AI news4
Nvidia agrees to buy Hugging Face for $12.93 billion
Nvidia will acquire Hugging Face, the platform hosting more than 18 million users sharing 3 million AI models used by over 200,000 companies. It's Nvidia's second-largest purchase ever, after the $20 billion Groq asset buy, and includes up to $1 billion in employee retention bonuses. CEO Jensen Huang pledged Hugging Face stays an open platform, with no obligation to use Nvidia products.
techcrunch.com ↗Why it matters
A chip maker best known for closed-model partners like Anthropic and OpenAI just bought the biggest hub for open-source AI, a bet that owning the open ecosystem matters as much as owning the silicon.
Cognition AI nears $47 billion valuation as coding-agent revenue surges
AI coding startup Cognition is closing a round that would value it near $47 billion, nearly double its $26 billion valuation from three months ago. It's raising about $1 billion against nearly $10 billion in investor demand. Annualized revenue has grown from $492 million in late May to more than $900 million now.
techfundingnews.com ↗Why it matters
A coding-agent startup roughly doubling its valuation in three months on real revenue growth, not just hype, shows how much money is still chasing AI coding tools specifically.
Moonshot AI files for $3 billion Hong Kong IPO at $50 billion valuation
Beijing-based Moonshot AI, maker of the Kimi model family, confidentially filed for a Hong Kong listing targeting $3 billion. The filing follows a round that raised over $2 billion from Meituan and China Mobile, and required Moonshot to restructure out of its offshore holding company into mainland China. Rivals DeepSeek and Z.AI are currently valued at $74 billion and $66 billion.
techstartups.com ↗Why it matters
A Chinese AI lab choosing a public listing over further private rounds signals the market for AI-lab liquidity is maturing, putting a real price tag on China's frontier-model race.
Tether open-sources offline AI translation models for 19 African languages
Tether's QVAC division released TranslatePsy-AfriSLM, translation models covering 19 Sub-Saharan African languages including Swahili, Yoruba, Hausa and Zulu. Each language pair is just 21 to 35 megabytes, small enough to run offline on a phone, and processes a sentence in about 46 milliseconds, up to 78 times faster than comparable models. The release also covers 8 more African languages and 9 European languages under companion model families.
cryptotimes.io ↗Why it matters
A stablecoin company, not an AI lab, just shipped some of the most practically useful low-resource-language AI available, a reminder that useful AI-for-good tools increasingly come from unexpected places.