https://zubnet.ai/news/en/ AI business news by Sarah Chen. en Fri, 31 Jul 2026 17:44:33 +0000 https://zubnet.ai/sarah.png https://zubnet.ai/news/en/ https://zubnet.ai/news/en/alphabet-raises-2026-capex-guidance-205-billion/ https://zubnet.ai/news/en/alphabet-raises-2026-capex-guidance-205-billion/ Alphabet lifted its 2026 capital spending guidance to between 195 and 205 billion dollars, up from 180 to 190 billion, adding as much as 15 billion dollars to an already historic build out. The increase came with second quarter results in which Google Cloud revenue grew 82 percent year over year to 24.8 billion dollars, well ahead of estimates, and the company's finance chief said demand continues to outpace the investment. Investors were less enthusiastic than the numbers suggested, with the stock falling on the spending guidance despite the revenue beat, a sign the market is starting to ask when the capital comes back. Wed, 22 Jul 2026 21:00:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/moonshot-ai-seeks-50-billion-valuation-pre-ipo-round/ https://zubnet.ai/news/en/moonshot-ai-seeks-50-billion-valuation-pre-ipo-round/ Moonshot AI, the Chinese lab behind Kimi, reportedly plans to open a final pre-IPO funding round in August targeting a pre-money valuation of up to 50 billion dollars, up from the 31.5 billion valuation of the round it is closing now. The company is expected to pursue a Hong Kong listing within about six months, possibly before the end of the year. The leap is being driven by Kimi K3, the largest open model ever released, which pushed the company's annual recurring revenue to 300 million dollars in June and sent daily revenue up more than sixfold, straining its computing capacity so badly it had to pause new subscriptions. Wed, 22 Jul 2026 08:00:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/kimi-k3-demand-moonshot-suspends-new-subscriptions/ https://zubnet.ai/news/en/kimi-k3-demand-moonshot-suspends-new-subscriptions/ Three days after launching Kimi K3, Moonshot AI suspended new consumer subscriptions because demand was pressing against the limits of its computing capacity. Existing subscribers keep full access, and the company says it is adding capacity and will reopen signups in batches. The pause echoes DeepSeek's moment in early 2025, when a breakout Chinese model met a wall of GPU scarcity, and it lands as Reuters reports Moonshot is preparing an IPO. Mon, 20 Jul 2026 08:00:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/meta-ai-layoffs-discrimination-lawsuit/ https://zubnet.ai/news/en/meta-ai-layoffs-discrimination-lawsuit/ A group of 26 former Meta employees has filed a federal lawsuit in Oakland accusing the company of using AI systems to disproportionately target workers with disabilities, serious medical conditions, or those on approved medical and family leave during its mass layoffs. Meta cut roughly 10 percent of its global workforce this year, nearly 8,000 people, and the plaintiffs allege that the process leaned on automated scoring built from highly flawed metrics, systems that graded people on digital productivity signals and fed a termination list. The suit names specific internal tools, including Metamate, an AI assistant, an employee trained system it describes as a second brain that tracked workers' communications and documents, and a productivity score assembled from scanning keystrokes, screen content, emails, and browser history. The core claim is disparate impact, that scoring people on those signals inevitably flagged workers who were less active precisely because they were ill, disabled, or on leave, turning a health circumstance into a low score and then into a layoff. Meta's response is that people, not AI, made the final layoff decisions, and that framing is the heart of the case, does a system that ranks and effectively recommends who should go carry legal responsibility when a human signs the final list. These are allegations rather than proven facts, the plaintiffs are seeking to block the cuts, which start July 22, while they pursue claims in private arbitration, but it is one of the first high-profile tests of what happens when AI helps decide who keeps their job. Tue, 14 Jul 2026 21:00:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/publishers-sue-google-gemini-ai-training-copyright/ https://zubnet.ai/news/en/publishers-sue-google-gemini-ai-training-copyright/ A group of major publishers and a best-selling author have filed a class action lawsuit accusing Google of willfully using their copyrighted works to train its Gemini AI models without permission. Filed on July 13 in the US District Court for the Southern District of New York, the suit is brought by Hachette Book Group, Cengage Learning, and Elsevier, along with author Scott Turow, and it seeks statutory damages, a permanent injunction to stop further infringement, and an order compelling Google to destroy all unauthorized copies of the works. What makes this more pointed than a generic scraping complaint is the specific allegation, the plaintiffs say Google drew on content from its own restricted, scope limited programs, Google Books, Google Play, and Google Scholar, to train early versions of Gemini, meaning material the company was allowed to access under narrow terms was, they claim, repurposed to build a commercial AI system. It is a putative class action, so it could eventually cover a wide swath of authors and publishers, and it arrives as courts are still trying to settle the central and unresolved question of whether training a model on copyrighted books counts as fair use or infringement. These are allegations rather than findings, and cases like this move slowly and often settle, but the suit puts that question squarely to Google and adds a sharp twist, that the company with some of the largest book access programs in the world may also have some of the largest exposure. Tue, 14 Jul 2026 18:30:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/pixverse-series-c-extension-439m-valuation-2b/ https://zubnet.ai/news/en/pixverse-series-c-extension-439m-valuation-2b/ PixVerse, the Alibaba backed AI video generation startup, has raised its Series C to 439 million dollars in total after an extension round, and its valuation has climbed past 2 billion dollars, roughly double the level it hit when the round's first tranche closed only months ago. The company closed the initial roughly 300 million dollar Series C in March, led by CDH Investments, at a valuation above 1 billion dollars, so the story here is less any single number and more the pace, a video generation company doubling its worth in a matter of months. The extension drew a long list of investors including Alibaba, Mirae Asset, BlueFocus, and Eastern Bell Capital, alongside returning backers, and PixVerse says it will use the money to expand its world model work and reach customers across more regions. The product itself is a video platform with three model families, a consumer and API line, a professional line aimed at film and commercials, and an R series of world models for games, and it can generate clips up to 4k resolution with audio built in. PixVerse reports 150 million registered users and 15 million monthly actives, figures worth treating with the usual caution about vendor numbers. It lands in one of the most crowded and expensive corners of AI, where OpenAI's Sora, ByteDance's Seedance, Kling, Runway, and Luma are all spending heavily, and the raise is a clear marker of how fast capital is flowing toward video, and increasingly toward interactive world models. Mon, 13 Jul 2026 22:00:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/apple-sues-openai-hardware-trade-secrets/ https://zubnet.ai/news/en/apple-sues-openai-hardware-trade-secrets/ Apple has filed a federal lawsuit against OpenAI, alleging a coordinated campaign to steal the trade secrets behind its hardware, and the language is unusually blunt for a company that rarely sues. Filed on Friday, July 10, the complaint names OpenAI Foundation, OpenAI Group PBC, and io Products, the hardware firm founded with former Apple design chief Jony Ive, along with two former Apple employees, and claims that at every level, from members of its technical staff to its chief hardware officer, and in coordination with business partners, OpenAI has been stealing Apple's trade secrets and confidential information. The specifics are striking. Apple says a former senior engineer downloaded dozens of confidential files from its network while developing hardware for OpenAI, and that OpenAI's hardware chief Tang Yew Tan, himself a former Apple vice president, directed Apple employees interviewing at OpenAI to share Apple secrets during the hiring process, including bringing actual parts from Apple to interviews for show and tell sessions. The suit notes that more than 400 former Apple employees now work at OpenAI. These are allegations, not findings, and the case is just beginning, but the collision it represents is real, the world's most valuable hardware company moving to slow the AI company that is openly building a device meant to sit where the iPhone sits. Mon, 13 Jul 2026 17:30:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/sk-hynix-record-us-ipo-ai-memory-hbm/ https://zubnet.ai/news/en/sk-hynix-record-us-ipo-ai-memory-hbm/ SK Hynix has completed the largest ever initial public offering by a foreign company in US history, raising 26.5 billion dollars in its Nasdaq debut. The South Korean memory maker sold 177.9 million American depositary receipts at 149 dollars each, the offering was more than seven times oversubscribed with demand from over 500 investment firms, and the stock jumped about 14 percent on its first day, making this the second biggest share sale of any kind ever, behind only Saudi Aramco. The number is striking, but the reason investors piled in is the real story, SK Hynix is the world's leading maker of high bandwidth memory, the stacked DRAM that sits beside AI accelerators and feeds them data fast enough to keep up, and it holds roughly 56 percent of that market. HBM has become one of the tightest bottlenecks in AI hardware, and the company says its supply is sold out for the rest of the year, with its entire 2026 output of HBM, DRAM, and NAND already spoken for and the crunch expected to run into 2027. The proceeds are aimed squarely at AI memory capacity, funding fab and advanced packaging expansions in Korea and a first US production site in Indiana, which makes the biggest foreign listing in US history a direct bet on the physical supply chain underneath the AI boom. Fri, 10 Jul 2026 18:00:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/meta-model-api-muse-spark-1-1-paid-developer-access/ https://zubnet.ai/news/en/meta-model-api-muse-spark-1-1-paid-developer-access/ Meta has launched the Meta Model API and, with it, started doing something the company built its AI reputation on avoiding, charging developers for access to its best model. The API opens Muse Spark 1.1, which Meta calls its most capable model for real world coding and agentic tasks, to developers in the United States as a public preview, with 20 dollars in free credits and then pay as you go pricing at 1.25 dollars per million input tokens and 4.25 dollars per million output tokens. Muse Spark 1.1 can write and debug code, use software and external tools, understand text, images, and video, and carry out complex multi step tasks with less human intervention, and Meta frames it inside its broader personal superintelligence pitch. The upgrade to the model is real but incremental, the more telling move is the business one, Meta made its name giving models away as open weights with Llama, and a metered, hosted API is the company stepping into the paid inference market that OpenAI and other providers already occupy. It is only a US public preview for now, and Meta has not walked away from open weights, but putting its strongest model behind a paywall is a meaningful widening of strategy, and a signal about where the industry thinks the revenue is. Thu, 09 Jul 2026 15:00:00 +0000 sarah@zubnet.ai (Sarah Chen) https://zubnet.ai/news/en/ollama-raises-65m-series-b-local-ai-models/ https://zubnet.ai/news/en/ollama-raises-65m-series-b-local-ai-models/ Ollama, the open source tool that lets people download and run AI models locally on their own computers, has raised a 65 million dollar Series B led by Theory Venture, and says it now serves nearly 8.9 million developers every month. The round brings the company's total funding to 88 million dollars, after a 15 million dollar Series A led by Benchmark's Peter Fenton, and it lands as Ollama has quietly become one of the default ways developers run open weight models without sending their data to a cloud API. The pitch is simple, a free desktop app that downloads and runs models like Llama, Qwen, and Gemma with a single command, with the model and the data staying on the user's machine. Ollama says it is now used by developers in 85 percent of the Fortune 500, and the open source project has gathered 176,000 stars on GitHub. The money arrives alongside a business model question, the company is expanding into a paid cloud service with subscription tiers from free to 100 dollars a month, billed by GPU time rather than token counts, while promising that the free local product is not changing. It is not a new model or a benchmark record, but a well funded bet that a meaningful share of AI work will keep running on people's own hardware, for reasons of privacy, cost, and control. Thu, 09 Jul 2026 13:00:00 +0000 sarah@zubnet.ai (Sarah Chen)