The Daily Brief · Issue 10 · 27 August 2026Nvidia posts $96.2 billion in quarterly revenue and forecasts $108 billion next, as Anthropic signs a $45 billion cloud deal
Nvidia's results showed AI spending still accelerating, and Anthropic kept buying capacity ahead of its IPO. Z.ai revealed that the viral Ox Alpha was its GLM-5.3-Flash, Qwen previewed its Qwen4 design, and Salesforce plugged its CRM straight into Claude.
EarningsNvidia revenue hits $96.2 billion, up 106% year on year, with $108 billion guided next
Nvidia reported revenue of $96.2 billion for the quarter ending 26 July 2026, up 18% from the previous quarter and 106% from a year earlier. Data-centre revenue, mostly AI chips and systems, was $89.0 billion, up 117%. Gross margin was 75.0%. Nvidia expects about $108.0 billion next quarter, plus or minus 2%, without assuming any data-centre compute sales to China. It said Vera Rubin systems are in full production at partners including CoreWeave, Google Cloud, Microsoft Azure and Oracle. CEO Jensen Huang said many frontier labs are now scaling in parallel, not just one.
Why it matters: AI hardware spending is still doubling year on year, so fears of an imminent slowdown are not showing up in the numbers yet.
Source: NVIDIA
InfrastructureAnthropic signs a roughly $45 billion cloud deal with Nscale
Anthropic agreed a cloud deal worth about $45 billion with Nscale, a UK-based AI infrastructure company, CNBC confirmed. Sources say Anthropic will rent around 460 megawatts of computing capacity at an Nscale data centre being built in West Virginia, running Nvidia's Vera Rubin chips, with the site expected online at the end of 2027. Anthropic said earlier this year that demand had caused 'inevitable strain' on reliability at peak times, and it has since signed capacity deals with AMD, SpaceX, Google and Broadcom. It is valued at $965 billion and has filed confidentially for an IPO.
Why it matters: More capacity should mean fewer Claude slowdowns and usage limits, though this site will not help until late 2027.
Source: CNBC
Model launchZ.ai reveals the viral Ox Alpha as GLM-5.3-Flash
Z.ai confirmed that Ox Alpha, the anonymous model that topped usage charts on OpenCode and OpenRouter, is GLM-5.3-Flash. It is the first natively multimodal model in the GLM-5 series (it handles images as well as text) and has 320 billion parameters in total. Z.ai claims it beats GLM-5.2 on several benchmarks at a tenth of the price and approaches Claude Opus 4.8 on coding and agent tests. Z.ai says all the preview traffic was served on Chinese AI chips. OpenRouter now lists the model under its real name.
Why it matters: A cheap Chinese model served on Chinese chips won millions of developer sessions before anyone knew who made it.
Source: The Deep View
Enterprise AISalesforce and Anthropic launch Claudeforce, a CRM plugin for Claude
Salesforce and Anthropic launched Claudeforce, a plugin that lets Claude users work with Salesforce data directly. It ships with 37 pre-built sales skills for tasks such as writing emails and updating customer records. It is the first time Salesforce has put its 'force' suffix on another company's product. Dario Amodei said the firms are building 'Enterprise Frontier Safeguards' to keep customer data private. Claudeforce is with pilot customers now, with a preview next month and more skills later this year. Salesforce shares rose 12% after hours, alongside better-than-expected results.
Why it matters: Instead of fighting AI assistants, Salesforce is plugging into them, a sign of how business software may be used from now on.
Source: CNBC
Open weightsQwen3.8-Flash-Next previews the Qwen4 design: 125B parameters, only 6B active
Alibaba's Qwen team released Qwen3.8-Flash-Next as open weights, a preview of the architecture it plans for Qwen4. It has 125 billion parameters but uses only 6 billion for each token, about a third of the active compute of Qwen3.7-Plus (397B total, 17B active). Changes include a new sparse attention scheme and 51 billion parameters held in a separate lookup table keyed by two- and three-character text fragments, which is cheap to compute and easy to move onto memory-limited chips. The Next Web notes the benchmark numbers are Alibaba's own and that the licence may not qualify for the EU AI Act's open-source exemption.
Why it matters: Chinese labs are designing around chip export limits, which pushes them towards models that are cheaper to run everywhere.
Source: The Next Web
DealsGoogle reportedly in talks to pay over $1.5 billion for AI coding start-up Mechanize
Google is in advanced talks on a deal worth more than $1.5 billion to license technology from Mechanize and hire some of its staff, according to Business Insider as reported by gHacks. Talks were ongoing and terms could change. Mechanize, founded in 2025 by Tamay Besiroglu, Matthew Barnett and Ege Erdil, builds training environments, benchmarks and data to help AI agents handle complex software work. The structure mirrors Google's roughly $2.4 billion Windsurf licensing deal in July 2025, which brought that team into Google DeepMind.
Why it matters: Training environments for coding agents are now worth billions, showing where labs think the next gains will come from.
Source: gHacks