☀️ TRENDING AI NEWS 🤖 Salesforce x Nvidia: New open-weight Koa reasoning model targets sales, marketing, and customer support tasks. 🏢 Anthropic: Signed a deal for a $31.9B data center in western Queensland, Australia. 🛠️ Meta One: New AI subscription bundles now live globally across Facebook, Instagram, and WhatsApp. 🚨 AI Slowdown: Jensen Huang told Trump on stage 'we're not going to let an AI slowdown happen.' |
Picture this: the CEO of the world's most valuable chip company takes a live phone call from the President of the United States - on stage, in front of a packed conference room - and uses it to publicly torpedo an idea that three of his biggest customers spent all weekend championing. That's roughly what happened this week, and it tells you everything about the fault lines opening up inside the AI industry right now.
🤓 AI Trivia
Anthropic's new Australian data center deal is valued at $31.9 billion. But roughly how many parameters does Anthropic's flagship Claude 3 Opus model reportedly have?
🔢 34 billion
🔢 70 billion
🔢 2 trillion
🔢 No figure has been publicly confirmed
The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇
| ⚡ Jensen Huang Breaks Ranks on AI Safety - and Does It on a Presidential Phone Call | |
Nvidia's CEO picks a side, loudly
While Anthropic's Dario Amodei and OpenAI's Sam Altman have been calling for a coordinated slowdown in AI development, Nvidia CEO Jensen Huang went in the opposite direction this week. At the All-In Summit, Huang took a live call from President Trump on stage and told him directly: 'We're not going to let an AI slowdown happen.' Huang's position is that AI is not some alien threat - it's hardware and software, and safety should be engineered into products by the companies building them, not mandated by governments.
His argument has a clear commercial logic behind it. Nvidia sells the GPUs that power every major AI lab. A slowdown in model development is directly a slowdown in chip sales. But Huang's technical point is also worth engaging with: unlike nuclear weapons, AI systems can be updated, patched, and improved incrementally. He argues that regulatory frameworks risk locking in bad architecture decisions before anyone knows what actually needs fixing.
The safety-vs-speed divide goes public
Meanwhile, Anthropic, OpenAI, and Google DeepMind have been in active safety talks for weeks, according to TechCrunch. The three labs are discussing frameworks for audits and pacing agreements - even as Trump's White House dismisses those concerns and says development speed is a national security priority. The AI industry hasn't looked this publicly divided in years.
The bottom line
Nvidia's chip revenue depends on AI acceleration, which makes Huang the most powerful voice against any slowdown - and the one with the most to lose if pacing agreements actually stick.
| 🌍 China Isn't Playing Ball on the Safety Pact Either | |
Beijing sees a slowdown as a US competitive strategy, not a safety measure
If you've been following the AI regulation debate, here's the wrinkle nobody is solving: China. According to Wired, Beijing is deeply skeptical of any international agreement to slow AI development - not because China doesn't believe AI poses risks, but because it suspects the whole framework is designed to lock in American dominance at a moment when Chinese labs are closing the gap.
Both the US and China agree, at least nominally, that advanced AI carries serious risks. But from Beijing's perspective, a deal that prioritizes 'safety' while US companies remain at the frontier looks less like a safety pact and more like a cartel. That skepticism isn't entirely unfounded - The Verge ran a piece this week asking exactly that question about the big labs' slowdown agreement.
No multilateral deal, no global floor
Without China at the table, any slowdown agreement between Western labs is really just a voluntary gentlemen's agreement - and history suggests those don't hold for long when billions in revenue are at stake.
| 🏢 Meta Puts a Price Tag on AI Access with New One Subscriptions | |
Bundling social media and AI into one monthly bill
Meta launched its Meta One subscription bundles this week, rolling them out globally across Facebook, Instagram, and WhatsApp. The tiers are aimed at individuals, creators, and businesses, and they bundle expanded access to Meta's AI tools with premium features across its apps. Critically, Meta says the 'core experience' on its apps remains free - this is about unlocking higher limits and extra AI capabilities, not paywalling the basics.
The timing makes sense. Meta launched its do-everything AI assistant Muse just before rolling out these bundles, and now it has a clear upsell path. Developers also got something this week: a new WhatsApp Business MCP server that lets AI coding agents like Claude, Cursor, and ChatGPT handle WhatsApp Business setup, messaging templates, and troubleshooting automatically.
The bottom line
Meta is quietly building the same kind of AI subscription flywheel that Microsoft and Google have, and the WhatsApp Business MCP move means developers now have a faster path to building agents on top of the world's largest messaging platform.
| 🔬 OpenAI Is Paying to Build Biology Training Data From Scratch | |
Buying failed biotech data to fill a gap in medical AI
AI models are data-hungry, and biology is one of the areas where high-quality training data is hardest to find. OpenAI is now funding a novel solution: acquiring detailed regulatory filings, manufacturing strategies, and safety data from failed biotech companies - the kind of information that is normally locked up as trade secrets but becomes available when companies go bankrupt.
The idea originated with policy analyst Ruxandra Teslo, who proposed that by bidding at biotech bankruptcy proceedings, AI developers could get access to rich, real-world biological datasets that would otherwise never see the light of day. It's a creative approach to a genuine bottleneck: healthcare AI systems are only as good as the biology data underneath them, and most of the best biology data is proprietary.
The bottom line
If this works at scale, it could meaningfully accelerate drug discovery and clinical AI - and it gives OpenAI a dataset moat in medical AI that competitors will struggle to replicate quickly.
| ⚠️ 61% of Voters Oppose AI Data Centers - and Politicians Are Listening | |
Public opinion is turning into a political liability for the AI industry
A New York Times and Siena University poll of 1,503 likely voters found that 61% oppose building new data centers to power AI - and that polling was conducted before the recent wave of AI safety panic added even more negative press to the industry. In Philadelphia, officials are now pushing back against data center construction in a neighborhood already scarred by a defunct oil refinery, per TechCrunch.
The environmental impact angle is becoming harder to ignore. Separate research shows that US data centers could consume more natural gas than Germany and Japan combined by 2035 if current AI buildout rates continue. That is the kind of statistic that lands in a campaign ad, not just a policy brief.
On the infrastructure side, if you're building something AI-powered and want a fast web presence without spinning up a full data center contract, it's worth checking out
On the infrastructure side, if you're building something AI-powered and want a fast web presence without the overhead, 60sec.site lets you build and launch an AI-generated website in under a minute - no data center drama required.
The bottom line
When 6 in 10 likely voters oppose your core infrastructure, you have a political problem - and the AI industry is only now starting to reckon with how fast that sentiment can translate into local zoning battles and midterm campaign rhetoric.
| 🌎 Trivia Reveal | |
The answer is: No figure has been publicly confirmed! Anthropic has never officially disclosed the parameter count for Claude 3 Opus. Unlike some competitors, Anthropic keeps its model architecture details private - which makes the $31.9 billion Australia data center deal all the more striking as a signal of just how seriously they're scaling up, even without sharing the technical specs publicly.
| 💬 Quick Question | |
Jensen Huang says AI safety should be left to product makers, not governments. Dario Amodei says the industry needs to slow down and coordinate. Where do you land? Hit reply and tell me - I genuinely read every response, and I'm curious whether readers are more worried about moving too fast or too slow right now.
That's all for today - see you tomorrow with more from the AI frontier. And if you want to dig into any of these topics further, everything we've covered is searchable over at dailyinference.com.