Stripe Volume Hits $1.9 Trillion as Tech Drives Half of US Growth

Stripe交易额达1.9万亿美元，科技驱动美国半数增长 传统的循序渐进式跨国扩张策略已彻底瓦解，因为现代互联网企业如今从成立第一天起就面向全球启动。 该演讲重点介绍了Stripe在2025年取得的重大财务里程碑——总处理金额达1.9万亿美元，并在科技和企业领域占据了主要市场份额。演讲审视了加速发展的全球经济格局，其特征表现为快速的市场分化、“生而全球化”（global by default）的初创企业发布模式，以及在专注于支付的Tempo区块链孵化支持下B2B稳定币交易量的大幅激增。此外，Stripe还概述了“高收入模式”（high revenue mode）支付优化的未来，并推出了诸如Agentic Commerce Suite等结构性框架，以便为自主AI驱动的商业生态做好准备。 Stripe上的企业在2025年共产生1.9万亿美元的总交易额，同比增长34%，约占全球GDP的1.6%。 (0:00) 市值和利润的集中化正通过一台快速运转的“分拣机”加速推进，排名前三分之一的美国上市公司占据了总市值的约三分之二。 (1:47) 在AI赋能的快速创业和更快变现的推动下，2025届Stripe新企业的增长速度比2024届快50%。 (3:49) 传统的循序渐进式跨国扩张模式已被“生而全球化”的发布模式所取代，新型互联网企业从成立第一天起就在全球范围内实现变现。 (5:45) 尽管加密货币市场整体低迷，但2025年稳定币支付交易额仍翻番至约4000亿美元，其中60%由B2B交易推动。 (9:01)

In its 2025 annual letter, Stripe reports processing $1.9 trillion in volume—roughly 1.6% of global GDP—while arguing that the competitive landscape is rapidly bifurcating around AI adoption, borderless financial rails, and automated commerce. Market Concentration and the Rapid Sorting Machine (1:47) Stripe frames modern commerce around an accelerated 'sorting machine' that consolidates profits and market capitalization faster than in past eras. The most profitable third of publicly listed US companies now controls two-thirds of the total market capitalization—the highest proportion since records began in 1963—with the top 10% of the S&P 500 generating roughly 59% of index profits. In individual sectors, divergence is pronounced: three-year US ecommerce sales rose 30% against 5% for brick-and-mortar retail, and Delta and United captured virtually all US airline profits in 2025. Global-by-Default Startups and AI-Led Acceleration (3:49) Unlike historic corporations that took decades to expand abroad, contemporary internet and AI firms operate globally from inception. Among Stripe's 2025 cohort, 57% of new businesses originated outside the United States, and the number of startups reaching $10 million in annual recurring revenue within three months doubled year-over-year. International revenue for globally oriented Stripe merchants increasingly relies on smaller markets, with 30% stemming from outside home markets and the top 10 global economies. Development cycles have shortened alongside this shift; GitHub code pushes rose 41% year-over-year, and Atlas startups charging customers within 30 days reached 20% in 2025, up from 8% in 2020. Stablecoin Integration and Infrastructure Alternatives (9:01) While broader cryptocurrency valuations fell, stablecoin payment volume doubled to approximately $400 billion in 2025, with an estimated 60% flowing through B2B transactions. Traditional financial networks and early public blockchains face severe performance constraints, illustrated by a memecoin trading surge that delayed business payouts by 12 hours and multiplied transaction fees 35 times. In response, Stripe and Paradigm incubated Tempo, a payment-focused blockchain with sub-second finality and dedicated lanes, currently tested by Shopify, Visa, and Nubank. Protocols and Architecture for Agentic Commerce (18:33) Stripe outlines a five-level taxonomy for agentic commerce, progressing from basic form autofill (Level 1) and descriptive natural language search (Level 2) to persistent context (Level 3), autonomous delegation within spending parameters (Level 4), and proactive anticipation without prompts (Level 5). The company assesses that the current commercial ecosystem sits at the boundary of levels one and two. To prevent fragmented walled gardens, Stripe released the Agentic Commerce Protocol with OpenAI, introduced Shared Payment Tokens to safeguard credential data, and rolled out machine payments enabling autonomous agents to clear API microtransactions. Regulatory Friction and Capital Access Constraints (12:14) Drawing on economic historian Joel Mokyr's critique of institutional resistance, Stripe warns that a modern 'Republic of Permissions' threatens technological adoption through bureaucratic hurdles like the EU AI Act and slow clinical trial processes. This friction mirrors capital scarcity for small enterprises, where OECD per capita GDP growth dropped to 1% annually post-2008 and US small business loan approval rates declined to 41% in 2025. Stripe notes that embedded real-time lending can counter this drag, citing internal randomized data showing financed businesses grew 27 percentage points faster than unfinanced counterparts. Stripe concludes that the rapid advances in AI and financial automation represent an 'event horizon' where underlying conditions have fundamentally transformed, though the ultimate economic benefits depend heavily on reforming regulatory barriers and establishing open interoperability standards.

Stripe在2025年年度信中指出，全球经济与商业竞争正进入筛选机制加速与利润高度集中的新周期，2025年其处理交易规模达到1.9万亿美元（占全球GDP约1.6%）。面对跨国扩张周期缩短、AI加速落地与小微信贷收缩等变局，Stripe主张通过支付底层优化、稳定币与专用区块链开发，以及制定代理商业协议，为AI驱动的全球经济提供基础架构支撑。 市场分化与全球化创业范式转移 (1:47) Stripe指出，市场竞争的“筛选机制”运转速度前所未有，推动商业价值向极少数头部企业高度集中。目前美国市值排名前三分之一的上市公司占据了总市值的近三分之二，创下1963年有统计以来的最高纪录；标准普尔500指数中市值前10%的企业贡献了约59%的总利润。各行业分化同样剧烈：过去三年美国实体零售经通胀调整后仅微增5%，而电商增长达30%；达美航空与联合航空几乎包揽了2025年全美航空业的全部利润；软件、硬件与数据中心投资则贡献了2025年近一半的美国GDP增长。 伴随市场集中化的是初创企业扩张节奏的根本性转变。2025年加入Stripe的新企业中，57%位于美国境外，该群体平均增速比2024届快约50%，且创立三个月内年化经常性收入（ARR）突破1000万美元的企业数量翻倍。传统按部就班的跨国扩张模式已被颠覆——以往巨头如可口可乐、麦当劳乃至早期互联网公司拓展海外市场或支持多币种结算需历时数年甚至数十年；如今ChatGPT、Claude和Cursor等AI产品上线即面向全球。在面向全球市场的Stripe企业中，30%的海外收入来自本土市场与全球前十大经济体之外的“长尾”地区。 稳定币爆发与专用支付链Tempo (9:01) 信中强调，在加密货币二级市场回调的背景下，稳定币支付却迎来大规模落地。2025年比特币自10月高点下跌50%，而全球稳定币支付规模翻倍至约4000亿美元，其中约60%估计为企业间（B2B）结算。Stripe收购的稳定币编排平台Bridge交易量增长超四倍，已有初创公司直接以稳定币接收投资、存入Stripe金融账户并向跨国工程师支付薪酬。此外，Bridge与Visa合作推出支持稳定币余额的借记卡，结账时自动转换为法币；拥有2000万月活用户的Phantom钱包亦在通过Bridge推行此类卡片。 然而，现有公链专为DeFi与投机交易设计，吞吐量有限且容易出现拥堵。例如比特币每秒处理少于10笔交易，某种模因币交易狂潮曾导致Bridge用户的结算延误12小时，手续费暴增35倍。为此，Stripe与Paradigm联合孵化了专为支付打造的区块链Tempo。Tempo具备专属支付通道、亚秒级结算确定性及合规支持。目前Visa、Nubank和Shopify正在对其进行测试，Klarna亦成为首家在Tempo测试网上通过Bridge发行稳定币KlarnaUSD的银行机构。 支付基建优化与小微企业信贷供给 (12:14) 2008年金融危机后，OECD国家人均GDP年均增速从此前46年间的2.8%放缓至1%。Stripe分析认为，巴塞尔协议III和多德-弗兰克法案推高了大型银行资本门槛，导致小微企业信贷严重萎缩；爱尔兰银行对小企业的贷款在2011至2019年间锐减逾66%，2025年美国小微企业贷款获批率降至41%。针对此缺口，Stripe Capital利用实时交易数据进行授信并从每日销售额中自动扣还，2025年发放资金同比增长45%，覆盖超8.1万家企业。双盲对照研究显示，接受融资的企业在随后一年内的增速平均高出同行27个百分点，低信用分企业增速亦高出11至18个百分点。 面对宏观不确定性，企业押注传统营销往往成效难测，例如吉列因广告争议导致母公司宝洁计提80亿美元减值，印度铁路获奖的彩票营销亦未能拉动车票销量。相较之下，底层支付基础设施优化具有高度确定性。Stripe通过动态自适应、网络令牌及AI反欺诈等工具帮助微软提升了授权率；盖特威克机场接入后支付接受率提升2.5个百分点；远程医疗公司Ro授权率提升2个百分点且争议率下降3%，每年节省数千万美元。本地化支付工具BLIK（波兰）和Pix（巴西）则分别带来46%与31%的结账转化率提升。 代理商业五级演进与协议层构建 (18:33) 针对AI代理（AI Agent）代人类进行采购的趋势，Stripe将其演化划分为五个阶段：第一级是消除网页表单填写；第二级是基于多维度情境的描述性搜索；第三级是具备跨会话持久记忆；第四级是在预算与参数限定下由AI自主决策下单的委托执行；第五级则是完全无需提示词、基于日程与需求的自主预期采购。Stripe评估，当前行业正处于由第一级向第二级过渡的临界阶段。 为避免AI商业陷入各自为政的封闭围墙，Stripe主张构建开放的底层标准。Stripe与OpenAI联合制定了开放的代理商业协议（ACP），并推出“共享支付令牌”（Shared Payment Tokens），使代理能够在不暴露真实信用卡凭证的前提下发起安全交易。针对商户端，Stripe推出了代理商业套件（ACS），帮助Urban Outfitters、Coach等品牌同时对接ACP及Google发布的通用商业协议（UCP）。此外，Stripe还上线了“机器支付”功能，允许开发者通过稳定币微支付，按API调用或HTTP请求直接向自主AI代理计费。 许可体制约束与技术创新奇点展望 (23:51) 借用经济史学家Joel Mokyr的研究，Stripe指出工业革命不仅依赖物理资源，更依托于视现状为可改进的“改善型心智”。然而，当前社会正滑向由监管机构、委员会与司法审查构成的“许可体制”（Republic of Permissions），人为主观设置的准入门槛正在削弱先进技术的渗透速度。信中直言，若要释放AI制药潜能，必须改革漫长的临床审批流程；欧洲经济若要重获活力，需削减如《欧盟人工智能法案》等繁琐负担；下一代核能与自动驾驶亦需打破否决政治和地方性法规的围追堵截。 尽管存在制度阻力，信中依然列举了Mistral AI、Zipline与Varda等在严苛监管环境中推进技术落地的代表企业。Stripe在结尾中将当前的AI爆发比作“跨越黑洞的事件视界”：在跨越临界点的那一刻局部体感平稳自然，但未来的可能性空间已发生剧烈转变。Stripe预测未来十年将展现截然不同的商业与创业形态，并表示将继续押注技术基建与企业家精神。 Stripe的年度信依托其微观交易数据与跨国业务版图，描绘了由AI代理与稳定币重塑的商业远景。然而，其构想的顺利实现高度依赖于各科技巨头能否真正落实开放协议互通、专用区块链在主流金融中的渗透进程，以及各国监管机构是否愿意实质性放宽对新兴技术与金融创新的制度约束。

[0:00] host: full, proper me [1:47] the sorting machine is "whirring faster," [3:49] host: Someone should check on the App Store review team's sleep scores. [3:49] larger inflection in entrepreneurship and creativity, facilitated by advances in large language models [5:45] the domestic market for a new generation of internet businesses is the internet itself. [5:45] the long tail is much of the dog. [9:01] host: it may be a crypto winter, but it's a stablecoin summer. [15:09] evangelically animated about payments. [16:20] low revenue mode [16:20] high revenue mode [18:33] This is what most people mean today when they talk about agentic commerce. [22:12] cooped up in walled gardens, but they'll be zooming down the wide-open protocol highways. [23:51] improvement mindset [23:51] Republic of Permissions [23:51] puncture the regulatory permafrost. [26:53] you don't actually feel anything at the special moment when you cross the event horizon. The path is locally smooth, even though the space of possible futures changes dramatically.

[0:00] We’re interrupting your usual Cheeky Pint programming today to bring you Stripe’s [0:03] 2025 annual letter, read in full, by me. Not AI me. [0:07] Full, proper me. [0:09] Normal service will come back on Thursday when I'll be welcoming Reiner Pope, [0:12] founder of AI chip company, MatX, to the booth. [0:15] But on to the letter. [0:17] Last year, businesses running on Stripe generated $1.9 trillion in total volume, [0:21] up 34% from 2024, and equivalent to roughly 1.6% of global GDP. [0:27] Our programmable financial services now power more than 5 million businesses [0:31] directly or via platforms, including all the top AI companies, [0:35] many of the largest blue-chip companies, 90% of the Dow Jones Industrial Average, [0:39] most of the biggest tech companies, 80% of the Nasdaq 100, [0:43] and a significant fraction of freshly minted startups—25% of all Delaware [0:47] corporations are now created with Stripe Atlas. [0:50] Beyond payments, these businesses are using Stripe [0:52] to accelerate their growth with billing and subscription management, [0:55] tax compliance, fraud prevention, embedded finance, global treasury [0:59] management, and much besides. [1:01] Link, the easiest way to pay online, is now used by more than 200 million people. [1:07] Stripe remained robustly profitable in 2025, allowing us to continue investing [1:12] heavily in product development, with over 350 product updates over [1:16] the last year, as well as acquisitions. [1:18] Since our last update, we acquired Privy, which powers more than 110 million [1:22] programmable wallets, and Metronome, which powers the complex usage-based [1:26] billing models used by companies like OpenAI, Anthropic, Confluent, and NVIDIA. [1:30] Metronome joins our Revenue suite, comprising Stripe Billing, Invoicing, Tax, [1:35] and more, which is on track to hit an annual run rate of $1 billion this year. [1:39] All in all, 2025 was a strong year for the internet economy, [1:43] and we're delighted to see so many of Stripe's customers doing well. [1:47] Let's talk about what we've been dubbing the sorting machine. [1:51] At their heart, competitive markets are a sorting machine. [1:54] They direct profits, capital, and talent [1:56] to the places of greatest impact, as discerned by customers voting with their wallets. [2:01] Historically, this sorting happened methodically. [2:04] It typically took decades for a household name to be unseated or for a new [2:08] entrant to reach meaningful scale. [2:11] The sorting machine is now whirring faster: winners and losers are being [2:14] anointed more quickly and more intensely. [2:17] Today, the most profitable third of publicly listed companies [2:20] in the United States account for two-thirds of the total market [2:23] capitalization, the highest share since data began in 1963. [2:27] And much of this is a story of profit concentration, not just valuations: [2:32] the top 10% of the S&P 500 by market cap now accounts for roughly 59% [2:36] of the index's total profits, which is elevated relative to recent history. [2:41] Much of this is driven by bifurcation within industries. [2:44] In retail, for example, US brick-and-mortar sales grew just 5% [2:48] over the past three years, whereas ecommerce sales grew 30% [2:52] over the same period (both of those in inflation-adjusted terms). [2:56] In air travel, the "big 3" of American, Delta, and United all increased their [3:01] share of industry revenues and profits over the past decade. [3:03] They were big to start with, and now, Delta and United accounted for nearly [3:08] all US airline profits in 2025. [3:12] In healthcare, hospital and insurer profit shares have contracted significantly since [3:16] 2019, but health tech is on track to exceed $110 billion in EBITDA by 2029. [3:22] Each sector has its own particular dynamics, but the pattern is clear: [3:26] a cohort of companies is pulling away. [3:28] Economy-wide, demand for software, computers, and data center investment [3:32] drove nearly half of all US GDP growth in 2025 and will likely soon [3:36] be a majority of all US growth. [3:38] Based on what we can already tell from the set of businesses that started [3:42] on Stripe in 2025—a remarkable cohort— there are no signs of the sorting [3:46] machines slowing down anytime soon. [3:49] In 2025, many more new companies joined Stripe than ever before, with more than [3:54] half of them (57%) based outside the US. [3:57] This new cohort is by far the highest performance programming and fastest moving [4:01] we've ever seen, growing around 50% faster than the 2024 cohort on average. [4:06] The number of companies reaching $10 million in ARR within three months of [4:09] launch was double that of the 2024 cohort. [4:13] This seems to be part of a larger expansion and acceleration [4:16] in our industry. [4:17] After years of relative calm, the number of iOS apps released [4:20] in December 2025, jumped by 60% year over year. [4:24] Someone should check on the App Store review team's sleep scores. [4:27] Even code production is accelerating. [4:29] Pushes to GitHub, which grew roughly 10% to 12% in prior years, jumped 41% [4:35] between Q3 '24 and Q3 '25. [4:38] As building gets easier, we're working on making Stripe even [4:41] simpler to integrate, including for the agents. [4:44] We recently introduced claimable Stripe sandboxes, [4:47] which lets you start using Stripe directly from your AI coding tools like Manus, [4:51] Base44, Replit, and Vercel. [4:53] When you're done with v1 and your product is ready to launch, that sandbox converts [4:57] into a live Stripe account with its configuration intact. [5:00] More than 100,000 sandboxes have already been created this way. [5:03] We're also improving Stripe Atlas, the world's easiest way to incorporate [5:07] a business, which saw a 41% increase in company formations last year. [5:11] Atlas companies are monetising sooner. [5:13] In 2025, 20% of Atlas startups charge their first customer within 30 days, [5:17] up from 8% for that same number in 2020. [5:19] As we look at these figures, there's an obvious question: is 2025 an [5:23] anomaly or the beginning of a new regime? [5:25] Time will tell, but our best guess is that the 2025 acceleration is the start [5:30] of a larger inflection in entrepreneurship and creativity, facilitated [5:33] by advances in large language models. [5:36] We have an ambitious roadmap of improvements planned. [5:39] Stripe will be the best way to build a business in the era of AI. [5:45] Now, let's talk about how businesses are increasingly global by default. [5:48] For those with aspirations to go global, the conventional playbook used to be [5:51] a steady, sequential progression. [5:53] You win at home and then you push abroad. [5:56] It took Coca-Cola 20 years to bottle its first soda in Cuba, [5:59] while McDonald's and Starbucks waited 27 and 16 years, respectively, [6:02] to serve their first customers in Canada. [6:04] After the arrival of the World Wide Web, free services tended to launch globally, [6:08] but their monetization machinery still operated on a time [6:11] delay to catch up. [6:12] When Facebook changed from a college-only network to a public platform in 2006, [6:17] anyone with a browser could create an account, but anyone with money [6:19] couldn't necessarily advertise. [6:21] Support for international currencies didn't arrive until 2009, [6:24] five years after the company was founded. [6:26] For its part, Google only accepted its first payment in British pounds [6:29] from an advertiser in the UK, which was actually a live lobster mail [6:32] order firm, no joke, in 2002, four years after launching [6:36] its search product globally. [6:38] Over the last few years, the country-by-country expansion [6:40] model has just melted away. [6:42] The domestic market for a new generation of internet businesses [6:45] is the internet itself. [6:47] Nearly every AI product you've heard of is all the rage in every [6:50] country you've heard of. [6:51] ChatGPT, Claude, Replit, Lovable, Base44, Vercel, Cursor, Midjourney, and many [6:56] more launched globally by default. [6:59] This isn't merely about incremental revenue from a long tail [7:02] of international users. [7:03] In many cases, the long tail is much of the dog. [7:07] Of businesses on Stripe that earn most of their revenue internationally, [7:10] 30% of that revenue comes from countries that are neither their home market nor [7:14] one of the top 10 global economies. [7:17] This is possible largely due to infrastructure that no longer [7:20] makes foreign demand feel foreign. [7:23] Last year, we enabled businesses to launch a localized checkout in more than 100 [7:27] countries simultaneously, complete with localized pricing to maximize [7:29] conversion, more than 120 payment methods, and local tax compliance [7:33] supported out of the box. [7:34] Sometimes, improved infrastructure is only felt after decades, but other times, [7:38] pent-up demand reveals itself overnight. [7:41] Gamma is a California-based AI platform used by 70 million people [7:44] to create presentations. [7:45] When they joined the first cohort of Stripe users to accept UPI payments [7:48] in India, their Indian revenue jumped 22% in that very same month. [7:53] Ironically, the tech companies most constrained by barriers of financial [7:57] geography have tended to be fintech themselves. [7:59] Twelve years after launching Chime, you can only open a new Chime account [8:03] from inside the United States. [8:04] Nubank, founded in 2013, served only Brazil for its first six years [8:09] and added only two other countries in the last six. [8:11] These are some of the most successful fintechs out there. [8:14] Even at Stripe, our Issuing product is available in only 22 countries, [8:18] some seven years after launch. [8:19] That's faster than most, but not as fast as we would like. [8:22] This is also changing. [8:24] The latest cohort of fintech companies— Sling Money, DolarApp, Félix, and KAST, [8:29] to name just a few— are building global financial [8:31] apps right out of the gate. [8:33] Similarly, last year, we launched our first globally native [8:36] product, Financial Accounts, which businesses can use to hold, send, [8:39] and receive funds. We made it available to businesses [8:41] in more than 100 countries on day one. [8:44] This progress is in large part due to stablecoins, [8:46] where borderlessness allows fintechs to set up infrastructure [8:51] that works everywhere. [8:53] Global-by-default financial services are, for the first time, a real possibility. [9:01] Now, let's talk about stablecoins. [9:03] It may be a crypto winter, but it's a stablecoin summer. [9:06] After a decade of stablecoin volumes tracking the undulations of the crypto [9:10] asset prices, last year saw a clear divergence. [9:13] In 2025, the price of Bitcoin dropped precipitously and is now 50% down [9:17] from October, but stablecoin payments volume doubled to around 400 billion, [9:22] 60% of which is estimated to represent B2B payments. [9:25] Bridge, the stablecoin orchestration platform we acquired, [9:28] saw volume more than quadruple. [9:30] Stablecoin payments are marching quietly and inexorably forward as [9:34] real-world uptake continues apace. [9:37] This growth has been catalyzed by a raft of new capabilities. [9:40] A Y Combinator founder can now receive funding in stablecoins, [9:43] hold them in a Stripe financial account, and use them to pay their first engineers [9:47] who could be anywhere in the world. [9:48] Enterprises leaning on stablecoins to expand internationally, now with better [9:52] tools to embed digital wallets directly into their core products. [9:54] With Privy, companies like Ramp and Deel have a single API to provision easy-to-use [9:58] wallets in both custodial and non-custodial models, [10:01] which allows fully global products to be built from day one. [10:04] The interoperability between crypto and fiat is also rapidly improving. [10:09] In April, Bridge partnered with Visa to introduce cards that allow businesses [10:11] and consumers to spend their stablecoins just like any other card. [10:15] The payment is deducted from a stablecoin balance and automatically [10:18] converted into the local currency. [10:20] The business receives funds just like any other payment, blissfully unaware [10:23] of the stablecoin origins. [10:25] Phantom, one of the most popular crypto wallets with 20 million monthly active [10:28] users, is using Bridge roll out stablecoin-backed cards [10:31] to their customers. [10:33] As we prepare for a world of massive stablecoin adoption, we spent [10:36] time thinking about blockchains. [10:38] Today's blockchains have been designed for trading and DeFi, [10:42] and the attributes that matter for payments, [10:43] including throughput, reliability, cost, predictability, and privacy, have not [10:47] been a significant focus. [10:49] Bitcoin processes fewer than 10 transactions a second. [10:52] Last year, a memecoin trading frenzy on one of the major blockchains delayed [10:56] payouts for one Bridge user by 12 hours and spiked per-transaction [11:00] prices 35 times. [11:02] While such operational issues are already significant, they'll only intensify, [11:06] for we expect the appetite for transactions to grow a great deal. [11:09] In our view, agents will most likely soon be responsible for most internet [11:13] transactions and will likely need blockchains that support more than [11:16] a million, or maybe even more than a billion, transactions a second. [11:19] In September, we unveiled Tempo, a blockchain purpose-built for payments [11:24] incubated together with Paradigm. [11:26] With Tempo, businesses get dedicated payment lanes, sub-second finality, [11:30] opt-in privacy, and interoperability with compliance and accounting systems. [11:34] These features might sound prosaic, but they matter a great deal [11:36] for infrastructure that supports real-world economic activity. [11:41] Companies like Visa, Nubank, and Shopify are already testing Tempo [11:44] for a number of use cases, including global payouts, [11:46] embedded finance, and remittances. [11:48] Klarna, whose CEO was once a self-proclaiming crypto skeptic, [11:52] became the first bank to launch a stablecoin—KlarnaUSD— [11:55] on the Tempo testnet, using Bridge's Open Issuance to facilitate [11:58] faster and cheaper cross-border settlement. Tempo's architecture is also particularly [12:02] well suited to agentic payments and microtransactions. [12:06] Tempo's mainnet will be launching soon, and we look forward to seeing what [12:08] ambitious businesses decide to build with it. [12:14] Now, let's talk about the availability of capital. [12:17] Since 2008, GDP per capita growth has been a sluggish 1% in OECD [12:21] countries, down from 2.8% [12:23] a year on average for the 46 prior years when data begins. [12:27] It's always tempting to reach for local explanations—Japan's aging workforce, [12:32] or Brexit, or Europe's energy mix, or all these things. [12:34] But there's good evidence that a [12:36] significant culprit is the sharp global drop in the availability of capital. [12:40] Indeed, in most OECD countries, capital requirements for banks went up [12:44] and capital access for small businesses went down following [12:47] the Global Financial Crisis. [12:49] Basel III reforms raised capital standards for large banks around the world. [12:53] In Ireland, bank lending to small businesses dropped by more than 66% [12:57] between 2011 and 2019. [13:00] In the UK, small business lending contracted in 2012 [13:03] and remained weak thereafter. [13:04] The US fared slightly better with easier access to capital from [13:08] nonbank lenders and 1.7% [13:09] per capita GDP growth over the last 15 years. [13:13] But the broad trends, post-Dodd-Frank, are the same: [13:16] tightening of bank rules and reduced access to capital for small businesses. [13:20] Only 41% of small business loan applications were approved in the US [13:24] last year, down from 50% in 2015. [13:27] We created Stripe Capital to help solve the global paucity of access to [13:31] capital, especially for small businesses. [13:34] For businesses that use Stripe to accept payments, the real-time revenue data makes [13:37] for a simple input to a lending decision. [13:39] Businesses pay back loans as a small percentage of their subsequent sales, [13:43] effectively drawing working capital forward from their own growth. [13:47] Last year, Stripe Capital distributed 45% more funds compared to 2024 [13:51] to more than 81,000 businesses. [13:53] Many of these businesses had access to Stripe Capital funds directly through [13:57] the vertical SaaS platforms on which they run—GlossGenius for beauty salons, [14:00] Tekmetric for auto shops, Pixieset for photographers, and so on. [14:04] Over the last two years, we ran a randomized study to understand [14:07] the causal impact of Capital: [14:09] when we help businesses grease their wheels, how fast can they grow? [14:14] Turns out, a lot faster. [14:16] Businesses that received and accepted Capital offers grew on average [14:19] 27 percentage points faster over the following year than the otherwise [14:23] equivalent businesses who didn't. [14:25] And the averages conceal a wide spread. [14:28] The fastest-growing decile of financed businesses grew more than three [14:31] times faster than comparable peers. [14:34] The next decile grew nearly 100 points faster. [14:37] A representative example, Xirsys, a server hosting business based [14:41] in California, used financing from Stripe Capital to set [14:43] up additional servers in China, India, and Japan, subsequently doubling its revenue. [14:48] Notably, even businesses with low credit scores grew 11 to 18 percentage points [14:52] faster after receiving financing. [14:54] It's speculative, but we wonder whether access to capital [14:57] will become a more important factor in economic outcomes over the coming years, as advances [15:02] in artificial intelligence increase the returns on investment. [15:09] Now, let's talk about what we've taken to calling low revenue mode. [15:13] For mature businesses, there's notably few guaranteed growth [15:17] levers hiding in plain sight, and the most enticing ones don't always work out. [15:21] Take advertising. [15:22] In 2019, Gillette won a Silver Lion at Cannes for their ad, "The Best Men Can Be." [15:27] Later that year, P&G took an $8 billion dollar write-down on the business, [15:31] as the award-winning campaign failed to ignite growth. [15:34] More recently, a campaign for Indian Railways won a Grand Prix and six [15:39] Gold Lions at Cannes in 2025 for its golden-ticket-style daily lottery where [15:43] each day they'd have a train ticket number that was eligible to win 10,000 rupees, [15:47] but the campaign was shut down after eight weeks with no discernible [15:50] increase in ticket sales. [15:51] This is why, 15 years in, we are evangelically animated about payments. [15:56] Unlike more speculative paths to growth, optimizing your payment setup is almost [16:00] guaranteed to yield extra revenue and to be among the highest ROI growth [16:04] activities that you could undertake. [16:06] Microsoft, for example, evaluates the precise performance of each [16:09] of their payment service providers on a monthly basis. [16:12] Between June and October, Stripe delivered a meaningful increase [16:15] in authorization rates for Microsoft using combination of Adaptive Acceptance, card [16:19] account updater, network tokens, and more. [16:20] As a result, Microsoft is now routing a larger share of its payment [16:24] volume through Stripe. [16:25] Gatwick Airport had been dealing with a string of failed [16:27] payments, disputes, and customer complaints under its previous payment provider. [16:31] After switching to Stripe, payment acceptance rate rose by 2.5 percentage points. [16:36] Following an A/B test against a former payment provider, FICO implemented Stripe [16:39] and saw a 1 percentage point increase in auth rates. [16:42] Telehealth company Ro saw a 2 percentage point increase in auth rates and a 3% [16:47] decrease in dispute activity over the last 12 months with Stripe, resulting [16:51] in tens of millions of dollars annually. [16:52] Most businesses are today operating in what we call low revenue mode, [16:56] running on old, unoptimized payments infrastructure that's [16:59] leaking dollars left, right, and center through conversion, auth, [17:02] and fraud prevention rates that should be so much better. [17:05] High revenue mode looks like a checkout that adapts differently to each customer. [17:09] You should be pricing your products in local currency and offering [17:10] the right set of the hundreds of local payment methods that are most relevant. [17:15] Presenting BLIK to your customers in Poland drives an average 46% increase [17:20] in checkout conversion. [17:21] Offering Pix does the same in Brazil with a 31% average conversion uplift. [17:26] This is all powered by a decades-long investment in AI, [17:29] including our Payments Foundation Model and AI-powered services like [17:32] Stripe Radar, Optimized Checkout Suite, and Authorization Boost that quietly [17:36] optimize billions of dollars of transaction volume every day. [17:40] And yet there's still so much more room for improvement. [17:42] Last year, we started testing the ability for consumers to tap their card [17:44] on their own phone for authentication, which validates the card through the NFC [17:49] chip and all but guarantees the card holder actually has possession [17:52] of the physical card. [17:54] DoorDash, a partner in this trial, has seen a meaningful increase [17:56] in conversion, versus their previous fraud checks, while reducing chargeback rates. [18:00] We also built new models for Radar to handle a wider range of fraud vectors [18:03] emerging in response to the AI boom— in particular, the very common attempts [18:06] to steal AI inference through abusing free trials or similar mechanisms. [18:10] So CEOs, splashy ad campaigns. [18:13] They're good fun, but don't overlook the revenue growth right under your nose. [18:17] Your heads of payments almost definitely deserve more recognition. [18:20] So let's maybe throw them an awards ceremony in the south of France [18:23] for payments optimizations. Or if you want a foggier version of that, [18:26] we'd love to see them in San Francisco this April for Stripe Sessions. [18:33] Now, let's talk about agentic commerce and the different levels of agentic commerce. [18:38] Most of the recent excitement in AI has been around some [18:40] form of tool use: models able, not only to cogitate on data within the pre-training corpus, [18:45] but to go out and search the web or use a browser, deploy code, or similarly, [18:49] take some action on the wider internet. [18:51] The form of tool use most relevant to our world is agentic commerce: [18:54] the idea that your AIs will soon be buying stuff on your behalf. [18:58] Now, like most in AI, agentic commerce suffers from having [19:01] been a bit overhyped too early. [19:04] People paint this utopian picture of autonomous agents running around, [19:07] planning your whole life, executing commerce without even talking to you. [19:12] We find it helpful to build up a broader vision of agentic [19:15] commerce in small chunks. [19:17] Level one is just eliminating web forms. [19:20] You research what you want to buy, but filling out web forms [19:23] is no one's idea of fun. [19:25] It would be really handy if you could just send the URL to your agent and have it [19:28] fill out the form, the payment and shipping details, and come [19:31] back to you with the confirmation. [19:32] The system isn't making any decisions for you. [19:34] It's just typing and clicking "buy" on your behalf. [19:37] Then level two is descriptive search. [19:40] You stop searching for products or specific attributes or keywords [19:44] and start describing the situation. [19:46] You might say, "I need back-to-school supplies for a third grader in Chicago, [19:50] including clothes, nothing too itchy or tight, pencils, [19:53] notebooks, a lunch box, my son likes KPop Demon Hunters and [19:56] tennis, and school starts in late August. [19:58] The system can reason across weather and materials and size and durability [20:01] and price and taste and reviews and delivery timelines and everything. [20:04] Specialized and long-tail products, they become easier to find. [20:08] Annoyingly blunt keyword searches become a thing of the past. [20:12] Then level three is about layering in persistence. [20:15] You can stop reintroducing yourself every single time. [20:18] Each search is not a brand new scenario. [20:22] You can say instead, "Find me options for my back-to-school clothes for Bobby." [20:27] The system already knows your preferences and remembers any requirements and infers from your [20:32] previous conversations and purchases. [20:33] You're still deciding what to buy. [20:35] You're clicking confirm on every purchase, but you're choosing from a set of options [20:38] that reflects you and your taste and budget and history. [20:41] Level four is where you get into delegation. [20:45] You stop having to make all the choices. [20:47] You could say, "Get the back-to-school shopping done and keep it under $400." [20:51] The system handles the search and the evaluation process [20:54] and the purchases on your behalf. [20:56] You trust it, like you might your partner, that it'll weigh the trade-offs. [21:00] And it'll choose things that your son would like. [21:02] All you do is determine the budget. [21:04] This is what most people mean today when they talk about agentic commerce, [21:08] is this level-four level of delegation. [21:12] Level five is anticipation. [21:14] Maybe, you don't do any prompting. [21:17] The system already knows the school calendar, your son's preferences, [21:20] your typical budget. All you do is receive a notification. [21:23] "Here's the back-to-school list of everything that we've already bought." [21:25] This is the most futuristic vision where things are showing up right before you [21:29] need them without you having to ask. [21:31] But today, the industry is hovering right on the edge of levels one and two. [21:37] With agentic commerce, we're reminded of those few years [21:40] in the mid-1990s when the structure of the internet we use today was hashed out. [21:44] Netscape developed graphical web browsers. HTTP, HTML, URLs, DNS [21:48] all gained prominence. [21:50] At the time, no one knew exactly which protocols or players would win out. [21:53] There was an AltaVista for every Google. [21:56] We're in a similarly rare moment now with agentic commerce, which is the [21:59] potential to be generationally impactful. [22:01] But as with the early internet, the future success of agentic commerce is [22:04] contingent on universal interoperability. [22:07] Our ascent through those five levels we discussed depends on our [22:10] ability to all work together. [22:12] To that end, we've been busy. [22:14] A few highlights from our last year include: [22:17] With OpenAI, we developed the Agentic Commerce Protocol [22:20] to establish a shared technical language between AI platforms and businesses. [22:24] It's open by design, working across payment providers and across AI platforms. [22:28] We introduced Shared Payment Tokens, a new payment primitive that lets agents [22:32] initiate payments without exposing credentials. [22:34] You don't want to be flinging your credit card all around the web. [22:36] Even businesses that don't process payments with Stripe can forward these [22:39] tokens to their own vaults or other processors as secure credentials. [22:42] We launched Agentic Commerce Suite, which provides tooling for businesses [22:46] to sell across multiple AI interfaces and protocols, including our own ACP [22:50] and the Universal Commerce Protocol that Google unveiled last month, [22:52] with a single integration. [22:54] Checkout, payments, and fraud protection continue to work predictably [22:57] underneath the hood. [22:58] Brands already onboarding to ACS include Anthropologie, Urban Outfitters, Etsy, [23:02] Coach, Kate Spade, and a bunch more. [23:04] We launched machine payments, a way for developers to charge agents [23:08] directly for API calls, MCP usage and HTTP requests using [23:11] stablecoin microtransactions. [23:13] Autonomous agents themselves are emerging as a new customer type [23:16] for internet businesses to sell to. [23:18] And we partnered with OpenAI to power the first shopping experience inside ChatGPT. [23:23] This is huge. [23:23] We're also collaborating with Microsoft to bring similar capabilities [23:26] to Copilot and with others. [23:29] There's no forecasting exactly where agentic commerce will be by the end [23:33] of 2026, but it's clear we've already moved well beyond pure hype into a phase [23:36] of building and real-world experimentation, and the pace of change [23:39] will likely only accelerate from here. [23:41] If all goes well, the little critters, they're not going to be cooped up [23:44] in walled gardens, but they'll be zooming down the wide-open protocol highways. [23:51] Lastly, let's talk about entrepreneurship more generally. [23:55] Today's entrepreneurs and innovators have tools and reach that prior generations [24:00] of industrialists could not have fathomed. [24:02] We'll hopefully soon witness the combinatorial effect of human [24:03] ingenuity playing out in the form of productivity gains and improved [24:08] living standards everywhere. [24:10] Last year, Joel Mokyr was awarded the Nobel Prize in Economic Sciences. [24:14] Mokyr is widely known for emphasizing the importance of culture relative [24:18] to the traditional economic inputs of capital, labor, and technology. [24:21] Eighteenth-century industrialists, they didn't just have coal [24:23] or geography on their side. [24:24] They had a new culture, an "improvement mindset" that saw the status [24:28] quo as imperfect and correctable. [24:31] In the political economy of technological change, Mokyr also observed that new [24:35] technologies have in the past often failed despite their economic superiority, [24:39] because technological decision-making implicates not only suppliers [24:43] and customers, but also a broad variety of non-market "aggregators"—regulators, [24:48] committees, courts—that influence what is adopted. [24:51] As AI and the internet expand the scope of what's possible, synthetic impediments [24:55] to adoption and adaptation will become increasingly costly. [24:59] Our bifurcating economy shows that growth is contingent on the application of useful [25:04] knowledge and not some preordained result as a result of its abstract availability. [25:08] AI harbors the promise of enormously improving drug discovery... [25:12] but the potential will only be realized if we make the regulatory processes, [25:15] including clinical trials, faster and cheaper. [25:18] Entrepreneurs in Europe can boost tepid economies with new tools... [25:21] But only if well-intentioned yet counterproductive burdens such [25:24] as the EU AI Act are curtailed. [25:27] Next-generation approaches to nuclear energy could usher an energy abundance... [25:30] but only if we overhaul vetocratic regulatory regimes. [25:35] Autonomous transport and logistics from long-haul trucking to drones could [25:38] dramatically reduce the cost of physical goods... provided we don't let a slurry [25:42] of local ordinances and rent-seeking harden into a blockade. [25:47] Mokyr wrote about the importance of the Republic of Letters [25:50] in catalyzing the Industrial Revolution. [25:52] But today, we inhabit a Republic of Permissions: a filtering [25:55] sieve of non-market aggregators. [25:57] While many of our structures are sensibly motivated, it's more important than ever [26:00] to ensure that they carefully balance the benefits achieved [26:03] with the possibilities foreclosed. [26:05] We're privileged to support many businesses with the tenacity [26:09] to show what's possible. [26:10] Mistral AI and Bending Spoons are proving that world-class European talent can [26:14] puncture the regulatory permafrost. [26:16] Zipline and Varda are earning permissions for complex new hardware inch by inch, [26:20] while Spring Health and Maven Clinic are stitching together a new software [26:24] layer for modern healthcare. [26:26] We continue to believe in the importance of ideas fueling economic progress [26:29] and believe that many of the best are undervalued. [26:32] The Stripe Press catalog is home to many of the people, stories, [26:35] and models we think can contribute to the next set of improvements. [26:37] The team recently celebrated 1 million books sold. [26:41] Works in Progress, our magazine of underrated ideas [26:44] to improve the world, has also recently branched out [26:45] into print subscriptions, and you can get yours over at worksinprogress.co/print. [26:51] We highly recommend it. [26:53] To wrap up, much of this letter has been dedicated [26:56] to advances in AI, which can seem hard to keep up with at times. [27:00] The qualitative difference between the just-released products and last [27:02] year's state of the art is huge. [27:05] We're reminded of the phenomenon of falling into a large black hole. [27:08] I hope this never happens to you, but if you do ever experience [27:11] that particular misfortune, you don't actually feel anything [27:14] at the special moment when you cross the event horizon. [27:17] The path is locally smooth, even though the space of possible futures [27:20] changes dramatically upon the point of crossing the threshold. [27:24] We're sharing this letter on what might well turn out to be the advent [27:27] of a different and hopefully much more beneficient singularity. [27:31] While much around us in 2026 feels similar to prior years, it's locally smooth, [27:35] it's also clear that the next decade will look very different to those just gone by. [27:40] We're as enthusiastic as ever about how vibrant entrepreneurship and wise cultures [27:45] can contribute to more successful future societies, and we hope that Stripe [27:49] can play a small role in that. [27:51] If you're propelling economic growth yourself, whether as an entrepreneur, [27:54] a business leader, or a financial [27:55] infrastructure builder, we hope to welcome you at Stripe Sessions in April. [27:59] As always, there's so much to discuss.