The AI Brief

Nvidia reported revenue for the second quarter ended July 26, 2026, of $96.2 billion, up 18% from the previous quarter and up 106% from a year ago. Data Center second-quarter revenue was $89.0 billion, up 18% from the previous quarter and up 117% from a year ago. GAAP and non-GAAP gross margins were both 75.0%; GAAP earnings per diluted share were $2.46 and non-GAAP were $2.22.

Third-quarter revenue guidance is $108.0 billion, plus or minus 2%. Nvidia is not assuming any Data Center compute revenue from China in its outlook. Huang said AI "reached its inflection point," noting that the number of companies needing large GPU clusters has expanded dramatically: "This time last year, one lab alone was driving the buildout," he said. "Today, we have a golden age of new AI labs and startups, multiple frontier labs scaling in parallel, a thriving open-model ecosystem and physical AI coming online."

Nvidia announced the Vera Rubin platform is ramping into full production with racks running at partners including CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. Nvidia stock jumped 8% during premarket trading on Thursday following the company's earnings release. Results were reported after the close on August 26. Nvidia's fiscal Q2 2027 ended July 26, 2026.

Anthropic has agreed to spend $45 billion to rent AI cloud computing power from Nscale's flagship data center development in West Virginia, the latest move to secure capacity for its expanding business in advance of going public. Nscale will use Nvidia's Vera Rubin chips, which start coming online late next year, to meet Anthropic's needs; the commitment runs six years and represents about 460 megawatts of power.

Nscale was founded in May 2024 by Josh Payne, who previously co-founded Arkon Energy, an infrastructure company tied to crypto mining; the London-based firm raised just over $4.9 billion in a little over a year and reached a $14.6 billion valuation after a Series D round. Nscale is also preparing a US listing as soon as next month. Anthropic is projecting 2028 revenue of roughly $190 billion to $200 billion, per Reuters.

Bloomberg reports Anthropic has also signed other deals for computing power in recent months, including ones amounting to $50 billion with Fluidstack, $10 billion with Volta Infra Holdings, and $45 billion with SpaceX. The Nscale deal, first reported by Bloomberg, was confirmed by Reuters and CNBC citing separate sources familiar with the agreement.

Disclosure: Claude, which generates this brief, is built by Anthropic.

In initial benchmarks, Jalapeño delivered 1.5×–1.9× higher throughput per kilowatt and 1.7×–3.6× lower end-to-end latency than Nvidia's GB200 and GB300 rack systems, per Tom's Hardware reporting on the Hot Chips presentation. Richard Ho, OpenAI's head of hardware, said on a press call: "The bottom line is that the results show a very, very significant performance advance over state of the art."

Since announcing Jalapeño, OpenAI's first custom inference chip, the company has been testing the chip and the system built around it; Jalapeño delivers both higher throughput and lower latency with one architecture, where existing hardware systems often have to make a tradeoff between the two. Each Jalapeño package combines its compute die with six HBM4 stacks, delivering 216 GiB of memory and 15.4 TB/s of bandwidth.

Jalapeño is the first step in a multi-generation compute platform designed for initial deployment by the end of 2026 and expanding in the years ahead, combining OpenAI-designed accelerators with Broadcom silicon implementation, networking, and connectivity technologies; and Celestica's board, rack, and system expertise. Analysts told CNBC that it could put pressure on Nvidia in the fast-growing inference market and reduce OpenAI's reliance on Nvidia for some workloads. Jalapeño is OpenAI-captive silicon, not a product OpenAI sells, with no API, no rental market, no instance type; the chip serves OpenAI's own API traffic.

Caveat Benchmark figures are from OpenAI's own Hot Chips presentation tested on SemiAnalysis' InferenceX benchmark. Independent third-party replication has not been published.

Nvidia has agreed to buy Hugging Face for $12.9 billion, a deal that would put the world's dominant AI chipmaker in control of one of the most important distribution hubs for open AI models; The Information first reported the agreement, citing a person familiar with the deal, and Reuters later confirmed it. The deal comes just two days after reports that Hugging Face was exploring a potential sale at a valuation of $13 billion or more.

The New York-based company hosts more than three million open-source AI models and one million datasets and had engaged a bank to gauge buyer interest. Hugging Face last raised in 2023 at a $4.5 billion post-money valuation in a round led by Salesforce Ventures with participation from Alphabet, GV, IBM Ventures, and others. The company earlier this year turned down a $500 million investment from Nvidia that would have valued it at $7 billion, saying it didn't want a single dominant investor to sway decisions.

The acquisition would give Nvidia something it cannot manufacture in a semiconductor fab: a massive community of AI developers and a repository that has become a default destination for models, datasets, and machine learning tools. The deal has not closed; Nvidia and Hugging Face did not immediately respond to requests for comment per Reuters.

China's Zhipu AI launched GLM-5.3-Flash, previously code-named Ox Alpha, saying the system ran entirely on a cluster of 100,000 domestically produced chips during a high-profile stealth trial. Before formal release, the model processed 62 trillion tokens on OpenRouter and agent platform OpenCode; on OpenRouter, the system processed more than 11 trillion tokens in its first three days, making it the platform's biggest launch to date.

GLM-5.3-Flash ranks 10th on the Artificial Analysis Intelligence Index, ahead of DeepSeek V4 Pro Max. Someone anonymously dropped Ox Alpha on OpenRouter and OpenCode on August 20, offering a 1-million-token multimodal context window, a 131,072-token output limit, and a capacity to process 100 trillion tokens per day entirely free of cost for a week. The model is 320B total parameters with 18B active, under MIT license, at $0.15/$0.50 API pricing.

The deployment marks a significant test of China's ability to handle large-scale global inference workloads on home-grown hardware, as Beijing seeks to reduce reliance on advanced processors from Nvidia amid tight export controls. Zhipu's shares closed more than 12% higher in Hong Kong on Thursday. Weights are published on Hugging Face under MIT License. Z.ai is the international brand name for Beijing-based lab Zhipu AI.

Caveat The claim that inference ran exclusively on domestic Chinese chips is from Z.ai's own launch blog. Independent hardware verification has not been published.