Has @OrnnExchange bought the domain of their main competitor @ComputeDesk to redirect to their site? It would be an unusual move for a US-regulated index provider. If you want to access Compute Desk, go to If you want to access Ornn, go to

Has @OrnnExchange bought the domain of their main competitor @ComputeDesk to redirect to their site? It would be an unusual move for a US-regulated index provider. If you want to access Compute Desk, go to If you want to access Ornn, go to

Navier-Stokes and other frontier math discoveries have shown that multi-agent ensembles are required to escape the typical creativity constraints of LLMs. Is this also applicable to algorithmic trading research, which similarly lacks direct answers in LLM training data? The general method for producing a proof of an unsolved math problem has been to set up large teams of agents to - create multiple formations of the problem - selectively research or exclude known results - search for counterexamples or use counterexample attempts to tighten problem bounds - randomly apply unrelated areas of mathematics - use formal verification methods to check other agents’ work - project manage and referee across all existing agents Single-agent systems tend to plateau to a mean knowledge level. This covers the majority of tasks at an efficient cost per token, as the answers to most queries lie explicitly in the latent space of LLMs. Clear examples where single agents fall short are complex system design and numerical analysis, both of which describe algorithmic trading research. The ensemble method has potential to discover new alpha in algorithmic trading, but at the cost of substantial compute and tokens. Frontier model companies won the competition for quant/engineering talent against large trading firms a few years ago, but are not yet deploying them in this direction.

At the current rate of compute financialization, CDOs on GPU-backed leases are likely coming to market. Non-investment-grade neocloud operators with 8-9 figure budgets are struggling to secure financing for under double-digit interest rates, and bundling is a short-term solution as a liquid residual value curve develops. A GPU-backed CDO could be structured as a portfolio of GPU leases across operators, chip generations, and geographic locations. The tranches could be • Senior notes sized to contracted utilization and investment-grade offtakers • Mezzanine notes that take on contract renewal risk • Subordinated/equity notes that absorb residual value depreciation Ratings agencies, including startups specializing in the assessment of compute quality, face the challenge of modeling default correlation with very little historical data. Any analysis would need to take into account offtaker credit-worthiness, GPU-hour rental prices, new accelerator manufacturing, memory and electricity costs, and political factors affecting buildouts. Listed compute futures and options are the solution the US financial system is waiting for, but that shouldn’t prevent financial institutions from deploying every available innovation to address the growing credit crisis, even if inputs are nascent.

Architect is hiring for three lateral roles: options trading quantitative researcher, compute capital market sales, and structured derivatives specialist. Details on what we’re building and the new team members we’re looking for Quantitative Researcher: We’re building: automated high-frequency options trading strategies on legacy and new(er) exchanges What you’ll work on: volatility modeling, ML for parameter optimization, order placement simulation, new venue exploration Stack: Rust, Python, AWS, ClickHouse What we’re looking for: field experience in derivatives trading and research, strong system design instincts, multiple approaches to options pricing theory, bias towards shipping Nice to have: familiarity with the latest ML toolkits and exchange microstructure Location: your preference Compute Capital Markets Sales: We’re building: the first US-regulated exchange purpose-built for compute futures, options, swaps, forwards, and capacity, plus AI supply chain commodities What you’ll work on: building relationships and facilitating trades with neoclouds, lenders, insurance providers, inference companies, datacenters, AI labs, interdealer brokers, retail brokers, hedge funds, and asset managers What we’re looking for: network of established and startup AI companies, fluency with compute capacity pricing/structures, translation skills between finance and AI, aversion to slop, commitment to competitive markets for compute Nice to have: experience in compute procurement at a hyperscaler, neocloud, or lab Location: SF Structured Derivatives Specialist: We’re building: specialized contracts for trading and hedging compute, compute derivatives, and a range of physical commodities supporting the US datacenter build-out What you’ll work on: modeling and structuring OTC products with Architect’s traders/technologists and AI counterparties for compute, memory, inference, and energy trading What we’re looking for: experience trading OTC derivatives contracts at banks or trading firms, sell-side institutional network, enthusiasm for working at startups Nice to have: relevant FINRA/NFA certifications or passion for test-taking Location: your preference

Ever since the announcement, there's been some interesting ideas shared about how Nvidia is functioning like a central bank of the AI industrial/infra build out. One main criticism is that there's increasing concentration risk. A liquid and robust derivatives market would do a lot to mitigate concentration risk. It's fascinating to see in real time because similar stories were prominent case studies from my formal education. The GPU residual value floor reminds me of similar terms that are common in the airline industry, big ag heavy machinery, and the auto industry. There are fair criticisms of circular financing, concentration risk. This is ultimately why a third party exchange like The AI Exchange would be a more sustainable way to legitimize chips and make AI infra an investible asset class. It would also give Nvidia an opportunity to offset these types of risks.

Thrilled to announce @BrettHarrison, founder and CEO of @Architect_Fi, is joining The Ledger Lab! He'll sit down with Clément Salaün (CTO, @formancehq) and Mai Trinh (CEO, @netbackyard) on September 30 to talk building a market for compute GPUs. RSVP:

Nvidia’s GPU backstop deals are equivalent to writing put options at the residual floor and buying a smaller notional of call options at the same strike. If Nvidia could sell GPU rental price forwards or futures they could hedge the partial synthetic long.

The mechanics and complexities of futures expirations, and how HFTs and market makers handle them from a technical perspective. • A single futures contract is associated with more than one expiration-related date. Physically settled contracts in particular have first notice dates, last trading dates, last notice dates, last delivery dates, final settlement dates, and exchange-defined “roll dates” that may be none of the above. The underlying asset type determines which dates are most relevant for tracking expiration effects. In addition, an exchange’s current trade date is not necessarily the calendar date. For example, CME advances the trade date to the next business day at 16:00 CT. HFTs maintain accurate metadata including all relevant expiration dates by building proprietary symbology databases. Firms reconcile across multiple input sources including real-time binary-encoded symbol definitions from live market data feeds, daily universe files from FTP, and third-party normalization services such as Bloomberg and FactSet. • The official front month is not necessarily the most-traded front month. Volume and open interest routinely migrate days to weeks ahead of last trade date or first notice date. For example, Treasury futures roll most open interest in the last 10 business days before the delivery month starts. Liquidity rolls in energy and agricultural products are determined by commercial hedging and index fund activity. Market makers decide which futures contracts to reference based on empirical liquidity measurements. Live and historical market data capture services are essential for tracking when the market decides to shift to the next expiry. • ETFs with futures holdings will convert front month positions to back month positions gradually over a predefined roll period. Arbitrage market-making systems that accurately price the fair value of futures-based funds need to reference multiple contract expiries in the right quantities based on the trading day. The indexes used to settle commodity perpetual futures use a similar liquidity rolling strategy. ETF market makers download basket definition files directly from issuers and encode the roll mechanics from prospectuses to ensure the fund holdings are modeled correctly. Sometimes the issuers publish mistakes in their own basket files, providing an additional source of edge for participants savvy enough to catch them. • Efficiently trading out of front month futures positions and legging into the next expiry requires an understanding of market microstructure. Trading firms time execution to the optimal combination of order book depth and spread tightness in the front month contract, back month contract, and atomic calendar spread if supported by the exchange. HFTs build models that predict calendar roll prices and spreads as a function of time-to-expiry. Firms also deploy algorithms that gradually execute roll strategies while minimizing alpha leakage. Some firms will pre-position ahead of the liquidity roll based on known positions of large funds in order to take advantage of the expected flow. • Most futures brokerages automatically roll or liquidate positions to avoid expiry, especially in the case of physically settled commodity futures. The specific date used as the deadline depends on the clearing firm, with first notice and last trade the two most common cutoffs. Auto-liquidations incur extra fees and often result in large amounts of slippage from inefficient execution. The solution for HFTs is to become direct members of futures exchanges and avoid any intermediation by third-party clearing firms or brokerages. In place of external processes, the internal back-office teams at HFTs and market makers build monitoring tools to ensure futures trading desks do not unintentionally take positions to expiry or delivery. Exchange membership is prohibitively costly and time-consuming for non-HFT market participants. Solving the above issues with futures expirations has become a hard-won source of edge for many HFTs and market makers. While perpetual futures reduce complexity and deadweight cost, they also remove profit opportunities for trading firms and fee capture for exchanges, which is an under-discussed source of the current pushback against listing commodity perpetuals in the US.

We’re launching the first tungsten perpetual futures on AX. Volatility of tungsten prices has uniquely affected the entire compute supply chain, as it’s used in a range of physical forms: - ultra-thin metallic films that connect microscopic transistors and fill contacts inside advanced processors - chemical vapor deposition during chip production in the derivative form tungsten hexafluoride - drill bits required to precision-drill the tiny multi-layered holes in printed circuit boards - word-line and gate metallization inside 3D NAND and high-bandwidth memory stacks - tungsten-copper composite wire drawn to hair-thin diameters for compact high-density modules - wear-resistant tooling that packages finished AI chips and boards Ammonium Paratungstate prices have surged 600% year-over-year, and realized volatility has entered a different regime from standard industrial metals. Tungsten represents a disproportionate risk in the AI supply chain basket. Liquid futures combined with cross-margin opportunities for other metals derivatives provide an overdue, capital efficient solution for dampening forward price risk.

Tungsten is coming to Architect’s AX. Perpetuals, not cubes. Hedge metals supply chain risk at every stage of compute manufacturing: metals mining, chip and memory fabrication, gas generation, electricity consumption, GPU capacity procurement, and inference creation.

The first reaction to finding a high-Sharpe strategy in simulation should be to assume there’s a mistake. Markets are adversarial, data is messy, and small modeling errors compound into misleading results. Common mistakes in training systematic trading models and some solutions: Omitting fees and slippage: Nearly all equities, options, futures, and digital asset exchanges charge higher fee amounts for removing liquidity than for adding liquidity. HFT strategies that hold positions for only seconds or minutes frequently see their expected edge disappear once realistic fees are applied. Another common mistake is assuming a liquidity-taking order is fully filled at the best bid/ask instead of walking the book, consuming successive price levels up to available size. This both systematically underestimates cost and overestimates the capacity of the strategy. Accurately modeling fees and slippage requires building automated reconciliation between predicted costs and exchange-reported fees and fill prices. In-sample contamination: Statistical arbitrage strategies that fit model parameters to historical data need to avoid evaluating performance on the period used for training. The naive method of splitting data is to hold back the most recent days from the training set, which may result in training the strategy on a different market volatility and momentum regime than the recent past. Simple K-fold cross validation, a common fix from machine learning, is inappropriate for time series as this method leaks future information into the past. The most successful data splitting techniques involve purged K-fold, walk-forward optimization, or combinatorial purged cross-validation. Incompatible clocks: Backtesters that reference timestamps based on the market data capture machine’s clock overestimate achievable fill ratios. A partial solution is to refer to exchange timestamps for order book events to determine whether liquidity still exists when orders are sent. However that solution is also unreliable as exchanges have wide distributions of latencies between matching engine actions and book update dissemination. A robust heuristic is to re-run the backtest under a range of artificial order-submission delays and examine the sensitivity of both fill ratio and P&L. Survivorship bias: The choice of instrument universe itself introduces survivorship bias. For example, choosing the constituents from today’s S&P 500 index systematically selects historical winners only, as those names that underperformed have already been excluded from the basket. One solution is to record point-in-time universes to use for future historical studies, and another is to fix today’s universe and implement a walk-forward simulation until enough data has been collected.

Excellent and comprehensive discussion of the critical market opportunities in compute derivatives. Appreciate Architect’s inclusion as US exchanges approach regulatory approval and begin building robust futures markets in compute.

Can inference costs be commoditized? On OpenRouter there’s an 11x spread between the cheapest and most expensive provider of a single model, DeepSeek V4 Flash. Other relevant data points on open-weight inference prices, speeds, availability: • Baidu serves DeepSeek V4 Flash at $0.049 per million tokens and at 124 tokens per second. 26 of the 30 other providers are both more expensive and slower, so it’s not a tradeoff of speed vs cost. • Across the 117 models on OpenRouter with three or more providers, the median price spread between the cheapest and priciest host is 2.2x, the most extreme is 14.4x (DeepSeek v3.2). • Among all endpoints serving the above, 37% of endpoints are strictly dominated by cost, speed, and uptime. The number goes up to 52% if considering only cost and speed.

Compute Markets referenced a couple of times in today's committee meeting. Everyone agrees that these markets should exist and fast-tracked. TLDR: @FalconXGlobal @2Ragu - AI CapEx hit $670B in a single-year. These will be the largest markets ever and an opportunity for the CFTC to define compute as an asset class. AI Labs going through IPO processes are being asked (for the first time) about their compute or datacentre expenses and they want to find ways to hedge to reduce their cost of capital. @multicoin @tushar_jain requesting for an innovation exemption for compute derivatives. @blockchain doubles down on Multicoin's idea for a sandbox environment for compute. Don Wilson says compute markets gives US capital efficiency in the AI race. Duffy highlights that Kalshi's compute markets are live whilst everyone else must abide by CFTC's request for comment period.

Accurately pricing derivatives when underlyings are closed is a universal source of edge in market making, from tier-1 firm graybox ETF trading to newer operations in 24-hour prediction markets and perpetual futures. Some widely applicable pricing strategies: Related futures and currency moves. When a derivative’s underlyings are closed, market makers use other instruments as proxies that have sufficient beta to the derivative. Examples for a US ETF on Japanese stocks: the USD/JPY pair trades 24/7 through a variety of FX ECNs, interdealer platforms, and futures exchanges. Nikkei 225 index futures also trade on multiple global derivatives exchanges during extended hours. Moves in these assets during the Japanese night will on average predict the opening prints of individual stocks listed on JPX. Home market index moves. An ETF moves with a non-trivial correlation factor to other US names simply because it’s a US-listed security. The effect is easily observable during heightened volatility. It’s a common industry saying that in a market-wide selloff, all correlations go to 1. Both narrow- and broad-based indexes of US names explain some of the signals in mid-frequency alphas. US index returns comprise a small but meaningful component of derivatives’ multifactor beta models. News. A key requirement for pricing derivatives on foreign stocks is processing local news and earnings releases that relate either to the particular stocks or relevant stocks in the same sector. Quantitative trading firms use automated translation tools to process local foreign-language news and suggest idiosyncratic adjustments to traditional factor models. Recent advances in LLM-based NLP have made sentiment analysis viable for blackbox trading systems to react instantaneously to news-based signals. Microstructure. High-frequency trading firms successfully and counterintuitively price derivatives by ignoring the underlyings’ characteristics. Firms extract short-term alphas from order book characteristics, recent returns, microprices, and other microstructure features. From the perspective of a reinforcement-learning non-linear model trainer, unlabeled feature sets and time series data result in ETFs, ADRs, and common stocks being treated as mathematically equivalent. Based on the above, here is a rough model of how large market-making firms predict mid-frequency returns when traditional underlying markets are closed: ΔP = β_FUT · ΔFUT + β_FX · ΔFX + β_IND · ΔIND + [News Term] These firms also heavily prepare for circumstances under which this model fails. FX and futures have idiosyncratic moves, home-market betas can break during regime shifts, news sentiments still have positives, and microstructure signals decay fast. The market shift underway in 24/7 traditional derivative and prediction contract trading will test the model’s longevity for market making and statistical arbitrage going forward.

The top trading firms in compute markets are going to be firms that are already leasing GPU compute, training their own models, managing their own infra. It's an incredibly young market with no way to warehouse risk. Having skin is the only way to get the invaluable experience with the nuances of the market as it's developing. Firms that hesitate will be too far behind considering that this experience is the byproduct of training bespoke models.