NVIDIA's first earnings report after Jensen Huang's departure: revenue of $115.8 billion; OpenAI releases GPT-6.2 Astra and Sol
NVIDIA released its first earnings report after Jensen Huang's departure, reporting revenue of $115.8 billion for the third quarter of fiscal year 2027 ended October 25, up 103% year over year, including data center revenue of $107.6 billion, up 110% year over year; its revenue guidance for the next quarter is approximately $130 billion and does not include data center computing revenue from the Chinese market. OpenAI released GPT-6.2 Astra and Sol on November 17, aimed respectively at complex professional tasks and everyday work, available to Plus, Pro, Enterprise and Edu users of ChatGPT Work and Codex, and usable by developers via API.
Body · AI translation
Nvidia delivers first earnings report after Jensen Huang's departure, OpenAI releases GPT-6.2 | AI Morning Express Phoenix Tech Download the app for exclusive early access November 18, 2026 08:00:00 From Beijing Updated November 18, 2026 08:36:58
Phoenix Tech News, November 18 — Nvidia delivered its first earnings report since Jensen Huang's departure, with quarterly revenue up 103% year over year. OpenAI released GPT-6.2, and Google opened Gemini task features to free users in six countries. Domestically, the Ministry of Commerce responded to U.S. restrictions on remote access to advanced semiconductors, Xiaomi disclosed progress on testing robots in its factory, and Xingdong Jiyuan's logistics robots have been deployed at 11 postal district nodes in Guangdong.
Today's 10 core AI developments ⏱ 30-second overview
▍Overseas developments
01 Nvidia delivers first earnings report after Jensen Huang's departure, revenue $115.8 billion
02 OpenAI releases GPT-6.2 Astra and Sol
03 Microsoft launches MAI Code 2 and other models, targeting programming and cybersecurity
04 Google opens Gemini task features to free users in six countries
05 Microsoft Copilot opens a new work entry point, Autopilot begins rollout
06 U.S. data centers save money by shaving peak electricity use, dispute arises over how to share grid costs
01 Nvidia delivers first earnings report after Jensen Huang's departure, revenue $115.8 billion
Beijing time in the early hours of November 18, Nvidia announced results for the third quarter of fiscal 2027, ended October 25. Revenue was $115.8 billion, up 103% year over year; data center revenue was $107.6 billion, up 110% year over year. The company expects revenue of about $130 billion next quarter, and its guidance does not include data center computing revenue from the Chinese market.
A closer look
Data centers contributed about 93% of quarterly revenue. The assumptions in next quarter's guidance about the Chinese market help readers understand the boundaries of this growth expectation.
02 OpenAI releases GPT-6.2 Astra and Sol
OpenAI on November 17 local time launched GPT-6.2 Astra and Sol, aimed respectively at complex professional tasks and everyday work. The company said the two models are available to Plus, Pro, Enterprise and Edu users of ChatGPT Work and Codex, with enterprise and education workspaces requiring administrator enablement; developers can also use them via API.
A closer look
The two models target complex tasks and daily use respectively. For ordinary users, what is more worth comparing is actual task quality, usage quotas and response speed.
03 Microsoft launches MAI Code 2 and other models, targeting programming and cybersecurity
Microsoft on November 17 local time unveiled new models including MAI Code 2 at its Ignite conference. The company said Code 2 targets everyday enterprise programming and is already available in GitHub Copilot and Foundry; the cybersecurity model MAI Cyber 2 Flash is used for security tasks, and Reuters said it will be offered to a limited set of customers.
A closer look
Microsoft emphasizes optimizing capability and cost for professional tasks. Vendor-provided benchmark results against competitors still need to be tested against real workloads and payment models.
04 Google opens Gemini task features to free users in six countries
Google announced on November 17 local time that Gemini's task features are beginning to be offered free to users in the United States, India, Canada, Australia, Japan and South Korea. Users can have Gemini handle multi-step tasks or repetitive matters in the background, such as planning trips or organizing information, and receive a notification when it is done.
A closer look
The task feature is expanding from paid tiers to free users in some regions. When email, calendar and other personal information are involved, users still need to pay attention to the scope of authorization.
05 Microsoft Copilot opens a new work entry point, Autopilot begins rollout
Microsoft said on November 17 local time that Copilot Home, which integrates Chat and Cowork, and Office in Copilot are now officially available. Autopilot, which can automatically handle some work, has begun rolling out to Frontier customers, while the Code feature is planned to open officially in the coming weeks. The scope and timing of availability differ by feature.
A closer look
Putting chat, collaboration and office entry points together helps reduce switching; before tasks are executed automatically, permissions and result checks still need human oversight.
06 U.S. data centers save money by shaving peak electricity use, dispute arises over how to share grid costs
Reuters reported on November 17 that some U.S. data centers briefly reduce power draw during periods of highest grid demand to lower capacity or transmission fees. Power company NRG estimates that a large data center in Texas could pay $33.4 million less in transmission fees a year if it avoids one critical peak hour. Industry associations say peak shaving can also ease pressure on the grid.
A closer look
07 European cloud provider OVH releases small model, plans to open weights
European cloud provider OVH on November 17 local time announced its first model family, Olala, and its small model Olala mini 7A1B. The company said the model has about 7 billion total parameters and activates about 1 billion parameters each time, and can run on a graphics card with 8GB of memory; it plans to release it on Hugging Face under the Apache 2.0 license.
A closer look
A lower operating threshold is conducive to developers trying local deployment, but when the open-source weights will be available and how the model actually performs on tasks still depend on subsequent releases and testing.
II | Domestic AI developments 08–10
08 Lei Jun discloses Xiaomi robots have entered factory testing, can work 8 hours
On November 17, Xiaomi CEO Lei Jun said at the APEC CEO Summit that Xiaomi's embodied intelligence robot has recently been tested in a factory and can complete an 8-hour shift, recharging itself twice during that time. He also said Xiaomi is exploring letting industrial agents autonomously discover production problems. The testing processes, task volume and frequency of human intervention were not disclosed.
A closer look
Completing a full shift shows the test duration. Whether it can replace a specific process still depends on task success rate, number of human interventions and long-term operating costs.
09 Xingdong Jiyuan robots already operating at 11 postal district nodes in Guangdong
Xingdong Jiyuan's logistics robot M7 has been deployed at 11 postal district nodes in 8 cities including Guangzhou and Shenzhen covered by China Post's Guangdong branch, mainly handling feeding of sorting lines. The project began with a pilot in Guangzhou in March. According to reports, during regular operations the average feeding efficiency exceeds 1,500 items per hour.
A closer look
From the Guangzhou pilot to deployment in multiple places in Guangdong, the next thing worth watching is whether each node can maintain the same efficiency during peak season, and the frequency of human intervention.
AI Trends