AI Payment Undercurrents: Google Brings 60 Allies, Stripe Builds Its Own Entire Road

marsbit发布于2026-02-23更新于2026-02-23

文章摘要

The AI payment war is intensifying as major tech companies race to control the infrastructure for AI-driven transactions. Google has formed an alliance with over 60 traditional financial and tech companies, including Mastercard and PayPal, to establish the "AI Agent Payment Protocol." Meanwhile, Stripe has taken a more independent approach by acquiring key companies like Bridge (for stablecoin capabilities) and Privy (for wallet technology), co-developing the Tempo blockchain with Paradigm, and launching the Agentic Commerce Protocol (ACP) with OpenAI. This allows AI platforms like ChatGPT to enable seamless, in-chat payments without redirecting users. At the heart of the conflict is the "toll" for processing AI transactions. Stripe’s strategy involves building a full-stack solution—from stablecoin accounts and blockchain infrastructure to banking licenses—while Google’s coalition relies on established financial networks. Notably, Circle and its USDC stablecoin emerge as a likely winner regardless of which camp dominates, as both ecosystems depend on compliant, auditable digital dollars for settlement. The broader implication is the need for a financial system capable of supporting autonomous AI agents conducting economic activities. While Stripe envisions a future where AI handles end-to-end transactions, Google’s alliance prefers integrating AI with existing human-centric systems. Regardless, the adoption of stablecoins for AI payments is accelerating, with regulatory and...

By Lin Wanwan

Money already lives in the code.

Six months ago, AI payment was just a concept on presentation slides. Now, AI is becoming the "checkout counter".

Open ChatGPT today, search for any product, and you'll see a blue Buy button. Enter your address, pay, and it ships. The entire process happens without redirects or opening any web pages.

Last week, Google followed suit, integrating products from Etsy and Wayfair into its Search and Gemini, allowing checkout directly within the conversation. Microsoft's Copilot simultaneously launched its shopping checkout feature. Meta's Zuckerberg just announced a full shift towards AI agent commerce.

But a more covert business battle is also quietly unfolding. The fight over AI payment tolls can be traced back to the two major AI payment camps in the fall of 2025.

On September 16th, Google gathered over 60 companies and released an "AI Agent Payment Protocol".

The list was full of familiar names from traditional finance: Mastercard, PayPal, American Express, plus a few allies from the tech circle.

On the 29th of the same month, Stripe, jointly with OpenAI, released another protocol, the Agentic Commerce Protocol (ACP). Stripe simultaneously announced it was testing ACP-based agent commerce solutions with AI-native players like Microsoft Copilot, Anthropic, and Perplexity.

The two lists had little overlap. Coinbase appeared in both Google's AP2 ecosystem and is a long-term partner of Stripe.

What these two camps are fighting over is a question that sounds mundane but is worth trillions: When AI spends money on behalf of humans, through whose pipe does the money flow?

You might think this is far from you. But think about it: you now ask ChatGPT to book flights for you, use an AI assistant to compare prices and buy things, or have an Agent automatically purchase office supplies. These scenarios are visibly becoming reality. Every transaction needs a pipe to move money from your pocket to the merchant.

Whoever builds this pipe can charge a toll on every transaction.

This is the essence of this war.

12 Months Changed by a Roundtable

The story starts with a dinner.

In the summer of 2024, Stripe hosted then-U.S. Deputy Treasury Secretary Wally Adeyemo at its San Francisco headquarters for a fintech roundtable.

A group of payment company CEOs sat around chatting. Among them were two people who had never met before: Stripe's CEO Patrick Collison, and a young man named Zach Abrams.

Abrams had significant credentials. He and his partner Sean Yu were serial entrepreneurs who sold their first company, Evenly (a P2P transfer app, similar to the American Venmo) to Square (now called Block) in 2013.

Later, Abrams became the head of consumer products at Coinbase and also served as Chief Product Officer at Brex; Yu worked as an engineer at DoorDash and Airbnb. In 2022, the duo teamed up again to found Bridge, helping businesses integrate stablecoin payments. Clients included Coinbase and SpaceX.

The roundtable discussion was originally broad, but Abrams later recalled being shocked: over 90% of the time was spent talking about stablecoins, even though he represented the only stablecoin company present.

Before that, Bridge had been pursuing Stripe as a client, hoping to integrate its technology into Stripe's payment system. But after that roundtable, the direction changed. Collison started frequently meeting with Abrams, not to discuss cooperation, but acquisition.

In October 2024, Stripe announced the acquisition of Bridge for $1.1 billion. Bridge had just raised a $40 million Series A (its first institutional round) in March 2024, valuing the company at $200 million.

The acquisition price was 5.5 times the valuation, potentially over 100 times revenue multiples. In a post-investment note, Sequoia Capital said they believed Bridge would join the ranks of Instagram, YouTube, PayPal, and WhatsApp as "one of those companies that realized its full potential after being acquired."

In February 2025, the deal officially closed. Bridge's 60-person team moved into Stripe's San Francisco headquarters and participated in Stripe's bi-weekly new employee boot camp.

This was just the first step.

What happened next occurred rapidly. In May 2025, Stripe launched stablecoin financial accounts, allowing businesses in 101 countries to directly hold stablecoin balances and send/receive payments globally using stablecoins.

In the same month, ChatGPT launched a shopping recommendation feature, allowing users to search for products, compare options, and then jump to the merchant's website to place orders.

In June, it acquired the wallet company Privy.

Privy did something simple: allow any app to have a built-in digital wallet, enabling users to complete on-chain payments without downloading additional cryptocurrency wallet software. Over 75 million accounts were already using it.

Patrick Collison tweeted a straightforward statement: "Money has to reside somewhere, and Privy builds the world's best programmable vaults."

In September, it jointly incubated the Tempo chain with crypto investment giant Paradigm, a brand new blockchain specifically designed for payments. Paradigm's co-founder Matt Huang (also a Stripe board member) personally led the team.

The list of companies joining the Tempo design camp read like an all-star game of the payment industry: OpenAI, Anthropic, Deutsche Bank, Visa, Shopify, Standard Chartered Bank, Brazil's largest digital bank Nubank, DoorDash, Revolut, South Korean e-commerce giant Coupang.

Stripe CEO Patrick Collison stated that Tempo could process tens of thousands of transactions per second, with sub-second confirmation, fees of less than 0.1 cents per transaction, and transaction fees denominated in USD stablecoins, eliminating the need to hold highly volatile native tokens.

In the same month, Stripe and OpenAI officially released the ACP protocol, simultaneously launching ChatGPT's Instant Checkout feature—users could see recommended products in the dialog box and directly place orders and pay with one click, without redirecting or swiping cards.

The first to support it were Etsy merchants, followed by Shopify's millions of merchants.

In October, Tempo completed a $500 million first round of financing, led by Greenoaks and Thrive Capital, with participation from Sequoia, Ribbit Capital, and SV Angel, valuing it at $5 billion. A blockchain project less than two months old, valued at $5 billion. Stripe and Paradigm themselves did not participate in this round.

In December, Tempo opened for public testing. UBS Group AG, Mastercard, and European buy-now-pay-later giant Klarna joined the partner list.

Bridge's Zach Abrams simultaneously announced that Bridge had applied for a national bank trust charter in the U.S. to comply with the requirements of the stablecoin regulatory act signed into effect in July 2025, the GENIUS Act.

String these events together: $1.1B for coin issuance capability, creating stablecoin financial accounts, acquiring a wallet company, incubating a dedicated blockchain, applying for a banking license.

From issuing coins to building chains to making wallets to setting protocols to obtaining licenses, Stripe did every layer itself.

In contrast, Google: over 60 alliance members, one open protocol, one code repository. Google has everything, except its own chain, its own stablecoin, its own wallet.

An alliance is the product of people sitting down for meetings. What Stripe built is a system that can be launched by one person's decision.

The month Google released AP2, Tempo was already in testing.

Whoever Wins, Circle Wins

There is one player in this war smarter than Stripe.

It doesn't pick sides, doesn't fight, and barely speaks. But no matter who wins, it wins steadily.

This player is called Circle.

Circle issues a stablecoin called USDC, currently the most compliant digital dollar globally.

Another company, Tether, issues USDT which has a larger scale, but regulators have argued for years over whether its reserves are sufficient and its reliability audited. Retail investors might not care, but in the AI world, where there might be hundreds of thousands of automatic transactions daily, each one needs to be auditable. No serious company would dare build its AI transactions on a stablecoin with questionable compliance.

Circle? A company listed on the NYSE. The U.S. SEC has seen its books, it discloses quarterly financial reports, and the whole world can see how much of its reserves are U.S. Treasuries and how much is cash.

So you see an interesting situation: Stripe's stablecoin financial accounts support USDC. OpenAI uses USDC through Stripe. Coinbase in Google's camp also connects to USDC.

The two camps are fighting bloody battles over the "entrance"—who controls the interface and protocol for AI spending. But no matter who controls the entrance, the money ultimately needs to be converted into stablecoins to run on the chain. And in the compliant stablecoin market, USDC has almost no rivals.

The two camps fight over the entrance; Circle gets the settlement volume.

Look at some data. In 2024, the global stablecoin transfer volume reached $15.6 trillion. To put that number in perspective: it's roughly equivalent to Visa's annual transaction volume.

A thing born less than a decade ago has already caught up with the network Visa built over sixty years.

And AI transactions are just beginning. Consulting firm Edgar Dunn & Co. predicts that by 2030, AI-driven transactions will reach $1.7 trillion. Every single one of these $1.7 trillion transactions will most likely go through the stablecoin pipe.

U.S. Treasury Secretary Scott Bessent said publicly during a Senate hearing in June 2025 that a stablecoin market capitalization reaching $2 trillion was a "very reasonable expectation".

Patrick Collison himself said: The average interest rate on U.S. bank deposits is only 0.40%, and $4 trillion in bank deposits even earn zero interest.

He believes this consumer-unfriendly approach is a "loser strategy," and young people will eventually move their money to higher-yielding stablecoins.

He was talking about the trend. And Circle happens to be standing right in the middle of that trend.

Epilogue

Finally, let's zoom out the lens.

On the surface, this battle over AI payment standards is about two commercial camps fighting for territory. But what it reflects is a deeper question: when AI starts participating in economic activities independently, is the financial system we designed for humans still sufficient?

Patrick Collison sees one future, where AI Agents are the main participants in economic activity. They compare prices, procure, pay, settle—the entire process doesn't require a human to press any button. This is the ultimate efficiency, but also the boundary of risk.

The alliance of Google and traditional finance sees another future: AI should be grafted onto the existing human financial infrastructure, constrained by existing human regulatory rules, and operate within the existing human trust framework.

Two futures, two logics, two camps.

But no matter which future arrives, one thing is certain: AI needs to spend money, money needs to run on the chain, and on-chain settlement requires stablecoins.

So Circle keeps winning. Stripe and Google keep fighting. Regulators keep chasing. Merchants keep integrating. Consumers continue not knowing which pipe their money actually flowed through.

Until one day, something goes wrong with what the AI bought for you, and you find that no one, not even the AI, knows who to ask for a refund.

On that day, everyone will suddenly remember the questions no one answered today.

But by that day, the pipes will already be built, and the tolls will already be collected.

History is always like this: get on the bus first, then buy the ticket.

Only this time, the bus is moving too fast.

相关问答

QWhat are the two main AI payment protocol alliances mentioned in the article, and who are their key members?

AThe two main alliances are Google's 'AI Agent Payment Protocol' (AP2) with over 60 members including Mastercard, PayPal, and American Express, and the Stripe-OpenAI alliance's 'Agentic Commerce Protocol' (ACP) with members like Microsoft Copilot, Anthropic, and Perplexity.

QWhat was the significance of Stripe's acquisition of the company Bridge in 2024?

AStripe's $1.1 billion acquisition of Bridge, a stablecoin payments company, was a key strategic move. It gave Stripe the technology to enable businesses to hold stablecoin balances and make global payments, forming a foundational piece of its end-to-end AI payment infrastructure.

QWhat is the Tempo chain, and what are its claimed advantages for payments?

ATempo is a new blockchain co-incubated by Stripe and Paradigm specifically designed for payments. Its claimed advantages include the ability to process tens of thousands of transactions per second, sub-second confirmation times, transaction fees of less than $0.001, and fees denominated in dollar stablecoins instead of a volatile native token.

QWhy is the company Circle considered a 'winner' regardless of which alliance succeeds in the AI payment war?

ACircle is considered a winner because it issues USDC, the most compliant and trusted digital dollar stable币. Both the Google and Stripe alliances rely on stablecoins for settlement, and USDC is the primary choice for major, regulated companies, meaning Circle benefits from the transaction volume no matter which payment 'entrance' is used.

QWhat core philosophical difference does the article highlight between the two approaches to AI payments?

AThe article highlights a difference in vision: Stripe's approach envisions a new, AI-native financial infrastructure built from the ground up for autonomous AI agents. Google's alliance aims to嫁接 (graft) AI onto existing human-designed financial infrastructure, governed by current regulations and trust frameworks.

你可能也喜欢

生成模型也能端到端训练了?核心竟是一个for循环

长期以来,深度学习在图像分类、检测等任务上凭借端到端训练取得了巨大成功,但生成模型领域却始终是个例外。当前主流的自回归、扩散模型等训练时只学习预测“一小步”,推理时则需要反复展开数百上千步,这种训练与推理的不一致导致了“暴露偏差”等问题,使得生成模型一直未能实现真正的端到端训练。 最近,一篇来自UIUC与哈佛大学的论文提出了“探索式建模(Explorative Modeling,XM)”新范式,其核心是一个简单到近乎朴素的for循环:在每个训练步骤中,模型生成K个候选样本,然后仅选取其中最接近真实数据的一个样本来计算损失和回传梯度。这种做法巧妙地绕开了传统生成模型中因使用重构损失(如平方误差)而导致的“模态模糊”问题——即模型倾向于输出多个可能答案的平均值,而这个平均值往往不属于任何真实的数据模式。 XM通过这种方式,在不拆分生成过程的前提下,直接提升了模型的“生成表达力”,即模型捕捉数据分布中多个独立模式的能力。论文将“探索”验证为继模型参数量、数据量之后的第三根可扩展轴。实验表明,随着模型规模、数据量和计算量的增大,探索带来的性能收益越发显著,在图像、视频、语言等多种任务上均能带来效率与质量的提升。 更重要的是,当探索程度足够时,XM可以实现真正的端到端生成。在机器人控制等任务中,仅需一次网络前向传播的XM策略,其性能可媲美甚至超过需要上百次前向的扩散模型策略,实现了推理效率的飞跃。 尽管“最佳K样本选择”的思想并非全新,但该研究的贡献在于清晰地揭示了这一简单机制如何直接解决生成模型的核心瓶颈,并为实现高效、端到端的生成模型开辟了新的可能性。随着模型规模的持续扩大,探索这一新维度可能将发挥越来越关键的作用。

marsbit8分钟前

生成模型也能端到端训练了?核心竟是一个for循环

marsbit8分钟前

年薪百万抢电工,Meta急到自己办技校

AI竞赛正面临新的瓶颈:工地。美国正面临严重的电工、建筑工等技术工人短缺,这已成为微软、Meta、OpenAI等公司快速建设超大规模AI数据中心(如OpenAI耗资160亿美元的“星际之门”项目)的“头号障碍”。 尽管AI公司愿支付高薪(如电工年薪可达24-28万美元),远超传统行业,但熟练技工仍供不应求。麦肯锡预测,美国在2023-2030年间需额外培养13万名电工和24万名建筑工,而劳工统计局预计每年仍有8万个电工岗位空缺。这种短缺导致项目延迟,每月可能造成数百万美元的收入损失。 AI数据中心建设复杂,需应对巨大功耗(一座设施耗电堪比数十万户家庭)、复杂的配电系统以及高密度散热(需液冷技术)等挑战,因此亟需大量技术娴熟的工人。 为此,科技巨头开始亲自下场培养人才。例如,Meta投入1.15亿美元建立建筑工人培训学校,提供免费培训及生活补贴,以快速输送工人上岗。OpenAI则与建筑工会合作,提前锁定熟练劳动力。同时,企业也将招聘目光投向高中生,鼓励年轻人投身技工行业。这些举措已见成效,Z世代对技工职业的兴趣显著上升。 然而,更深层的挑战在于电力。AI数据中心用电量正以惊人速度增长,已推高部分地区的电价。此外,数据中心建设是项目制的,建设期需要成千上万的工人,但建成后仅需少量常驻人员。这意味着未来可能面临熟练工人短期过剩、并流向其他行业压低薪资的风险。如何实现劳动力与电力资源的长期平衡,仍是悬而未决的问题。

marsbit1小时前

年薪百万抢电工,Meta急到自己办技校

marsbit1小时前

交易

现货
活动图片