Top Stories
Anthropic Closes $30 Billion Series G at $380 Billion Valuation
Anthropic announced its Series G round on Wednesday: $30 billion at a $380 billion post-money valuation. That’s more than double the $183 billion valuation from its Series F and makes this the second-largest private tech fundraise ever, behind only OpenAI’s $40 billion raise last year.
Singapore’s sovereign wealth fund GIC led the round alongside Coatue, with D.E. Shaw Ventures, Peter Thiel’s Founders Fund, and Abu Dhabi’s MGX co-leading. Accel, General Catalyst, Jane Street, and the Qatar Investment Authority also participated. The company’s annualized revenue has hit $14 billion, with Claude Code alone running at $2.5 billion - more than double where it started the year. Customers spending over $100,000 annually have grown 7x in the past twelve months.
The timing is deliberate. Anthropic just expanded Claude’s free tier with features previously locked behind its $20/month Pro plan - file creation, app connectors, and custom skills - while OpenAI started showing ads to free and Go-tier ChatGPT users. The contrast was not subtle. Anthropic’s Super Bowl ad literally said “Ads are coming to AI. But not to Claude.” With $30 billion in fresh capital, the company can afford to keep that promise for a while.
Sources: TechCrunch, CNBC, SiliconANGLE
Chinese AI Stocks Explode as Zhipu and MiniMax Drop New Models
Chinese AI companies triggered a market frenzy on Thursday as a wave of model releases hit just ahead of the Lunar New Year holiday.
Zhipu AI launched GLM-5, an open-source model that the company says surpasses Google’s Gemini 3 Pro on coding and agentic benchmarks, and approaches Anthropic’s Claude Opus in some tests. The stock surged nearly 30% in Hong Kong, closing at HK$405. Zhipu immediately hiked pricing on its GLM Coding Plan by 30%, betting that demand will absorb the increase.
MiniMax followed with M2.5, which it claims is the first production-grade model natively built for agent workflows. With only 10 billion activated parameters, the model supports throughput of 100 tokens per second while keeping inference costs low. MiniMax shares jumped 14%.
The broader context matters as much as the individual models. An MIT Technology Review analysis published Wednesday noted that Chinese open-source models have now surpassed U.S. models in total downloads on Hugging Face. Alibaba’s Qwen family overtook Meta’s Llama in cumulative downloads, and the gap between Chinese releases and the Western frontier has compressed from months to weeks. MIT’s conclusion: expect more Silicon Valley products to quietly ship on top of Chinese open models in 2026.
Sources: CNBC, SCMP, MIT Technology Review
ByteDance’s Seedance 2.0 Launches - and Immediately Gets Partially Shut Down
ByteDance launched Seedance 2.0, its most capable AI video generation model, and within days had to suspend one of its marquee features after a tech reviewer demonstrated it could reconstruct his exact voice and speaking mannerisms from nothing more than a photo of his face.
The model itself is a significant leap: it generates 20-second video clips with audio using a dual-branch diffusion transformer, producing dialogue, ambient sound, and effects synchronized frame-by-frame. ByteDance claims a 90%+ usable output rate. Elon Musk commented “this is happening too fast” on the demo clips.
But the Face-to-Voice feature crossed a line. Reviewer Tim Pan showed that the AI could accurately replicate his specific timbre, cadence, and intonation using only a facial photograph - no audio sample required. The implications for deepfakes and identity theft were immediately obvious. Social media responses ranged from “creepy” to detailed descriptions of abuse scenarios. ByteDance suspended the feature on February 10, imposed live verification protocols, and barred the use of real-person photos and videos as reference subjects for avatar generation.
The episode captures the current moment in AI video generation: the technology is now good enough to be genuinely dangerous, and the companies building it are shipping first and asking questions after the backlash arrives.
Sources: TechNode, No Film School, Social Media Today
Quick Hits
-
Claude’s free tier gets a major upgrade: Anthropic unlocked file creation, app connectors (Slack, Notion, Google Workspace, Figma, and more), and custom skills for all free users. These were previously Pro-only features. Free users also get longer conversations through automatic context compaction. The move came the same week OpenAI began inserting ads into ChatGPT for non-paying users. MacRumors, CNBC
-
Meta breaks ground on $10B Indiana data center: The 4-million-square-foot campus in Lebanon, Indiana will deliver 1 gigawatt of capacity when it comes online in late 2027 or early 2028. Meta is self-funding the project as part of a planned $135 billion AI infrastructure spend in 2026. The site will create 4,000 construction jobs and 300 permanent positions. Yahoo Finance, Meta Newsroom
-
AI already making cybercrime easier - and cheaper: MIT Technology Review published an analysis showing that criminals are using LLMs mainly for productivity: writing malware, crafting phishing lures, conducting reconnaissance, and translating across languages. Microsoft blocked $4 billion in AI-assisted scams in the year through April 2025. At least half of spam email is now LLM-generated, according to researchers at Columbia and the University of Chicago. MIT Technology Review
-
Chinese open-source models overtake U.S. in downloads: Alibaba’s Qwen family has surpassed Meta’s Llama in cumulative downloads on Hugging Face. Chinese firms are releasing frontier-competitive models within weeks of Western launches, and the lag keeps shrinking. MIT predicts more Western products will quietly ship on Chinese open-source foundations this year. MIT Technology Review
Worth Watching
Anthropic’s $30 billion raise is staggering on its own terms, but the strategic positioning is what makes it interesting. The company is simultaneously expanding free access to Claude, running anti-ad messaging, and stockpiling enough capital to sustain a long war of attrition with OpenAI. The bet is that users will choose the platform that doesn’t monetize their attention - and that Anthropic can scale revenue fast enough through enterprise and API customers to never need ads. With annualized revenue at $14 billion and growing, the math might actually work. But $30 billion buys a lot of runway either way.
The Chinese model releases deserve close attention too. When Zhipu can launch an open-source model that credibly benchmarks near Claude Opus and immediately hike its API pricing by 30% - and see its stock jump 30% the same day - that’s a market signaling that the value gap between frontier Western models and Chinese open-source alternatives has narrowed enough to matter commercially. For any company building on top of LLMs, the question of which models to use just got a lot more complicated.