AI news, 15 August: Anthropic IPO valuation tied to $190-200 billion revenue forecast

Anthropic IPO talks rest on huge 2028 revenue projections. According to a Reuters exclusive citing people familiar with the matter, Anthropic is projecting roughly $190 billion to $200 billion in revenue for 2028 as it prepares what could be one of the largest IPOs on record. Bankers and investors are valuing the company using enterprise value-to-revenue multiples based on those forward forecasts rather than nearer-term figures, a less common approach that reflects the firm’s rapid growth and heavy infrastructure spending. Anthropic’s revenue run rate had reached about $47 billion by May; comparables under discussion include Cloudflare, Palantir and SpaceX. The company did not immediately comment. The figures underscore how investors are betting that today’s massive AI build-out costs will eventually yield much higher margins at scale.

Nvidia scales back planned OpenAI data-center guarantee. The Wall Street Journal reported, and Reuters confirmed, that Nvidia has revised plans to financially backstop a massive OpenAI data-center project in Ohio. The chipmaker is now expected to initially guarantee less than $120 billion, down from the roughly $250 billion previously discussed, covering only the first phase. The change followed investor concerns about Nvidia’s risk exposure. A deal could be signed as soon as this weekend. The full 10-gigawatt campus, developed by a SoftBank energy subsidiary, would be among the largest AI infrastructure projects announced if completed. Nvidia and OpenAI did not provide detailed new comments.

Apple trains its own China-specific AI model with Alibaba’s help. Reuters reported exclusively that Apple has developed a large language model tailored for the Chinese market, trained with support from Alibaba, marking a shift from relying mainly on third-party Chinese models. Apple Intelligence features are expected to roll out in China in the coming months after regulatory registration. The move would give Apple greater control in a key market where it faces strong local competition and where U.S. models such as ChatGPT and Claude are unavailable. It could also make Apple the first foreign firm approved to offer a proprietary AI model in China amid ongoing U.S.-China tech tensions. Apple and Alibaba did not comment.

Google launches Gemini 3.7 Flash for coding and agents. Google released Gemini 3.7 Flash, describing it as its most capable “workhorse” model yet for software engineering, agent workflows and knowledge work. The update arrives three weeks after the previous Flash version and includes improved coding, debugging and multi-step task performance, with an introductory API price cut. Reuters and Google’s own announcements position it as a lower-cost option for businesses building more autonomous AI systems. No firm timeline was given for the next flagship Pro model.

AI designs working viruses in lab breakthrough. Researchers at Stanford University and the Arc Institute reported that AI models (Evo 1 and Evo 2) designed novel bacteriophage genomes—viruses that infect bacteria. Of hundreds synthesized and tested, 16 proved fully functional, some overcoming bacterial resistance that defeated the natural version. The work, covered in Science and recent recaps, points toward potential new tools against antibiotic-resistant infections but also raises biosecurity questions, as the method itself does not inherently limit targets. Experts have noted a gap between the pace of the science and existing safeguards.