Ray Xiao@_RayXiao · 2026/08/25观点输出RT @okxchinese: https://t.co/uSCTewJrot
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Ray Xiao@_RayXiao · 2026/08/25观点输出RT @okxchinese: https://t.co/ButyynZJP2
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Ray Xiao@_RayXiao · 2026/08/25观点输出@andyyy at least much more people are now seriously thinking about the revenue meta rather than just paying lip service to it.
AI：偏平台经营与生态扩张，强调交易平台和应用入口的长期位置。
Ray Xiao@_RayXiao · 2026/08/24观点输出Full discussion here:
https://t.co/nmvyCtP8Pf
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Ray Xiao@_RayXiao · 2026/08/24政策影响Crypto regulation is finally moving from enforcement toward market design.
At the CFTC Innovation Advisory Committee’s first meeting, Chairman Selig framed the question clearly. Blockchain, AI and prediction markets are reshaping finance. Where will these markets be built, and can U.S. regulation keep pace?
People spoke candidly about lawsuits, debanking, conflicting rules, teams moving offshore and no clear path for DeFi.
A meaningful change in direction.
AI：偏政策推进，强调合规落地与规则明晰，对监管主线更敏感。
Ray Xiao@_RayXiao · 2026/08/21观点输出What's there to be happy about? https://t.co/EuAZaKaaGl
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Ray Xiao@_RayXiao · 2026/08/20政策影响A few thoughts on how AI compute finance may develop:
AI compute is becoming a productive capital asset. Jensen Huang recently put it more directly: “In AI, compute is revenue.”
The next phase is the development of a capital market around that asset—one capable of financing capacity, pricing future compute delivery, and distributing rental rate and residual value risk.
• Debt will be the first major catalyst for the global GPU market.
As more AI infrastructure is financed with credit, lenders need to underwrite utilization, rental rates, customer concentration, technological obsolescence, and collateral value.
The missing layer is a credible term structure for compute. The market has few reliable forward-price benchmarks and almost no instruments for hedging future rental income or GPU residual values. As a result, lenders still depend heavily on sponsor balance sheets, customer guarantees, and bespoke contracts.
A functioning forward curve would change the nature of underwriting. Capital providers could begin pricing the cash flows of compute itself, rather than relying primarily on the credit quality of the operator behind it.
• Open weight models will be the second catalyst.
Until recently, advanced model compute demand was concentrated within hyperscalers and leading AI labs, so much of it never appeared in the merchant market.
Open-weight models are beginning to loosen that structure. As more enterprises deploy models independently, a growing share of compute demand is moving into arm’s-length capacity and inference markets.
Each route creates compute exposure. Capacity buyers face changing rental prices and availability. Inference providers carry compute costs inside their gross margins. For both groups, compute becomes a variable operating input that must be procured, budgeted, financed, and eventually hedged.
• Timing still matters.
AI compute finance remains nascent because the physical market is fragmented, heterogeneous, and heavily OTC.
A GPU-hour is not yet a fungible commodity. Its economic value depends on the chip generation, cluster configuration, networking, power availability, location, software environment, service level, and counterparty delivering it.
Financialization will follow contract standardization. The market is likely to begin with narrowly specified instruments tied to particular hardware, delivery periods, locations, and service levels. Broader benchmarks and derivatives can develop once enough comparable transactions exist around those standardized claims.
• Crypto native markets may have some advantages in one specific part of this stack.
We think the most credible opportunity lies in what we call Inference Capital Markets: markets for financing, transferring, and settling claims on future compute and inference services.
GPU credits and transferable claims on future service are natural starting points. They are digitally deliverable, programmable, divisible, and potentially usable as collateral. Onchain rails could make prepaid capacity easier to distribute, allow service claims to trade before delivery, and open the market to buyers and capital providers outside the hyperscaler ecosystem.
The hard problems remain offchain. A viable claim still needs enforceable service rights, standardized delivery terms, credible usage verification, and clear default resolution. Tokenization supplies the financial wrapper; contract standardization and reliable delivery give the asset its value.
Our full breakdown below👇
Ray Xiao@_RayXiao · 2026/08/11观点输出RT @a16zcrypto: First Principles Ep. 6 with Shafi Goldwasser
What if you could prove something is true without revealing why it’s true? Tu…
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Ray Xiao@_RayXiao · 2026/08/08政策影响A sharp read on what’s taking shape in the trenches and why it really matters to all of us building here.
The regulatory building blocks are stacking up, and we’re getting closer to the point where incremental progress becomes a step change. CLARITY is one piece, parallel work across agencies and growing competition across jurisdictions are adding to the same momentum. That is laying the foundation for the next phase of market evolution.
And that next phase is coming into view: internet capital markets bringing capital formation, trading, and new asset creation to internet speed and scale.
AI：偏政策推进，强调合规落地与规则明晰，对监管主线更敏感。
Ray Xiao@_RayXiao · 2026/07/16政策影响A bank runs on three things: a franchise (the brand plus the stack), underwriting authority (which assets to list, what LTVs and caps to run), and capital, the operator's own money that stands junior to depositors.
Watch DeFi lending through that lens and the past year gets interesting. Onchain there have always been two ways to run a bank: an integrated shop that handles the stack, the underwriting, and the liquidations, or a platform where the base layer provides the rails and specialist teams operate on top. This year the two converged. The integrated model started renting out the whole franchise, underwriting authority included (EtherFi's whitelabel Aave V4 instance). The platform model started writing rules for its operators (Morpho Vaults V2: role separation, caps, timelocks).
Now take stock. The franchise is for rent. Underwriting authority is changing hands. The rulebook keeps getting thicker. The capital line is still not written into any contract. Everyone onchain has opened a bank, and every one of them runs on depositors' money. Nor is the pattern confined to lending protocols. Building out our own chain's ecosystem, we keep catching it in miniature: every listing, every cap, every incentive decision is a quiet underwriting call.
Meanwhile the underwriting work itself is losing pricing power. Blue chip parameter management is commoditizing fast, and Wintermute just walked in at 0% management and 0% performance fees.
To be fair, capital light and high turnover is exactly why this industry could bootstrap at all. Curator first loss has been tried; it never scaled, never became the standard. But if onchain credit wants to carry real size, that layer gets built eventually. The open question is trade offs and timing, not direction. Top curators are one step from underwriters. Banks, insurers, and RWA shops already run balance sheets, capital charges, and loss management; what they're missing is onchain product and a regulatory path.
Step back and the industry structure writes itself. DeFi spent years unbundling the bank: rails from underwriting from capital. The rails commoditized first; the underwriting work is commoditizing now. The one piece nobody has rebuilt is the piece that made a bank a bank: capital standing junior to depositors. That is where the durable fee will sit, because it is the only part of this business that does not scale like software. If you want the underwriting fee, take the first loss.
0
评论
暂无评论