Amazon Web Services CEO Matt Garman said AWS AgentCore Payments is now officially live. The managed payments module integrates Coinbase and Stripe and is designed to give AI agents a controlled way to make payments. According to the announcement, the service can be used to pay for APIs, content, data sources, and even other agents. It also supports spending limits, adding a layer of control over how agents handle payment flows. The update ties AWS infrastructure to both traditional online payments through Stripe and crypto-related payment rails through Coinbase, focusing on operational controls for AI-driven transactions rather than open-ended autonomous spending.

Andreessen Horowitz’s New Media team used its latest Charts of the Week to examine the AI infrastructure trade through a wider lens than headline demand growth. Moses Sternstein focused on neocloud companies such as CoreWeave, Nebius, and Applied Digital, arguing that the market’s question is no longer whether AI needs more compute, but whether providers can turn that demand into durable cash flow. The piece says many neocloud players entered the AI cycle with an advantage built during the crypto mining era: power access, data center capacity, cooling systems, and experience running dense compute loads. That legacy helped them scale revenue quickly, with CoreWeave reaching $2.6 billion in revenue in about 25 quarters versus 40 quarters for AWS after launch. Still, investors have not rewarded growth evenly. Over the past year, CoreWeave shares were down about 16%, while Nebius stayed closer to prior highs, highlighting concerns over capital intensity, depreciation, and rising interest expense. Sternstein also argues that AI is reshaping software unevenly rather than destroying SaaS across the board. Atlassian’s cloud revenue rose 31% year over year, and customers using its AI assistant Rovo were spending at nearly twice the growth rate of non-Rovo users. Databricks, meanwhile, said its Smart Router can cut average task costs by more than 30% by matching tasks with different model tiers. The article closes with data on widening enterprise AI spend gaps and diverging hiring patterns at OpenAI and Anthropic.

U.S. stock index futures and a range of AI-linked hardware names moved lower in premarket trading on Aug. 18, with weakness showing up across storage and optical communications stocks. According to BIT (Bit.com) market data, the Nasdaq fell 0.32%, the S&P 500 lost 0.52%, and the Dow slipped 0.51% before the opening bell. Among individual names, SanDisk, Seagate Technology, Western Digital, Micron Technology, and SK Hynix all declined, while Corning, Coherent, Marvell Technology, Lumentum Holdings, and Nokia also traded lower. The move came as markets revisited Anthropic’s reported revenue trajectory. Bloomberg had previously reported that the company’s annualized revenue run rate reached about $65 billion as of the end of July. That figure landed below some third-party data points and optimistic expectations in AI circles that had pointed to more than $80 billion, prompting questions about whether the pace of growth is slowing. Part of the dispute centers on ARR, or annual recurring revenue. ARR annualizes current revenue pace and is not the same as audited full-year revenue. Sacra said Anthropic’s annualized revenue was about $47 billion in May and rose to $65 billion in July, while also noting that revenue from cloud channels including AWS, Google, and Microsoft may be recognized on a gross basis, a factor that can make the revenue scale appear larger and draw more scrutiny to margins and revenue quality.

A viral meme titled "The SF Social Contract" has become a sharp shorthand for how money moves through San Francisco’s AI startup economy. Originally created by X user @chiefofautism and later unpacked by BlockTempo, the chart traces venture capital from startup funding rounds into model and API bills, cloud and GPU costs, Bay Area rents, high-cost professional lifestyles, and finally acqui-hire deals by large tech companies. The framing is satirical, but the article argues that the underlying cash loop is real. BlockTempo cites 2026 first-half U.S. venture investment of $412.7 billion, with 86% flowing to AI. OpenAI and Anthropic alone accounted for $217 billion, according to the article. It also says AI startups are spending 40% to 50% of revenue on APIs and inference, sending a large share of operating cash back to the model providers they depend on. Even those providers face heavy infrastructure costs: Anthropic’s annualized compute spending in mid-2026 was estimated at $4 billion to $4.5 billion, or about 60% of annualized revenue, while OpenAI’s ratio was said to exceed 75%. The piece then shifts to Bay Area landlords, citing California’s Proposition 13 and San Francisco rent data from Zumper, before turning to the meme’s dating-and-lifestyle stereotypes and the final exit point in the loop: acqui-hire structures used by Google, Meta, Microsoft, and others to bring in AI founders and research teams without conventional acquisitions.

Anthropic has signed a 20-year AI data center compute hosting agreement with Riot Platforms worth $9.1 billion, according to Bloomberg. Riot said it will provide 191 megawatts of capacity for Anthropic’s growing Claude workloads, with the first 96 megawatts expected to come online in December next year and full deployment scheduled for June 2028, based on SEC filings. The news sent Riot shares up more than 25% in after-hours trading. The deal highlights how listed bitcoin miners are being pulled into the AI infrastructure buildout as large model developers look beyond traditional cloud providers and move to lock in long-term power, land and compute capacity. The article says Anthropic has been assembling that network through long-dated build-to-suit agreements, alongside partnerships involving AWS, Google Cloud, TeraWulf, Volta Infra Holdings, xAI, Hut 8 and Fluidstack. PANews argues Riot won the contract because of three factors already in place: energized power access at its Rockdale, Texas site; execution capability shown in its earlier data center work with AMD; and a balance sheet that includes more than $1.2 billion in liquid assets, about $549 million in cash, 11,380 BTC and a $573 million transitional financing agreement with Morgan Stanley. The report says the transaction may push investors to value some miners less like high-beta crypto proxies and more like infrastructure operators with long-term contracted cash flow.

A commentary published by Foresight turns the current wave of AI-agent infrastructure into a set of direct tests rather than a sales pitch. It opens with a simple problem: major firms are now giving software agents the ability to hold stablecoin wallets, operate under spending budgets, and use payment protocols. Cloudflare has rolled out a stablecoin wallet for AI agents, Coinbase has integrated x402 and stablecoin wallets into AWS agent environments with enterprise budget and governance controls, and Google has released an agent payment protocol. Once software can place orders, sign deals, and move money, the question is no longer whether the tooling works. It is who carries the consequences when something goes wrong. The article then shifts to real-world assets and argues that the recent growth in onchain RWA is real but concentrated in financial products, not industrial assets. As of Aug. 7, 2026, total onchain RWA market value stood at $37.94 billion, with 1.6294 million holders, up 55.34% from the prior month and more than 560,000 net new holders in a single month. Yet the assets driving that growth are Treasuries, money market funds, private credit, gold, fund shares, and equities. Machine tools, production lines, and supply-chain orders are still largely offchain. The final challenge is social scoring: if influence can be priced and AI agents can work around the clock to build relationships, does the resulting score still measure the person, or the machine working on that person’s behalf? The piece leaves those questions open and ties them to an Aug. 22 event at Suzhou University of Science and Technology.

Citrini analyst Jukan said a recent overseas electronics report from Guosen Securities viewed Intel’s $20 billion equity offering as a positive signal, pointing to direct participation by CEO Chen Liwu and his family of about $12 million. Guosen reiterated a Buy rating and a $136 target price, while lifting its 2026 and 2027 EPS forecasts by 3% and 1%, respectively. The report said the offering was upsized from an initial $15 billion to $20 billion, priced at $95 per share, with the greenshoe fully exercised and reported institutional demand exceeding $100 billion. Guosen also said management participation could support Intel’s 2027 capital spending plans. On operations, the report projected Intel’s foundry business would break even in the fourth quarter of 2027, with margin leverage expanding in 2028. It also cited an estimated 18A yield of about 80%, volume ramp progress for Clearwater Forest, and potential high-volume progress for Apple 14A. In advanced packaging, the report said Intel’s EMIB customer base continues to widen, with AWS Trainium3 expected to adopt EMIB-T in 2027 and additional programs from Google, AWS and Microsoft possibly contributing from late 2027 into 2028.

Altimeter Capital partner Clark Tang argues that Nvidia is no longer just a chip supplier to the AI industry. In his view, the company has been building the two pillars that historically defined hyperscalers: an operating layer that abstracts and manages infrastructure, and a financing layer that funds capacity ahead of demand. Tang says this combination is turning Nvidia into a “synthetic hyperscaler,” one that is starting to displace Amazon, Microsoft, and Google in parts of the AI compute supply chain. His thesis begins with a shift in infrastructure economics. Traditional hyperscalers built strong margins by converting enterprise capex into opex and using software to maximize utilization of shared hardware. Tang says AI workloads break that model. Large-scale training depends on tightly synchronized GPU clusters, while inference is highly sensitive to tokens per watt and time to first token. In that setup, virtualization and networking layers that worked well in the cloud era can become a drag on GPU performance. He also points to the rise of neocloud providers, which offer lower-margin, AI-focused infrastructure but often lack the balance sheet strength to finance aggressive buildouts. Tang says Nvidia has moved to close that gap with software such as DSX OS, Mission Control, Omniverse, and Dynamo, while also standardizing hardware and bringing in third-party capital from firms including Apollo, BlackRock, Blackstone, Goldman Sachs, and KKR.