China is reportedly preparing a $295 billion plan to build massive AI data centers across the country. This isn’t just another infrastructure project, this is a full-scale national bet on compute dominance. If executed, it would dramatically accelerate China’s ability to train and run large-scale AI models and agents locally, reducing reliance on foreign chips and cloud services. What this means long-term: - The global compute war just entered a much more intense phase. - We’ll likely see faster proliferation of AI agents and multimodal models coming out of China. - Supply chain pressure on GPUs, TPUs, energy, and data center components will only get crazier. The age of “whoever controls the most compute wins” is here. And China is no longer playing catch-up, they’re going all-in. This move could reshape the entire AI landscape in the next 3–5 years.

Anthropic just dropped a serious policy package alongside CEO Dario Amodei’s new essay. They’re launching: - A $200M fund focused on AI labor research - A $150M national fellowship program - And two other major initiatives While most AI labs are racing purely for capability, Anthropic is putting real money into understanding and shaping how AI will impact the workforce and society. This feels like a mature move. They’re not just building powerful models, they’re actively trying to influence the broader consequences of the technology they’re creating. Whether you agree with their direction or not, credit where it’s due: they’re thinking several steps ahead. The era of “build fast and figure out the societal impact later” is slowly ending.

Google just released Gemini 3.5 Live Translate, their new real-time speech-to-speech translation model. It supports over 70 languages, starts translating the moment you speak, and maintains natural tone, pacing, and flow with very low latency. This isn’t the old “speak then wait for translation” anymore. This is fluid, near-simultaneous conversation across languages. The barrier of language in real-time communication just got lowered dramatically. Whether it’s international meetings, travel, or global collaboration, the way we connect with people who speak different languages is about to feel much more natural. This is one of those quiet but extremely impactful releases. The agent communication layer is evolving fast

The AI price war just escalated again. According to WSJ, OpenAI is seriously considering drastic price cuts on its token costs to stay competitive with Anthropic. This is getting interesting. After years of premium pricing, the frontier labs are now fighting hard for users and developers. The era of “our model is smarter so we can charge more” is being challenged by “our model is good enough and much cheaper.” We’re moving from a quality-only game to a quality price/performance game. This is ultimately bullish for builders and users. Lower costs = more experimentation, more agents, and faster adoption. The race isn’t just about who has the smartest model anymore, it’s about who can deliver the best intelligence at the best price.

Claude Fable 5 just dropped and it’s quietly cooking. In the latest benchmarks, Fable 5 / Mythos 5 is absolutely dominating: - Agentic coding - Complex knowledge work - Spatial reasoning - Legal understanding - Multidisciplinary reasoning - Even Cybersecurity and Health benchmarks It’s not just beating Opus 4.8, it’s putting noticeable distance between itself and GPT-5.5 & Gemini 3.1 Pro in several key areas. Google and OpenAI are still going for raw intelligence, while Anthropic is clearly optimizing hard for agentic performance and real-world usefulness. The gap in usable intelligence is widening. Fable 5 feels like the first model truly built for the agent era. This is impressive. Google just released Gemini 3...

Google just released Gemma 4 12B, a unified, encoder-free multimodal model designed to run locally on your laptop. Key highlights: - Runs smoothly on just 16GB VRAM - Strong reasoning performance, close to their 26B model but at half the memory - Native multimodal (vision language) with a lightweight 35M vision module - Fully open source under Apache 2.0 This is huge for builders. Real agentic workflows and advanced intelligence no longer require a massive GPU cluster. Local AI just got a serious upgrade. The edge era is here.

Meta just launched Meta Business Agent: an AI that handles customer service 24/7 across Messenger, WhatsApp, and Instagram. The good: Businesses can now scale support and operations dramatically with almost zero extra headcount. Real efficiency unlock for SMEs. The concerning: We’re steadily replacing human customer service with AI at massive scale. While it saves costs, it also risks turning customer experience into something colder, more robotic, and occasionally hallucinated. This is the double-edged sword of the agent era in full display. Productivity goes up. Human connection? That’s another conversation.

The brutal reality of the AI gold rush right now: Big Tech has spent a combined $1.4 Trillion on AI… but the entire industry has only generated $613B in revenue so far. Amazon: spent $313B → made $22B Google: spent $287B → made $25B Microsoft: spent $266B → made $31B Meta: spent $230B → still very little AI revenue Almost everyone is bleeding cash to build infrastructure as fast as possible. The only clear winner? NVIDIA, sitting on ~$478B AI revenue and over $250B in cumulative profit. Everyone else is still in heavy investment mode. The AI infrastructure layer is printing money, while most application & model layers are still deep in the “spend now, profit later” phase. This divide is becoming clearer every quarter

Microsoft just dropped Scout: a real AI executive assistant that lives directly inside your email and calendar. It doesn’t just reply to messages. It acts like an actual chief of staff: manages your schedule, drafts emails, makes decisions, and shows up exactly where you need it. We’ve officially gone from “AI helps you work” to “AI works like a real employee beside you.” The agent layer is no longer experimental. It’s entering the workplace. This is actually huge fr

Bernie Sanders just dropped a nuke: He’s proposing that companies like OpenAI, Anthropic, and xAI should hand over 50% of their equity to a federally managed public fund. Basically saying the American people should own half of the frontier AI labs. Whether you see this as visionary wealth redistribution or government overreach on steroids, one thing is undeniable: The AI era is forcing society to have very serious conversations about ownership, power, and economic structure much faster than anyone expected. This is going to be a wild debate

Google DeepMind, Anthropic, and Meta are now seriously researching whether AI can become conscious. We’ve officially moved from “make AI smarter” to “can AI actually wake up and know it exists?” This isn’t sci-fi anymore. Some of the biggest AI labs in the world are actively exploring machine consciousness. The day we have to ask “Is this AI sentient?” is coming faster than most people realize. Mind-blowing stuff

SoftBank just dropped a monster move: $87 billion (75 billion euros) into AI data centers across France. This is reportedly the largest single AI infrastructure investment ever announced in Europe. The plan includes up to 5 gigawatts of compute power, with the first phase already at 45 billion euros focused on major sites in northern France. Europe is no longer sitting on the sidelines. The global compute war just got significantly bigger. This is the kind of scale we’ll need if the agent economy predictions are even half right

A developer just deleted 3 months of AI-generated code because he realized he didn’t actually understand his own project. He shipped fast, felt super productive, everything looked clean… until he had to add one real feature and couldn’t trace the logic. This is the silent trap of over-relying on AI. Speed is addictive, but real ownership and deep system understanding still can’t be fully outsourced. The best builders will use AI as a 10x multiplier, not a replacement for thinking.

OpenAI and Anthropic are pouring billions into enterprise partnerships and forward-deployed infrastructure plays. The model layer is getting crowded and they know it. Morgan Stanley charted the same pattern during mobile internet. Semis ran first. Infra and devices followed. Then software and services dominated for years and made every earlier wave look small. We're in the infra phase of AI right now. The labs are locking up enterprise distribution before the application layer runs because whoever owns the distribution channel into enterprise owns the moat, not whoever has the best model. The uncomfortable read for builders: the labs going upstream into enterprise accounts aren't just your API provider anymore. In some verticals, they're going to be your competitor.

The AI price war is getting absolutely brutal. Xiaomi just cut MiMo V2.5 Pro prices by 99%. DeepSeek made V4 Pro permanently cheaper too. Chinese labs are basically giving away high-level intelligence at this point. The real question isn’t how good the models are anymore, it’s how the hell are they making money at these prices? This price war is wild.

Introducing Claude Opus 4.8: it builds on Opus 4.7 with sharper judgment, more honesty about its own progress, and the ability to work independently for longer than its predecessors. Available today at the same price.

A top UK law firm, Pinsent Masons, was publicly reprimanded by London’s High Court after submitting AI-generated content containing inaccuracies without any human verification. Lawyers literally took AI output, hallucinated facts and all, and presented it directly to a judge. This is exactly why “AI is just a tool” narrative exists. Powerful, but still needs serious human oversight. The legal industry is learning the hard way that blindly trusting AI can get expensive real quick.

Today we’re announcing our $113M Series B led by @CapitalGVC. Over the last 6 months, weekly volume on OpenRouter grew from 5T to 25T tokens as AI rapidly shifts from experimentation into production. We’re excited for what comes next.

Goldman Sachs is predicting AI agents will consume 120 quadrillion tokens per month by 2030. That's a 24x increase from today. Let that sink in for a second. We're not in "interesting tech trend" territory anymore. The compute appetite we're about to see from autonomous agents running 24/7, spinning up sub-agents, calling tools, retrying failures... it compounds in ways that are genuinely hard to model. Token demand is about to become one of the most important numbers in tech.

Google DeepMind’s AI agent just autonomously solved 9 Erdős problems including two that had remained unsolved for 56 years. Out of 353 of the hardest open problems in mathematics. This isn’t just “AI is getting better at math”, AI starting to push the boundaries of human mathematical knowledge on its own.