Learning about US stocks Day 2: Why are there only a few winners when everyone is selling shovels? A couple of days ago, we talked about the people selling shovels; today I want to think through this a bit more. Because when you actually draw out an industry chain, you find a very realistic problem: there are too many people selling shovels. AI chips are being developed by Nvidia, AMD, Intel, and even Google and Amazon are doing in-house development. Servers are made by Dell, Super Micro, Lenovo, and many others. Optical modules, liquid cooling, data centers—every segment is crowded with companies. Logically, the AI boom's dividends should be shared by everyone. But in reality, the biggest profits are concentrated in very few companies. Why? Let's take Nvidia as an example. Many people think Nvidia wins because its GPUs are good. The hardware is indeed strong, but if you only compare hardware, AMD's chips aren't necessarily much worse. What really sets Nvidia apart from everyone else is CUDA. CUDA is a software development platform Nvidia launched back in 2006. At that time, most people still thought graphics cards were just for gaming, but Jensen Huang had already started using GPUs for scientific computing and data processing. Later, when AI exploded, the whole world found this tool perfectly usable. Over more than a decade, almost everyone working on AI—writing code, training models, tuning parameters—runs almost entirely on CUDA. Various toolkits, optimization frameworks, tutorials, and documentation have all laid the foundation of AI development on CUDA. So the real issue isn't whether AMD can make good GPUs. It's that if a company wants to switch from Nvidia to AMD: code has to be rewritten, toolchains replaced, teams retrained, and previously tuned models might have to be redone. The migration cost is so high that almost no company is willing to bear it. What Nvidia sells is not just chips, but an ecosystem you can't live without. This is the moat. Simply put: others want to do the same thing but can't, or even if they do, they can't catch up. It's not that competitors can't enter, but if they do, they can't beat you. This kind of thing exists in every industry: Apple. iPhone, App Store, iCloud, Apple Watch form a complete ecosystem. If you switch to Android, you have to transfer photos, repurchase apps, miss messages, and your watch might not connect. Just thinking about it is a headache. Users don't want to leave, but they can't. Coca-Cola. You can replicate the exact taste in a lab, but you can't sell it at that price. Because people drink Coca-Cola not just for the sugary water, but for the brand trust built over more than a century. Visa. Tens of millions of merchants worldwide accept Visa cards, and billions of users use them for payments. The more merchants, the more users want to use it; the more users, the more merchants have to accept it. Once this flywheel starts spinning, new payment networks simply can't break in. ASML. The only company in the world that can make EUV lithography machines. This isn't something you can catch up to just by throwing money at it—it requires decades of technological accumulation plus the coordination of a global supply chain. High-end chips from TSMC, Samsung, and Intel all have to buy from it. Notice? These companies don't lack competitors, but their competitors find it very hard to truly threaten them. Buffett once said he likes companies that can make money even if run by a fool. Because if the moat is deep enough, you don't need a genius CEO to survive. Truly great companies maintain their competitive advantage even with an ordinary manager. So how to judge if a company has a moat? I summarized a very simple test: If this company raises prices by 20% tomorrow, will customers leave or just endure it? If customers can't go anywhere else, then this company's moat is probably really deep. Of course, moats aren't permanent. Kodak once had a brand moat, Nokia once had a market share moat, but times changed and moats can dry up. Finding a moat is just the first step; you have to keep watching if it's still there. Industry trends determine how big the pie is; moats determine who gets the biggest slice. Next time you see a hot sector, don't just ask if the industry will explode. Ask one more question: If the industry really explodes, who has the deepest moat? Who is the irreplaceable link? $XNVDA $XAAPL $XASML

OKX.AI Trading Hackathon Season 1: This time, let the Agent prove itself with real trading The last OKX.AI Genesis Hackathon focused on Agent products and real usage. This time, it’s all about real action—trading with real money in the market. Participants don’t submit a demo or a video; they develop a trading ASP that connects the Agent to a wallet or trading account to run strategies with real funds. Rankings are based on return rates—the more profit, the higher the rank. Trading ASP, the first challenge The name "Season 1" is interesting. Since it’s called Season 1, there’s a high chance there will be Season 2, Season 3, and so on. This season’s theme is trading, testing whether Trading ASPs can read market data, manage funds, execute strategies, and deliver results in real markets. Will there be future seasons for research analysis, on-chain data, content creation, payments, or even gaming and social ASPs? It’s uncertain. But from a platform development perspective, adding different ASP categories season by season to the ecosystem is a very reasonable path. So the significance of this competition might not just be to pick a few profitable Agents, but more like OKX.AI screening the first batch of trading ASPs that can truly go live. It’s not about submitting strategies, it’s about running them. Previously, in quantitative competitions, you had to write your own code, connect APIs, rent servers, and handle market data, accounts, orders, and risk control all by yourself. This time, OKX.AI integrates all these steps into the Agent system. Your trading logic registers as a Trading ASP, links to a wallet or trading account, and the Agent can trade automatically during the competition. Registration closes at 12:00 noon on August 11, and the competition runs from August 11 to August 25, two weeks, UTC+8. The leaderboard ranks by PnL%, i.e., return rate. This time, it’s not about how good your PPT is or how pretty your interface looks; it’s about whether your Agent truly understands the market and executes strategies to make money while controlling risk. This is the most direct way to validate a trading ASP. Two paths, think carefully before choosing The competition offers two participation methods; you choose one when registering and cannot change it after submission. First: Onchain OS. Link Agentic Wallet and trade on-chain via Onchain OS. The asset scope is more open—token rotation, trend following, arbitrage, etc. If you know the on-chain market better or want the Agent to manage on-chain assets directly, choose this path. Second: Agent Trade Kit. Link OKX UID and trade via CLI or MCP, mainly tracking USDT perpetual contracts. This suits strategies like trend, grid, moving averages, funding rates, and long/short signals. Trades done by other methods may not count toward rankings. Prizes and exposure Prize pool: $20,000. 1st place: $5,000 2nd place: $4,000 3rd place: $3,000 Prizes also for 4th to 30th place. Top three also get official OKX social media and product-level exposure. For those building trading Agents, quantitative strategies, on-chain tools, or signal services, this exposure might be as valuable as the prize money. After the competition, strategies don’t retire; they can continue as Trading ASPs providing services for others to use or subscribe to. The end of the competition might just be the start of commercialization. What you need to prepare for registration The process isn’t complicated: Install Agent tools (OpenClaw, Hermes, Claude Code, Codex, or use third-party cloud hosting) → connect to Onchain OS → log in to Agentic Wallet via email → register Trading ASP → link wallet or trading account. For funds, prepare no less than 300U equivalent assets. This is not a registration fee but the initial capital for the Agent’s real trading. Real trading means real risk of loss. Before competing, it’s best to test in a simulation or with small amounts to check position management, stop loss, and Agent logic under abnormal market conditions. Don’t max out leverage just to climb the leaderboard, and don’t risk all your assets on an untested strategy. The last Genesis Hackathon tested whether Agents could solve real needs and had users. This one tests whether trading ASPs can manage money, run strategies, and survive in real markets. From helping users complete tasks to helping users execute trades, OKX.AI is pushing Agent capabilities toward more real economic activities. If you’ve completed ASP registration, just send your Agent: "Help me register for OKX.AI Trading Hackathon" If not, refer to: https://www.okx.ai/zh-hans/hackathon

Good morning, esteemed traders Today, while checking the Creator Center, I noticed that the Planet livestream weight has been increased For those who haven't applied for livestream permissions yet, you can apply via the link below https://www.okx.com/zh-hans/livestream/interest-form Also, I want to make a suggestion: Many people are a bit confused about Planet earnings and want to know how the weekly earnings ratio is calculated. The last rules were from a few months ago, and I personally feel they should have changed by now, but there is no related documentation available. @佳佳OKX @OKX中文 @你的爱播Misa @OKX星球 @OKX小天 @可乐Cola_OKX @米妮Minnie_OKX @米花Lilac_OKX

The earnings reports of six tech giants are all out, with only Nvidia left, expected in August. Currently, the results of these companies are quite interesting: Microsoft and Amazon surged; Google, Tesla, Meta, and Apple declined. Amazon has the highest revenue, Google’s cloud business grows the fastest, but the stock price reactions are completely different. The market’s attitude toward AI investment has changed. A year ago, as long as a giant announced increased AI investment, the stock price would rise first. Now the market is starting to check the books: Has the money spent actually turned into revenue and cash flow? So this round of earnings reports, on the surface, is about performance, but in reality, the market is regrading the growth quality of these six companies. In this lesson, we’ll score them: two top performers, two somewhat unfairly treated, and two with their own troubles. Two that rose: Money was spent, and it proved to be profitable Microsoft: AI has started generating rent Revenue $90 billion, up 18% year-over-year; Azure grew 43%, Microsoft Cloud grew 27%. More importantly, next quarter Azure guidance is about 45%, commercial contract liabilities $678 billion, up 84% year-over-year. The logic behind Microsoft’s rise can be summed up in one sentence: they dare to spend money and can prove that the money spent is making money. Enterprises want to use AI, so they buy Azure computing power, databases, Microsoft 365, Copilot—Microsoft isn’t just selling a model, but repeatedly charging across the entire enterprise AI chain. Plus, capital expenditure and cash flow expectations are not as bad as the market feared, so investors are willing to give a higher valuation. After the earnings report, the stock rose over 15%, adding nearly $450 billion in market value in one day. Amazon: AWS accelerates, proving computing power is truly being bought Sales up 20%, operating profit up 43%; AWS revenue $42.2 billion, up 37% year-over-year, the fastest in over four years; advertising also rose 26%, reaching $19.8 billion. What was the market’s biggest fear before? That Amazon would spend over $200 billion a year building data centers, only to have customers taken by Microsoft and Google. This time AWS suddenly accelerated, telling everyone: the new computing power is indeed being purchased, and supply can’t keep up with demand. So even with an annual investment plan mentioning $220 billion, the market accepted it. The stock rose over 12% pre-market after the earnings. Spending big money, Amazon and Google’s stock prices moved in opposite directions. The difference is: one proved returns with growth, the other exposed greater cash flow pressure. Two somewhat unfairly treated: Great performance, but burning cash too fast Google: Did very well, but the parent was scared by the bill Looking at the results alone, Google might be the best among the six: revenue $119.8 billion, up 24%; search up 17%; Google Cloud surged 82%, cloud business operating profit $8.8 billion. But the stock price fell. The reason is capital expenditure was raised again: from $180-$190 billion in 2026 to $195-$205 billion, with a clear increase expected in 2027, and about $5.9 billion cash consumed in Q2 already. The market’s thinking is easy to understand: Cloud grew 82%, yet cash flow is so tight, how much more investment will be needed? Plus, Google’s stock had already risen a lot with high expectations, so "good" is no longer enough; it has to be "better than the most optimistic expectations." Once spending is raised, investors take profits first. Meta: Advertising is strong, but cash flow is being rapidly eaten by AI Previously discussed Meta separately, here’s a brief recap. Revenue $60.8 billion, up 28%, ad impressions up 14%, price per ad up 12%, solid fundamentals. But costs rose 55%, EPS $6.18, below the expected $7.22; most strikingly, free cash flow dropped from $8.55 billion to $784 million, shrinking by about 90%. The money earned from advertising is being quickly consumed by AI infrastructure. And new businesses like AI assistants and computing power rentals don’t yet have clear revenue scale. The market wants to see a clear monetization path before it’s willing to add back valuation. Two with their own troubles Tesla: Selling more cars, but earning less per vehicle Delivered over 480,000 vehicles in Q2, deployed 13.5 GWh of energy storage, sales are back. But to boost volume, they continued price cuts and discounts, vehicle revenue per unit dropped from $45,345 to $42,730, operating profit only about $400 million, profit margin 1.4%, significantly below expectations. Tesla’s current high valuation relies on Robotaxi, FSD, and Optimus robots, not car sales. These new businesses are not rolling out fast enough, so the market can only look back at the car business — and car profits are declining, AI requires heavy investment, so the stock can only fall. Apple: Beat expectations this quarter, but the market focuses on next quarter Revenue $109.4 billion, up 16%; EPS $2.02, up 29%; iPhone revenue $54.25 billion, up 22%. The quarterly results are not bad, but problems lie in two areas. First, about two percentage points of gross margin this quarter came from tariff refunds, and EPS includes about $0.11 of this "extra" income — meaning part of profit growth is not from core business improvement. Second, next quarter guidance: revenue growth 9%-11%, below market expectation of 12%, and management mentioned tight supply of high-end chips, possibly limiting iPhone, Mac, and iPad shipments. Demand remains, but supply can’t meet it, so that demand temporarily doesn’t convert into revenue. Plus, Apple’s stock has risen a lot this year with a high valuation, so slightly missing guidance leads to capital pulling out first. The stock fell about 7% pre-market after earnings. Microsoft and Amazon: Spend a lot, but earn faster, so they rise. Google and Meta: Strong revenue, but cash burn is too large, valuation is suppressed. Apple and Tesla: One is stuck on next quarter guidance and supply, the other on profit margin. Looking at earnings reports is never just about how much revenue grew, but how actual results compare to market expectations. Microsoft proved AI can already generate orders and cash flow; Amazon proved there is demand for expanded computing power; Google and Meta have equally excellent growth, but investment scale exceeds market’s psychological tolerance. #财报观察员：亚马逊指引不及预期，股价却反涨9%

Apple fell, Amazon rose: Two earnings reports, one lesson This earnings season, it's particularly interesting to look at two companies side by side: Apple's earnings were good, but its stock price fell; Amazon's cash flow was negative, yet its stock price rose. Apple: Did well this time, but forecasted to do worse next time Let's first look at Apple's report card. Revenue was $109.4 billion, up 16% year-over-year; net profit was $29.8 billion, up 27%. Breaking it down, iPhone revenue rose 22%, Mac rose 29%. This is a very impressive result. Both phones and computers sold very well, and the company remains highly profitable. So the question arises: with such good results, why did the stock price fall? Investors look at earnings reports not just to see how much was earned in the past three months, but more importantly, whether growth can continue. To give an analogy: a student scored 90 this time, which is very good. But if he tells his parents, "Next exam, I’ll probably only score just over 80," how would the parents feel? Certainly not happy. Apple did exactly this: it forecasted next quarter revenue growth of 9%–11%, below market expectations. The services business was not as strong as imagined, and AI has not yet brought significant revenue. So Apple's problem is not that it isn't making money now, but the market worries that its growth rate may slow down. Amazon: Spending aggressively, but already seeing returns Now let's look at Amazon. The focus of this earnings report is not online retail, but AWS. What is AWS? You can think of it as a computing power rental company. Many businesses don't want to buy servers or build data centers themselves, so they rent from Amazon. AI companies especially need computing power, so AWS naturally becomes the most direct beneficiary. Looking at the numbers: total revenue was $200.6 billion, up 20%; operating profit was $27.5 billion, up 43%. AWS revenue was $42.2 billion, up 37%—the fastest growth in the past 18 quarters. But Amazon also spends aggressively: this year it plans to invest $220 billion on chips, data centers, and AI. Because of heavy spending, free cash flow over the past 12 months was negative $7.6 billion. Negative cash flow, so why did the stock price still rise? Another analogy: someone spends a lot to build a factory, so cash is tight initially. But as factory orders increase and revenue grows rapidly, lenders won't panic because they can see the money coming back. Amazon is in this state now. AWS's 37% growth signals to the market that the money invested is already turning into revenue. Why does Amazon rise while Meta falls despite both burning cash? Yesterday we talked about Meta. Both Amazon and Meta are burning cash to build AI infrastructure, and both have declining free cash flow, but the market's attitude is completely different. What's the difference? Amazon spends money but can directly sell computing power to customers through AWS; customers pay for servers, cloud services, and AI usage as they go, so money comes back immediately. Meta? It makes money from Facebook and Instagram ads, then invests that ad revenue into AI models and data centers. AI can improve ad targeting accuracy, but the returns versus investment are not clearly accounted for. Amazon spends money and earns money simultaneously; Meta spends first and the future earnings are still uncertain. The market naturally prefers the former. Apple's problem is not lack of profit, but possibly slower future growth; Amazon's problem is heavy spending, but AWS has proven the spending is worthwhile. This principle applies to all companies: spending money is not scary; what's scary is spending money that doesn't bring back more revenue and profit. Borrowed a chart from @你的爱播Misa to use here

Microsoft up 6.6%, Meta down 6.7%: Two ways to spend the same money Last night, Microsoft and Meta both released earnings reports; after hours, one rose 6.6%, the other fell 6.7%. The trends were completely opposite, but the market was actually asking the same question: Money has been poured into AI, when will it pay off? Microsoft’s earnings answered this question. Microsoft: Money spent, invoices returned This quarter’s revenue was $90 billion, up 18% year-over-year; net profit was $35.77 billion, up 31%. But what really impressed the market wasn’t the big numbers, it was Azure’s 43% year-over-year growth beating expectations and accelerating from last quarter’s 40%. Capital expenditure for the quarter was $41 billion, a figure that would normally draw criticism. But this time no one complained, because the investment has turned into revenue: Copilot paid seats exceeded 30 million, and commercial remaining performance obligations (RPO) reached $678 billion, up 84% year-over-year. This last figure deserves extra emphasis. An RPO of $678 billion means enterprise customers are lining up to sign contracts, and Microsoft is holding a large amount of unrecognized revenue. In other words, Microsoft’s current problem isn’t insufficient demand, but insufficient capacity. This is a completely different issue from other companies burning cash on AI. The logic is straightforward: invest in GPUs, build data centers, expand Azure capacity, then sell computing power through cloud services and Copilot. As long as Azure’s growth doesn’t slow, the market is willing to tolerate this spending. However, profits need to be examined separately. This quarter includes $3.2 billion in Anthropic investment gains; excluding that, adjusted EPS was $4.74, still beating expectations, but operational growth isn’t as impressive as the GAAP 31%. Incidentally, Microsoft is both partnering with OpenAI and investing in Anthropic, and the money bet on both fronts is already generating book returns—this spending is much smarter than some companies’ past cash burns on the metaverse. Meta: Not unprofitable, but earnings can’t keep up with spending Meta’s problem is different. Revenue was $60.8 billion, up 28%; ad impressions rose 14%, and ad prices increased 12%, so the core business is still solid. The ugly side is on the other end: total expenses rose 55%, capital expenditure was $31.08 billion, and free cash flow dropped to $784 million, down 91% year-over-year. The 6.7% drop after hours reflects this. The 14% net profit decline includes $2.4 billion in litigation fees and $1.18 billion in layoff costs; excluding these, operating profit is actually growing. So this report can’t be read as a deterioration in the ad business. What the market fears is looking ahead: 2026 capital expenditure guidance is $130-145 billion, with total expenses of $165-169 billion. In other words, cash flow pressure isn’t just a one-quarter issue, but a problem for the next year or two. The difference is: one is starting to collect rent, the other is still building the building. Microsoft and Meta are both building AI infrastructure, but at completely different stages. Microsoft can directly sell computing power to Azure customers, charge Copilot monthly, and distribute through an existing enterprise customer system. The monetization path is closed-loop. Meta still relies on advertising revenue, then funnels cash into models, data centers, AI assistants, and glasses. AI has indeed improved ad recommendation efficiency and user engagement, but there’s no clear correlation between these gains and the $30+ billion quarterly spending. The market can’t price what it can’t see, so it just punishes the stock. One often overlooked point: Zuckerberg has absolute voting control and doesn’t have to bow to quarterly earnings. This means Meta’s heavy investment will last longer than the market expects. Buying Meta is essentially buying Zuckerberg’s patience and vision. He lost a bet on the metaverse in 2022 but later saved the stock price through efficiency gains. This time’s script is similar, just with stakes several times higher. Microsoft’s premium is certainty: high Azure growth, scaled Copilot billing, orders lined up for the next generation. But certainty comes at a cost—the valuation is packed with expectations, and if Azure growth slows or capital expenditure rises, the valuation will be harshly punished. A well-known good company has very little margin for error. Meta is the opposite: the ad base is solid, this drop has released valuation pressure, but the market needs to see stabilized cash flow and clear revenue sources from AI investments before it’s willing to reprice. Until then, every earnings report risks a hit from expense guidance. So Microsoft’s rise isn’t because the market likes it spending money; Meta’s fall isn’t because the market rejects AI. The difference in one sentence: Microsoft is already collecting rent from AI, Meta is still using ads to fund AI. My own inclination: for stability and mid-to-long term, Microsoft’s logic is much smoother; for those who can tolerate volatility and believe Zuckerberg can turn AI into ad efficiency and new hardware revenue, this drop is actually a good observation point—watch cash flow, not stock price. Last earnings season, the market was counting who bought how many GPUs; this round, it’s about who first turns investment into free cash flow. Yesterday’s rise and fall was the first ballot cast at this watershed. #财报观察员：微软Meta亚马逊今夜交卷

Why does OKX.AI need so many payment methods? An Agent store selling functions is just the first step. The hard part is keeping users coming back and paying continuously. A one-time purchase can’t achieve this; users spend 0.1 USDT to call an Agent once, get the result, and leave—just like buying a temporary tool. For the platform and developers, closing a single deal is nothing; the real skill is making users return continuously. Turning one-time customers into long-term customers is what makes the business viable. Simple tasks: pay once, deliver once; high-frequency data services: if each call is billed separately, friction costs are too high, so batch payments are more cost-effective; complex tasks: pay based on usage, which is fair; long-term tools: subscriptions are the most convenient; for large amounts and long cycles, without escrow and acceptance mechanisms, who dares to pay upfront? Therefore, OKX.AI offers five additional payment methods besides one-time payments. One-time payment: the simplest and most common. Pay once, buy the service once. For example, spend 0.1 USDT to let an on-chain analysis Agent check a wallet address; the Agent analyzes its holdings, transaction records, profit and loss, and risk tags, then returns a report, ending the transaction. Generating resumes, making images, checking data, analyzing projects, calculating a meal’s calories—all these scenarios fit this model. Fixed price, clear results, one-time completion—this is currently the easiest way for users to accept. Batch payment: solves high-frequency small calls where one-time payment is unsuitable. A quantitative trading Agent may check BTC and ETH prices dozens of times per minute. If each query costs 0.001 USDT and is settled separately, the Agent wastes a lot of time waiting for settlements. Batch payment means: use the service first, then settle the bill collectively. Real-time market data, on-chain data, high-frequency APIs, and automated trading all fit. In the future, Agents will interact thousands of times daily; it’s impossible to handle each transaction like a normal person’s transfer. Batch payments will become increasingly important. Pay-as-you-go: pay for what you use. Some tasks can’t be priced upfront. For example, a market research Agent investigating a project: checking the official website, reading the whitepaper, breaking down the token model, reviewing the team background, checking on-chain holdings, comparing similar projects. Well-documented projects are quick; messy ones require extensive info gathering and more API calls, with costs varying several times. Uniform pricing is unreasonable. A better approach is prepaid deduction: users deposit 5 USDT, and the Agent deducts based on actual calls, token consumption, or execution time. If only 2.3 USDT is spent, the remainder is refunded. This is like mobile data or cloud server billing—not paying for fixed results but for actual usage. Suitable for complex research and multi-step workflows that can’t be estimated in advance. Subscription payment: from selling tools to selling services. Subscribe for 20 USDT/month to a smart money monitor, continuously receiving alerts on large transfers, address changes, new coin purchases, holdings changes, and risk warnings during the subscription period—no need to pay per view. Developers also gain stable income. Many Agents eventually adopt subscriptions. One-time payments sell tools; subscriptions sell ongoing services: AI investment research, on-chain monitoring, paid databases, trading signals, API packages—all fit monthly or yearly fees. Note not to confuse subscriptions with pay-as-you-go: subscriptions charge a fixed monthly fee regardless of use; pay-as-you-go charges based on usage. The former suits stable long-term products; the latter suits services with fluctuating costs. Escrow payment: not launched yet, but the most imaginative. Escrow payment solves trust issues. For example, someone offers 1,000 USDT for a developer to build a Web3 data analysis website. The user fears paying upfront without delivery; the developer fears delivering without payment. Escrow locks funds in a contract; payment is released after delivery and acceptance; disputes go to arbitration. This is no longer buying a tool but a full on-chain outsourcing transaction system. My view on escrow payment: I believe it will become one of OKX.AI’s most important payment methods. Current ASPs mostly offer lightweight services: pay once, get data or content in seconds or minutes. But Agents’ capabilities are growing; future tasks may last hours or days: website development, industry research, long-term social media management, contract deployment, or multiple Agents collaborating on a project. Such work can’t be done with one payment; results must be accepted. So after escrow payment launches, it will likely cover: - Staged payments: for a 1,000 USDT project, 20% on prototype delivery, 50% on main feature completion, balance on acceptance. - Deliverable management: service providers submit code, reports, files, demo links—not just say "done." - Dispute arbitration: if parties disagree, platform rules, third-party arbitrators, or dedicated arbitration Agents decide. - Reputation system: how many orders a developer has taken, delivery speed, refund rate, and reviews become credentials. Once implemented, OKX.AI won’t just be a place to buy tools but a true on-chain service marketplace: post requests, take orders, deliver, accept, and auto-settle. One-time payment makes Agents simple tools; batch and pay-as-you-go push them to high-frequency calls and complex tasks; subscriptions provide steady income; escrow payment advances them to project-based collaboration. At this stage, AI Agents complete the business loop of working—delivering—getting paid. Payment methods documentation: https://web3.okx.com/zh-hans/onchainos/dev-docs/payments/methods-overview @Star_OKX @OKX中文 @你的爱播Misa @OKX星球 @OKX小天 @可乐Cola_OKX @米妮Minnie_OKX @佳佳OKX

The Planet Certification Logo has changed color