. CalAI was a good example, targeting a niche group of protein-maxxing health-centric folks being able to visualize an output (generated by AI) will be important to expanding the current market of AI users

1/ When we looked at the economics of AGI, the key policy challenge was immediately clear: AI drastically lowers the cost of execution for anything easy to verify. For everything else, verification is the bottleneck.

presuming AI accelerates all work streams, we go from 5 work days (or 4 for some folks) to say 2/3 work days what would you do with the increased amount of free time? what would you buy (products/services) to occupy this new ‘leisure time’ sure, doom-scrolling/gamified finance e.g blind boxes prob takes up a good chunk of time but I believe people will default back to communities, searching for belonging / status in social groups, hyrox is a great example of this this will come easier than we think with personal agents adoption, the agent network will be the one identifying commonalities in the population - and this can potentially be worth a lot so what would you do?