Big Firms Invest, Sovereign Bonds Go Onchain: Tokenized Finance Picks Up Speed
Industry Flash
10 items
10 items
共 10 条



云巨头第一次把”按次计费”搬上链:企业给AI代理设稳定币预算,代理用USDC逐笔结算每次云端调用,月结发票的时代正在被改写。
10月7日,新加坡TOKEN2049与Sui Basecamp同时开幕,AI与链上金融的融合成为全场主线。当天Sui联合阿里云宣布,将阿里云的算力、存储与推理服务接入Sui Agent Payments,让AI代理用稳定币为每一次API调用单独付款,不再按月度发票结算。
同一周,Mysten Labs还和谷歌云推出Verifiable Agent Arbiter,用于验证代理是否真的留在被授权的范围内花钱。一个让机器花得了钱,一个让机器证明”这钱是它被允许花的”,两条线合起来,指向同一个判断:机器原生支付的结算层与问责层正在同步成形。
此前机器付款的主角是协议层玩家——Coinbase的x402、AWS的Bedrock AgentCore、Stripe与Tempo的自建协议。这次是阿里云把整条云服务目录接到链上,按调用计费,由代理自己持有稳定币钱包按次结算。
真正的信号不在单一合作,而在”云巨头+稳定币”这个组合。阿里云完全有能力自建账单系统,却选择把结算放到公链上用美元稳定币完成。这背后是对企业AI支出形态的一种预判:当算力消费开始由代理自主驱动,月结发票和人工对账将不再适配。
Sui并非先行者。Coinbase与Cloudflare的x402协议把HTTP 402″付款要求”状态码变成机器付款的标准动作,到2026年3月已在Base处理超1.19亿笔、Solana超3500万笔,年化约6亿美元,且Coinbase对协议收费为零。AWS推出Bedrock AgentCore Payments,Stripe则押注稳定币链Tempo。机构结算层也在同步补位——Solana在10月6日发布了联合摩根大通设计的DvP结算标准。
需求侧信号同样在放大。Robinhood Crypto副总裁在TOKEN2049披露,已开立15万个AI代理账户,99.3%的x402支付以USDC结算,Robinhood Chain上70%的交易发生在周末或常规交易时段之外。稳定币已成为AI代理间支付的默认结算货币。
AI支付题材在本次会议持续发酵。Tom Lee在TOKEN2049将”AI代理链上付款”列为牛市四大驱动之一,Arthur Hayes也推出面向AI代理经济的FLOP项目。但市场反应分化:Sui合作公布后交易量清淡,比特币在会议当日上午一度从约86590美元跌至83790美元,一小时内清算超4亿美元。
从估值看,机器付款仍处概念兑现阶段。真正需要跟踪的不是发布会,而是稳定币支付的实际放量、企业采购团队是否把代理钱包纳入预算审批,以及各协议在微支付场景下的真实成交。Coinbase零协议费的x402、AWS的云内闭环、Stripe的支付巨头路径,与Sui的”高吞吐结算+问责层”路线会形成怎样的事实标准,仍是这场竞争最值得观察的分野。
对普通用户来说,变化正在从交易端渗透到钱包端。Robinhood的15万个AI代理账户意味着,越来越多投资者开始授权代理在设定目标与条件后自主执行交易,而非自己逐笔下单。NEAR Intents等产品则让用户只描述想要的交易结果,由代理在多个链与中心化交易所之间寻找流动性完成成交。
隐私是这层迁移里最现实的坎。代理要读取用户完整持仓才能下单,敏感的财务信息因此暴露给模型。NEAR联合创始人强调需对代理本身做隐私保护,用机密计算对喂给AI的交易数据与请求做加密处理。对企业而言,Sui与阿里云的模式把决定权收回到”预算”层面——用户设定上限与授权范围,代理在边界内自主消费,越界即被链上规则拦截。
机器原生支付正在从协议实验走向商业结算,本次会议给出了三条可观察的线索:一是稳定币坐实了”机器原生货币”的位置,99.3%的x402支付以USDC结算即是佐证;二是竞争从”谁能花”升级为”谁能花且能被证明该花”,花钱层与证据层正被组合成一套完整的企业级方案;三是需求侧的真实放量仍落后于供给叙事,90%的微支付不足1美分、机构采用有限。
对从业者而言,值得跟踪两个方向:代理支付的稳定币结算量能否从”活跃钱包驱动”转为”广泛企业采用驱动”,以及各协议在微支付场景下能否沉淀出可持续的收入结构。当采购团队真正把代理钱包写进预算流程,供给侧的赌注才算得到验证。
共 10 条



HM Treasury closes its payments consultation on Oct 6, setting the first regulatory framework for agent-initiated payments and bringing UK qualifying stablecoins inside the perimeter of money.
At 11:59pm UK time on October 6, HM Treasury closed the 12-week consultation on Modernising Payment Services Regulation. Widely described by law firms as the most consequential payments rewrite in a decade, the document brings UK qualifying stablecoins into the regulatory category of money for the first time, while asking a novel question: when an AI agent initiates a payment on your behalf, how should authentication, consent and liability for unauthorised transactions be handled.
The same day, OpenAI said it would add invisible watermarks to ChatGPT and Codex text in the EU to comply with the AI Act. The two stories look unrelated but point to one thing: AI is beginning to actually spend money, and regulation is catching up with the machines.
The UK’s Payment Services Regulations 2017 predate the AI-agent boom, and no line mentions machine-initiated payments. The consultation uses 42 questions to cover stablecoins, tokenised deposits, Open Banking and agentic payments, built on the principle of same risk, same regulatory outcome: the risk, and therefore the result, should not depend on whether the payer is human or AI. Set against the global backdrop, the move carries real weight: the US CLARITY Act stalled in the Senate on September 15, while Britain is using one unified framework to accommodate fiat, stablecoins and tokenised deposits, and explicitly baking agentic payments into the premise of its rule design.
The data explains why the regulator cannot afford to wait. Research commissioned by the FCA found that around one in five UK adults, some 11 million people, would use AI to act within preset limits to book, switch, pay and manage money. The Payments Association’s survey is blunter: 58% of UK online merchants say an AI agent has already completed a transaction on their platform, but only 41% are confident in the liability framework supporting those transactions. Analysts value the global agentic-AI financial services market at roughly $7.8 billion in 2026, on course to exceed $40 billion by 2031, with a compound growth rate above 40%.
The underlying infrastructure is maturing too. Bank of England Deputy Governor Sarah Breeden has laid out a multi-money framework, where bank deposits, tokenised deposits, regulated stablecoins and a potential digital pound run in parallel inside one retail payment system; a pound is a pound, whoever issues it. The BoE published draft sterling stablecoin rules in June and plans to finalise them by year-end. Lloyds, NatWest and Barclays, in UK Finance’s Great British Tokenised Deposit project, have completed the first interbank tokenised-deposit transfers over blockchain rails.
From an allocation standpoint, the agentic-payments line is gaining both regulatory certainty and capital attention at the same time. On PitchBook’s count, UK VCs deployed more than £800 million into AI payments startups in the first three quarters of 2026. The Bank of England is set to finalise systemic stablecoin rules by year-end, and the FCA’s full cryptoasset framework is expected to land around October 2027, giving institutions predictable milestones.
The rivalry between tokenised deposits and stablecoins is also worth watching. The BoE keeps an issuance cap on private stablecoins and leans toward bank-issued tokenised deposits; Lloyds, NatWest and Barclays have already moved real money across tokenised deposits. This bank-first path suggests the underlying money of machine payments may come more from the regulated banking system than from purely private stablecoin issuers.
The October 6 consultation deadline in the UK is the first move to write AI-agent-initiated payments explicitly into the design of a payments framework. Together with OpenAI’s EU text watermarking and the BoE’s multi-money framework, it draws the outline of a world where machines actually manage money and rules are racing to catch up. For practitioners, two signals are worth tracking: how the UK’s stablecoin equivalence mechanism lands, and the final answer on where responsibility for agentic payments sits. These two points will decide how fast, and how far, machine-native payments can go.

英国财政部10月6日关闭支付新规咨询窗口,首次为AI代理付款设立监管框架,并把英国合规稳定币纳入”钱”的范畴。
10月6日英国时间23:59,英国财政部长达12周的《现代化支付服务监管》咨询窗口正式关闭。这份被多家律所称为近十年最重要支付规则重写的文件,把英国合规稳定币首次纳入”钱”的监管范畴,同时破天荒地抛出一个问题——当一个AI代理替你发起付款,认证、同意和未经授权交易的责任该怎么算。
同一天,OpenAI宣布将在欧盟为ChatGPT和Codex文本加隐形水印以履行《人工智能法案》。两条新闻看似无关,却指向同一件事:AI开始真正”花”钱了,监管正在追上机器。
英国现行《支付服务条例2017》诞生于AI代理爆发之前,条文里没有一个字提到”机器付款”。这次咨询用42个问题覆盖稳定币、代币化存款、开放银行和agentic payments四条路线,核心思路是”same risk, same regulatory outcome”——同样的风险,同样的监管结果,不论付款方是人还是AI。放在全球看,这个动作的分量很清楚:美国CLARITY法案9月15日在参议院受挫,而英国选择用一套统一框架同时容纳法币、稳定币和代币化存款,并明确把agentic payments写进规则设计的前提。
数据说明监管为什么等不起了。FCA委托研究显示,约五分之一、约1100万英国成年人会用AI代自己在预设范围内订票、换服务、付钱、管账。The Payments Association的调查更直接:58%的英国线上商户声称AI agent已在其平台完成交易,但只有41%对支撑这些交易的问责框架有信心。分析师把2026年全球agentic-AI金融服务市场估在约7.8亿美元,2031年预计冲上40亿美元以上,复合增速超40%。
底层基建也在成熟。英国央行副行长Sarah Breeden提出”multi-money”框架——同一套零售支付体系里,银行存款、代币化存款、受监管稳定币和潜在的数字英镑并行,”a pound is a pound”,谁来发行都一样可靠。BoE已于6月发布英镑系统稳定币草案,计划年底定稿。Lloyds、NatWest、Barclays在UK Finance的Great British Tokenized Deposit项目里,首次用区块链完成行间代币化存款转账。
从配置角度看,agentic payments这条线正同时获得监管确定性和资本关注。PitchBook口径显示,2026年前三季度英国VC已向AI支付初创投入逾8亿英镑。英国央行定于年底定稿系统稳定币规则,FCA的完整加密资产框架预计2027年10月落地,这些时间表给机构提供了可预期的落地节点。
代币化存款与稳定币的路线之争也值得留意。BoE对私人稳定币保留发行上限、更倾向银行代币化存款,Lloyds、NatWest、Barclays已在行间用代币化存款完成真实转账。这条”银行优先”的路径意味着,机器付款的底层货币,可能更多来自受监管的银行体系而非纯私营稳定币发行商。
英国10月6日的咨询截止,是全球首个把”AI代理付款”明确写进支付框架设计的动作。它和OpenAI为欧盟文本加水印、BoE的multi-money框架共同勾勒出一件事:机器真正开始管理资金,而规则正在追上。对从业者而言,值得紧盯的两个信号,一是英国稳定币等价认可机制怎么落地,二是agentic payments责任归属的最终答案。这两点将决定机器原生支付能走多快、能走多远。