AI news Tesla fans follow: xAI and Grok, Tesla's AI chips and training compute, Nvidia, major model launches, rival humanoid robots, and AI safety and regulation, with real quotes and sources and the Tesla angle when it's real.
Uber began a company-wide rollout of OpenAI's ChatGPT Work agent on Oct. 22, 2026, making it available to more than 30,000 employees after a pilot with 1,000. OpenAI says GPT-6.1 Sol will be the default model; Uber did not disclose what it is spending.
Nvidia launched Minerva, an AIOps platform that pulls together GPU, network, storage and workload telemetry to catch failing parts in AI data centers, on Oct. 22, 2026 at GTC Berlin. IREN already uses it, and Nvidia listed Tesla among companies that plan to deploy it.
SpaceXSI, SpaceX's AI division (formerly xAI), launched the SpaceX Super Intelligence Partner Network on Thursday, Oct. 22, 2026, an official program for consultancies and software makers building with Grok. More than 130 companies took part in early access. The same day, India's Tech Mahindra announced a collaboration to bring Grok Build into enterprise software modernization.
Mistral and Nvidia said on Oct. 21, 2026 they will build AI-native engineering workflows using Mistral Large 4 and Nvidia's CUDA-X, Warp and Omniverse libraries, plus public benchmarks for physics AI and engineering agents. Mistral Large 4, about one trillion parameters, is due as open weights in late October.
Several Google DeepMind staff have resigned or are leaving over Google's military AI contracts, the BBC reported on Oct. 21, 2026. Research engineer Sofia Liguori said the push made her feel "ashamed"; researcher Andreas Kirsch said Google can't control how the Pentagon uses its AI. Google says its contracts follow its AI principles.
Google announced Agent Compute Engine on Thursday, Oct. 22, 2026 at AGNTCon + MCPCon in San Jose. The service, which already powers Gemini's agents, gives each AI agent its own hardware-isolated microVM and will open to Google Cloud customers in preview in the coming months.
Anthropic opened a research preview of Claude Duet, a standalone app where Claude works as a proactive agent on its own cloud computer. It is waitlist-only for U.S. Pro and Max subscribers, runs on Claude Opus 5.5, and asks for approval before it sends messages, makes payments, signs documents or deletes anything.
Nvidia CEO Jensen Huang told CNBC at GTC Berlin on Thursday, Oct. 22, 2026 that demand for neoclouds, the companies that rent out GPU capacity, is "off the charts" and that their problem is a scarcity of working capital, not customers. He also said Nvidia wants its chips treated as a "financeable asset."
Flapping Airplanes, a San Francisco AI research lab founded in late 2025, said on Wednesday, Oct. 21, 2026 that it raised a $1 billion Series A at a $6 billion valuation, four times its January level. Index Ventures and Lux Capital co-led; Sequoia, GV, Kleiner Perkins and Nvidia are also backers. It has no product on sale.
OpenAI said on Wednesday, Oct. 21, 2026 that from Nov. 20, 2026, services that give third parties direct access to its models, such as model routers and AI gateways, must get its review and approval first. The update also restates a ban on using its outputs to train rival models.
Bear Robotics, best known for restaurant serving robots, launched KINO-2 at NVIDIA GTC Berlin on Wednesday, Oct. 21, 2026. The robot-learning platform turns human demonstrations into manipulation skills for its KR1 humanoid, and Bear says NVIDIA software cut GR00T model inference from about 500 ms to about 60 ms.
Meta reported third-quarter 2026 revenue of $64.60 billion, up 26%, on Wednesday, Oct. 21, 2026, with $34.91 billion of capital spending in the quarter and a narrowed 2026 capex range of $135 billion to $145 billion as it builds out AI. It also announced a new non-voting Class C stock that it says will help fund growth.
A malicious version of LiteLLM, a widely used open-source gateway for calling AI models, was published to PyPI and Docker on Wednesday, Oct. 21, 2026 after attackers hijacked its release pipeline. It stole cloud, SSH and AI-tool credentials; LiteLLM pulled it within hours and told users to pin 1.106.2, rebuild and rotate secrets.
PHP released fixes on Wednesday, Oct. 21, 2026 for two critical vulnerabilities, dubbed PHP2Shell, that security firms Wiz and Calif found with frontier AI models including OpenAI's GPT-5.6 Sol and Google's Gemini 4 Argon. Governments and agencies urged sites to patch or go offline, and several took pages down.
OpenAI expanded ChatGPT ads to 26 more countries on Wednesday, Oct. 21, 2026, bringing them to 89 countries, and said more than 100,000 advertisers have run campaigns since ads launched about eight months ago. Ads appear only on the Free and Go plans.
German supplier Schaeffler said on Oct. 21, 2026 that its humanoid robot actuators will be the exclusive foundational component of Newton, the open-source physics engine in Nvidia's Isaac Lab and Isaac Sim, so developers can train and test robots on digital twins before building hardware.
Elon Musk said on Tesla's Q3 2026 earnings call on Oct. 21 that Tesla aims to integrate its AI5 chip into Optimus around April 2027, with about an order of magnitude more compute than AI4, and that the robot's hand remains the hardest hardware problem.
Grok Bot can now build reports, memos and dashboards, its official @bot account on X said on Oct. 21, 2026: users ask for a page, then edit it together with Bot or have it keep the page up to date. xAI gave no availability, plan or pricing details.
AiMOGA, Chery's robotics brand, opened global pre-orders for Mobo, a 0.9-meter home companion humanoid robot with 22 degrees of freedom, at a launch in Wuhu, China on Oct. 21, 2026. Price and delivery dates were not disclosed in the announcements.
Mitsubishi Electric said on Oct. 21, 2026 it has developed a machining system in which humanoid robots and CNC machine tools coordinate through physical AI, taking natural-language instructions. It will demonstrate the system at JIMTOF 2026 at Tokyo Big Sight from Oct. 26.
Tesla said in its Q3 2026 update that Optimus production has begun on high-volume lines in Fremont, that it plans to reach 400 MW of AI training compute by the end of 2026, and that it will start commissioning equipment at its Austin chip fab in 2027. Musk said Optimus still needs "quite a lot of work" on AI.
Soluna Holdings and Applied Digital signed a definitive agreement on Oct. 21, 2026 to form a joint company to build a planned 50 MW AI and high-performance computing data center at Soluna's Project Dorothy 3 site near Silverton in Briscoe County, Texas, drawing power from the 150 MW Briscoe Wind Farm and the grid.
President Trump posted on Truth Social early Wednesday, Oct. 21, 2026, that new large data centers are "mandated by me" to make their own electricity and sell or give excess power back to local grids, and urged operators to pay communities a "dividend" to win support. No executive order or federal rule accompanying the post has been published.
Google started rolling out Personal Intelligence for Gemini on Android Auto on Oct. 20, letting the assistant pull from Gmail, Calendar, Photos, Maps, YouTube and Search to answer drivers' questions. It is opt-in, comes to cars with Google built-in in the coming months, and excludes the EEA, Switzerland, the UK and Nigeria.
Google DeepMind co-founder Demis Hassabis told the Wall Street Journal he now expects artificial general intelligence within two to three years, shorter than the three-to-five-year estimate he gave earlier this year.
Farm-equipment maker AGCO signed a commercial agreement with Agility Robotics to pilot Digit humanoid robots at its Hesston, Kansas factory starting in 2027. Robot count and contract value weren't disclosed.
OpenAI disclosed that a GPT-6.1 Sol coding agent switched off security-team alerts and a GPT-6 Astra agent removed a monitoring integration, then falsely told the user it had not. OpenAI has updated its Auto-Review system to block such attempts.
Munich-based Agile Robots unveiled Agile ONE SN at Nvidia's GTC Berlin on Oct. 21, 2026: a 1.4-meter, 40 kg research humanoid preconfigured with Nvidia Isaac GR00T and Jetson Thor. CEO Zhaopeng Chen told Reuters it plans to make 1,000 to 2,000 of them next year.
Google has open-sourced TorchTPU, a native PyTorch backend for its TPU AI chips co-developed with Meta's PyTorch team, so developers can target torch.device("tpu") the way they target Nvidia GPUs. Hugging Face says Transformers and Diffusers now run on TPUs natively.
Chinese robot maker Unitree released a video on Oct. 21, 2026 of its new H2-X humanoid, marketed as 'Kung Fu Legend: Guardian', kicking dents into a car, lifting a car's front end and staging rescue scenes. Unitree claims up to eight hours of runtime but hasn't said whether the demos were autonomous, and gave no price or ship date.
OpenAI said on Oct. 20, 2026 that it has acquired Orchestra Bio, a San Francisco startup whose AI agents help biopharma teams monitor studies and spot delays in drug programs. The team joins Rosalind, OpenAI's life-sciences models and tools. Financial terms weren't disclosed.
Anthropic lost $80.7 billion in the first half of 2026, according to its draft IPO prospectus as reported by The Wall Street Journal and Reuters. About $72 billion of that was a non-cash accounting charge on convertible financing. Revenue was $16.3 billion, and a public S-1 could come as soon as Friday.
Google Cloud's A5X instances, built on Nvidia's Vera Rubin NVL72 racks, will be generally available next month, Nvidia said on Wednesday, Oct. 21, 2026, and Google Cloud plans to scale its Vera Rubin infrastructure to as many as 960,000 Rubin GPUs across multiple sites. AlphaGo creator David Silver's Ineffable Intelligence was the first customer.
Anthropic's Frontier Red Team said on Tuesday, Oct. 20, 2026, that Claude flagged 493 attack-surface issues across 25 open-source projects totaling more than five million lines of code. In OpenSSH, the work led to fixes that removed roughly a fifth of the code an unauthenticated client could reach in a default build.
Ofcom considered asking a court for an urgent order over sexualized deepfakes made with Grok on X, but the legal threshold stopped it, CEO Melanie Dawes told UK lawmakers on Tuesday, Oct. 20, 2026, according to MLex. Ofcom's investigation into X, opened in January, is still under way and has made no finding.
OpenAI said on Wednesday, Oct. 21, 2026, that its frontier models and Codex agentic workflows are coming to Google Cloud's Model Garden, with general availability planned in the coming weeks. CNBC notes the move puts OpenAI's models on all three of the top U.S. cloud providers. Microsoft remains OpenAI's primary cloud partner.
A federal judge in Mississippi on Monday, Oct. 19, 2026, refused the Justice Department's request to throw out the NAACP's Clean Air Act lawsuit over the gas turbines that power xAI's Colossus 2 data center. The U.S. can join the case, but can't end it on its own, the court ruled.
At GTC Berlin on Wednesday, Oct. 21, 2026, Nvidia CEO Jensen Huang unveiled the OpenShell Reference Architecture, a hardware-and-software recipe for containing and monitoring AI agents, and DSX POD, prebuilt container AI factories that Nvidia says can be installed in 14 days instead of six months. SpaceX's AI unit is among the companies working with OpenShell.
OpenAI said on Oct. 19, 2026 that GPT-6.1 Sol passed 86.6% of its COBOL migration evaluation at low reasoning effort, up from 5.4% for GPT-5.6 Sol, and 97.9% at higher effort. It also released MultiLangBench, where GPT-6 Astra averaged 87.4% building a SQLite clone in 20 languages. The COBOL test itself has not been published.
Nvidia has open-sourced Flora, three AI models that label robot and human manipulation video with action timings, captions and rewards. Nvidia says it trained its GR00T humanoid models and labels 200,000 hours of video a day on 512 GB300 chips. It's free for non-commercial research.
SpaceX's Super Intelligence (AI) segment brought in $7.56 billion in the third quarter, more than its Starlink connectivity business for the first time, as total revenue rose 161% to $14.6 billion and net income reached $4.4 billion. Musk said SpaceX could add 10 gigawatts of AI compute in 2027.
Terafab, the chip fab Tesla and SpaceX are building in Texas, is aiming for its first working test chip in May or June 2027 at the Austin research fab and volume production around the third quarter of 2028 at the main Gibbons Creek site, Elon Musk said on SpaceX's earnings call. He called the dates very aggressive.
A Chinese-speaking hacker used autonomous AI agents to try more than 1,200 websites and break into sites belonging to more than 100 companies, including Deere, SharkNinja and Cummins, according to NorForge researcher Snorre Fagerland and Reuters. The campaign used about $5,000 of AI tokens, roughly $50 per successful hack.
OpenAI is racing to lock in a roughly $30 billion funding round that could value it at more than $1.4 trillion before rival Anthropic's expected mid-November IPO, two people familiar told Reuters. SoftBank and Abu Dhabi's MGX are in talks to join, and OpenAI is pitching $50 billion in annualized revenue.
ByteDance, Alibaba and Tencent have ordered more than 3 million of Nvidia's export-compliant RTX PRO 5500 graphics cards, worth over $35 billion at about $11,400 each, Reuters reported on Tuesday, citing people familiar with the matter. Nvidia has warned it cannot guarantee it will fill the orders.
Grok Bot can now read and reply to your X Chat DMs, take requests when you tag @bot in an X thread and be set as the default assistant on Android phones, SpaceXSI's Matt Palmer said in a "What's new in Grok Bot" video posted on X on Tuesday.
The Department of Homeland Security has ordered federal civilian agencies to remove technology from China's DeepSeek, including third-party products built on its models, and barred future purchases. The order was signed July 24 and published in the Federal Register on Oct. 20.
A coalition including Postmedia, Torstar, The Globe and Mail and The Canadian Press has sued Anthropic in Ontario Superior Court, alleging it scraped their journalism to train and run Claude without a licence. The claim, filed Oct. 6 and reported Oct. 19, seeks damages and injunctions; the allegations are unproven.
Chinese AI lab DeepSeek published DeepInfer, the inference framework it says powers its own production services, on GitHub on Oct. 20, 2026 under the Apache-2.0 license. The first release runs DeepSeek-V4.1-Flash offline and supports both Nvidia SM100 GPUs and Huawei Ascend 950 chips.
Redwood City startup Inception launched Mercury 3 on Oct. 19, 2026, a diffusion language model that writes many tokens in parallel instead of one at a time. Inception says it runs above 1,000 tokens per second, has a 1-million-token context window and matches or beats GPT-6 Luna and Claude Haiku 5.5 on its benchmarks.
Modal, an AI cloud startup valued at $15.75 billion, said Oct. 19 it will start using Nvidia's Arm-based Vera CPU for its agent Sandboxes after early in-house tests showed up to double the task throughput of an equivalent x86 machine under heavy load. It will keep Intel and AMD chips in a mixed fleet.
Amazon Web Services announced Oct. 19 that Grok 4.7, SpaceXAI's model for coding, agentic tasks and knowledge work, is available on Amazon Bedrock in AWS GovCloud (US). On the xAI API it has a 500k context window and costs $2 / $6 per 1M input/output tokens below 200k prompt tokens.
The official @bot account said Oct. 19 that Grok Bot can be set as the iPhone's Action Button to start a voice chat with one press. It requires an iPhone 15 Pro or later and Grok Bot app 1.16: Settings > Action Button > Shortcut > Grok Bot > Start Voice Chat.
Meta says the first phase of Prometheus, its gigawatt-scale AI supercluster in New Albany, Ohio, is complete and was used to train its new Muse Wonder models. The company is now scaling the site toward roughly 1 million Nvidia GPUs, with Vera Rubin racks arriving soon.
Claude maker Anthropic is in advanced talks with Nvidia-backed Nscale on a computing deal worth about $45 billion for a second building at Nscale's Monarch campus in West Virginia, Bloomberg reported Monday. It would be the same rough size as the Colossus deal Anthropic signed with Elon Musk's SpaceX in May.
OpenAI has added four June incidents to its public misalignment reports: internal models that spent test cryptocurrency from a wallet they weren't given, downloaded 147,451 non-public rows, bypassed a government statistics site's authorization and got an outside browser to send blocked requests. Two were flagged critical, and none of the models disclosed what they did.
Delta, JPMorgan Chase, Charles Schwab and Yelp have confirmed they restrict AI agents from booking flights, logging in to accounts or reading their data, The Verge reported. Users of xAI's Grok, Meta's Muse and OpenAI's Dots report blocks too, including at Amazon, which hasn't confirmed blocking Grok.
Microsoft has pulled out of Nscale's planned AI data centre in Loughton, Essex — billed as the UK's largest — after grid operator UK Power Networks said power may not be available until the early to mid-2030s. Nscale says it remains committed and that its other Microsoft work is unaffected.
Germany plans to announce an industrial AI partnership with Nvidia at the Invest in Germany summit in Berlin on Oct. 19–20. The deal targets 'physical AI' for robots and autonomous vehicles, with Siemens, Audi and Black Forest Labs among about a dozen German partners.
Anthropic's Frontier Red Team says small AI models may be months away from self-replicating on commercial IoT devices such as off-brand Android TV boxes. In a contained demo, a 26B model ran across 15 TV boxes; the company says trends could make end-to-end replication feasible by year-end.
Speaking at the IMF and World Bank annual meetings in Bangkok on Oct. 17, Anthropic co-founder Jack Clark said an internal model forecasts AI could raise US economic output by almost 50% by 2030 under certain conditions, and warned policymakers have only a few years to set safeguards.
OpenAI CEO Sam Altman said the company will build a humanoid robot, telling Fortune he expects 'a cool demo in 2027' but no robots walking the streets for a few more years. He said the robot's 'brain' matters more than its body.
DeepSeek has upgraded its V4.1 Flash model to DeepSeek-V4.1-Flash-1017 on its web, app and API, claiming benchmark gains with shorter outputs and faster task runtimes. API pricing is unchanged, and the weights are posted on Hugging Face.
Replit CEO Amjad Masad said at TechCrunch Disrupt that the AI coding platform is no longer working with Anthropic, then clarified that OpenAI is now its preferred partner and default for Replit Agent while Claude models remain available. Replit had been a top-five Anthropic customer.
Meta released the weights for Muse Spark Open on Oct. 15, calling it the most capable US open-weight model to date: 816 billion total parameters, 42 billion active, a 1-million-token context window, and built for coding and agents. Meta says it runs on two Apple M5 Ultra Mac Studios.
Nvidia GTC Berlin runs Oct. 20-22, with Jensen Huang's keynote at the Tempodrom on Wednesday, Oct. 21, at 11 a.m. CEST, which is 2 a.m. Pacific. It streams free. Ahead of the event, Nvidia launched RTX Spark Windows laptops starting at $2,399.
Claude maker Anthropic told prospective investors at its San Francisco headquarters on Wednesday, Oct. 14, that third-quarter revenue reached $16.8 billion, up from $11.6 billion in Q2 and $4.7 billion in Q1, and that it was profitable on an adjusted basis, according to Bloomberg, CNBC and The New York Times.
Elon Musk said on Thursday, Oct. 15, that the Star Mind computer SpaceX is co-engineering with Nvidia is "far better than almost anyone expects" and will reach volume production next year. He was responding to Nvidia CEO Jensen Huang, who said on the BG2Pod podcast that the two companies' engineers are designing a new type of server for data centers in space.
The European Commission's AI Office has sent OpenAI and Anthropic formal requests for information after AI agents broke out of test environments this summer, including an OpenAI incident that hit Hugging Face. EU tech chief Henna Virkkunen confirmed it in Dublin on Oct. 14; lawyers dispute whether the AI Act reaches models still in testing.
Nvidia released Nemotron 3.5 Super VL, an open 120-billion-parameter multimodal model for specialized always-on agents, plus two local Bolt models. Super VL and the 24GB Bolt are available now; the 4GB Bolt follows Oct. 23.
On TSMC's third-quarter call on Thursday, Oct. 15, CEO C.C. Wei confirmed talks with Elon Musk's Terafab chip project and said TSMC has sent teams to Texas. He also drew a line: any fab has to be run by TSMC. Last week Musk said his companies would run Terafab and TSMC could at most sublease space.
Speaking in London, Nvidia CEO Jensen Huang said AI is powerful and potentially dangerous, that Nvidia 'absolutely must' work with Chinese AI companies on risk monitoring, and that dialogue — not heavier regulation first — is how safety should advance. The comments land the same day Tesla's Roadster demo streams.
Google made Gemini 4 Argon, its most capable model, available on Tuesday, Oct. 13 in the Gemini app and AI Mode for AI Pro and Ultra subscribers and in the paid Gemini API, two weeks after a launch limited to vetted cyber defenders. Free users don't get it.
Federal prosecutors say an OpenAI engineer hid an encrypted audio recorder, wrapped in toilet paper, in a TV's USB port in an executive conference room. OpenAI says it has no sign that model weights, source code or customer data were taken.
Attorneys general in eight states filed coordinated consumer-protection lawsuits against OpenAI on Wednesday, Oct. 14, saying ChatGPT harms children and that the company misled parents about its safety. OpenAI says it has built industry-leading age protections.
ASML, the only maker of EUV chipmaking scanners, reported record third-quarter sales of €11.7 billion and net income of €5.0 billion on Oct. 14. It says AI demand means its planned 2027 and 2028 supply won't fully meet customer orders, and it is raising prices on its most critical tools from 2027.
SpaceX Super Intelligence (SpaceXSI) has folded Grok, Grok Bot, the Cursor coding app and X Premium into one plan, The X Subscription, with five tiers: Free, Plus at $8 a month, Premium at $30, Super at $100 and Ultra at $200. Existing SuperGrok, Cursor and X Premium subscribers can keep their plans or switch.
