|
|
|
Dec 19, 2025
|
|
|
|
|
Supported by
|
|
|
|
|
|
|
TGIF! OpenAI's 2025 revenue is on track to top $13 billion. ByteDance signs a deal to create a U.S.-based TikTok joint venture majority owned by American investors. Thinking Machines Lab plans to release its own AI models next year.
|
|
|
|
OpenAI is on track to beat its 2025 revenue goal of $13 billion, up from around $4 billion last year, The Information reported. Annualized revenue, or the last month’s revenue multiplied by 12, by last month topped $19 billion, more than triple the pace in January. The growth comes as company begins early discussions to raise as much as $100 billion at a valuation of around $750 billion before the investment, 50% higher than its valuation two months ago during the share sale. The discussions suggest that private investors still have appetite for
fast-growing AI startups even while worries about an AI bubble derail some public tech stocks.
|
|
|
|
TikTok owner ByteDance reached agreements with investors to create a U.S.-based TikTok joint venture majority owned by American investors, according to an internal memo to employees viewed by The Information. ByteDance signed a deal with Abu Dhabi-based MGX, Oracle and investment firm Silver Lake to create the new company and is aiming to have the transaction finalized by January 22, according to the memo–one day before a deadline set by the White House for a sale of TikTok’s U.S. operations. MGX, Oracle and Silver Lake will each own 15% of the new company, as part of a broader consortium holding a 50% stake. Affiliates of existing ByteDance investors will own 30.1%, and ByteDance will own 19.9%. The agreement is in line with a proposed deal mapped out this fall and brings ByteDance closer to ending the multi-year negotiations determining the fate of TikTok in the U.S. The new joint venture will be responsible for safeguarding U.S. user data in a cloud environment run by Oracle. A TikTok spokesperson did not immediately respond to a request for comment on the memo, which was first reported by Axios.
|
|
|
|
Thinking Machines Lab, the AI lab co-founded by former OpenAI Chief Technology Officer Mira Murati, is planning to release its own in-house models next year, co-founder and Chief Scientist John Schulman said on a podcast. So far, Thinking Machines has released Tinker, an application programming interface that allows developers to finetune, or tweak, open-source models. Murati has told investors that the company aims to produce models customized to a company’s performance indicators or business metrics. It also plans to release a consumer product, such as an AI-powered assistant consumers can speak
to. Thinking Machines is part of a growing group of startups called “neolabs” who hope to exploit new approaches to developing AI models and research they say major developers like OpenAI and Anthropic may have overlooked. The company was recently in discussions to raise between $4 billion and $5 billion in new funding at a valuation of at least $50 billion. It has so far raised $2 billion, most recently at a valuation of $10 billion, from investors including Andreessen Horowitz.
|
|
|
|
Brett Adcock, the CEO of robotics startup Figure AI, has started a new AI lab called Hark, which will be funded by $100 million of his personal capital, according to a copy of a memo Adcock sent to Figure employees and investors on Thursday seen by The Information. The AI lab will be focused on building “human-centric” AI, which can “think proactively, recursively improve and care deeply about people,” the memo said. Hark’s first cluster of graphics processing units went online on Monday, the memo said, though it couldn’t be learned how large the cluster was. Adcock will remain the CEO of Figure, in addition to his new role at Hark, according to a person with knowledge of the move. Hark continues the trend of AI startup CEOs founding new companies while remaining at their old one. Richard Socher, the CEO of
search engine startup You.com, is similarly launching an AI lab which he is raising $1 billion in funding for, The Information has reported. More broadly, well-known researchers and executives have raised billions of dollars for so-called “neolabs” in recent months, which hope to exploit new approaches to developing AI models and research
they say major developers like OpenAI and Anthropic may have overlooked. These neolabs are taking on the larger AI labs despite the fact that those firms are already generating billions of dollars in revenue. Figure has previously raised nearly $2 billion in funding from investors including Parkway Venture Capital, Nvidia, Microsoft and Bezos Expeditions and was last valued at $39 billion.
|
|
|
|
Meta Platforms is working on two new AI models—one focused on images and video, and the other on text—for release in the first half of 2026, the company’s chief AI officer Alexandr Wang told employees on Thursday. The Wall Street Journal said Wang made his remarks about the video and image model, which is code-named Mango, and the text model, code-named Avocado, in a company question and answer session. Wang told employees that one of the priorities for the Avocado model is to improve its software coding abilities. Meta’s focus on an image and video model is no surprise. The company has already launched a video generation tool called Vibes, which uses technology from the startup Midjourney. OpenAI and Google have headstarts on Meta in the category with their Sora video- and Nano Banana image-making tools,
respectively. Wang also told employees that Meta is in the early stages of working on world models, which process visual information to learn about physical environments. CNBC previously reported on the existence of the Avocado model.
|
|
|
|
| |
