@jvb_xyz

catch us live!

🔴 LIVE: Who's Coming For The Market In Q3? https://t.co/RlDbAIExlc

Compute trading probably won't happen anytime soon. ...and I really want it to. I also want to see @BrettHarrison win. I watched people try and fail to financialize LNG liquidity for years. As I study compute mkts, I am increasingly convinced they will go the way of LNG for the following reasons: 1/ Due to obsolescence, there is no enduring benchmark: A contract on H100s loses relevance as the world rolls to H200/B200/GB200/Rubin. WTI, meanwhile, has been useful for decades. 2/ The chip universe will fragment: NVIDIA won't be the only game in town for long. Soon you'll have AMD, Google TPUs, Trainium, custom hyperscaler ASICs, inference chips, sovereign AI chips, and workload-specific silicon competing for inference & training cycles. This will make benchmarking even more of a nightmare. 3/ The physical is too bespoke: Successful commodities futures have a solid benchmark w/ physical delivery at a hub (to prevent manipulation) plus a few relevant bases. In compute, it's the opposite: no solid benchmark and bases galore. PCIe vs SXM, InfiniBand vs Ethernet, 8-GPU node vs 4,096-GPU cluster, region, uptime, reserved vs spot, bare metal vs containerized, storage, egress, security, support, etc...too many taxonomies to financialize. 4/ Incumbents benefit from opacity: This one really murdered the LNG exchange startups. Just like Vitol/BP/Shell, the best-positioned compute purveyors make money from relationships & information asymmetry. They will not abdicate that edge for a few bps of startup equity. 5/ You need physical settlement for compute trading to be sustainable. There are already plenty of potential “hubs,” but until real capacity can be delivered, rejected, upgraded, discounted, or arbitraged against the screen, the futures contract is too prone to manipulation by the oligopoly of compute producers & marketers. Physical settlement, or a real EFP mechanism, is what forces convergence. Cash settlement should come after that, not before. --- Compute will probably financialize eventually...but not as "compute." It will start as narrow, ugly, basis-heavy contracts around specific chips, regions, terms, and vintages. These will probably be traded peer to peer, forming an OTC market more akin to treasury bonds than a futurized derivatives mkt akin to oil or crypto.

what do the "operations" look like @ Strategy? I imagine a mix between the early days of Stratton Oakmont and the oompa loompa scene in Willy Wonka

Displaced workers will hedge with their votes, not Kalshi. Nick nailed it.