Nvidia's fiscal second-quarter results landed after market close on August 26 and dominated the August 27 AI tape. The company reported $96.2 billion in revenue, up 106 percent year over year, and $89.0 billion in data-center revenue, up 117 percent.
The company guided for $108.0 billion in third-quarter revenue, plus or minus 2 percent, while saying it was not assuming any data-center compute revenue from China in that outlook. It also highlighted Vera Rubin entering full production and multiple cloud partners running or preparing racks around the platform.
The numbers made the AI buildout feel less theoretical. Whatever skepticism investors have about long-term returns, Nvidia is currently converting the boom into operating income, buybacks, dividends, and a bigger claim on the infrastructure stack beneath every frontier lab.
Salesforce and Anthropic announced Claudeforce, an expanded partnership built around putting Salesforce data, workflows, business logic, and governance directly into Claude. The first product is Salesforce in Claude, a plugin with 37 prebuilt sales skills for meeting prep, deal health, pipeline review, and governed CRM updates.
The partnership cuts both ways: Claude becomes available inside Agentforce and Slack experiences, while Salesforce exposes its business systems to Claude through AIforce, MCP servers, APIs, and CLI tools. Open beta for Salesforce in Claude is planned for September 2026.
This is the enterprise-agent thesis in one deal: the model reasons, the system of record enforces permissions, and the interface becomes whatever the worker is already using.
Wired reported on August 27 that the U.K. power grid is clogged by speculative or phantom data-center projects, many tied to AI demand projections but unlikely to be built. These applications can occupy grid-connection capacity and make the country's future electricity needs look larger and messier than they really are.
Ofgem's proposed reforms would require developers to show funding, customers, and more serious commitments before holding a place in the queue. The goal is to keep viable projects moving while reducing speculative grid hoarding.
For AI, this is the physical bottleneck in plain language. The industry can announce model roadmaps all it wants, but local power planning decides which clusters actually turn on.
Reuters reported that MiniMax posted first-half revenue of $116.6 million, up 283.1 percent from a year earlier, as demand rose for cheaper, open-source-based AI models and platform services from Chinese providers.
Its Open Platform and other enterprise AI services grew even faster, rising 703.1 percent to $73.9 million and becoming nearly two-thirds of total revenue. The company remains loss-making, but its losses narrowed from the prior year.
The western AI story is often told through trillion-dollar infrastructure plans. MiniMax is a reminder that another race is happening underneath: who can push capable models far enough down the cost curve for ordinary businesses to adopt them in volume.
Axios argued on August 27 that the Meta settlement and escalating fights over AI data centers point to a broader backlash against Big Tech's ability to build first and answer later. The pressure now runs through courts, state officials, local communities, and capital markets.
Data centers make the AI issue visible because they turn abstract model demand into land, water, power, noise, and tax debates. Delayed infrastructure is no longer just a community-relations problem; it can become a financial constraint on model companies and cloud providers.
The signal for builders is blunt: technical capability does not buy automatic permission. The next phase of AI deployment will be negotiated with judges, regulators, utilities, neighbors, and balance sheets.