Latest / AI Fire Daily / 🎙️ EP 293: Anthropic Joins $1.8B Climate Coalition & China’s DeepSeek Hits $50B Valuation
Transcript
- 0:00AI is no longer just text on a screen. Beat.
- 0:02It is demanding massive power grids. It is causing
- 0:06global geopolitical friction. And now, beat.
- 0:10It is literally growing hands and legs to walk
- 0:13around the physical world. Yeah, it is wild to
- 0:16see. I mean, we're watching code crash straight
- 0:18into reality here. We are. Two sec silence. Welcome
- 0:21to today's Deep Dive. I am glad you are here
- 0:24with us. We're taking a stack of fresh newsletter
- 0:26sources to explore how AI is breaking out of
- 0:28the software layer and crashing into physical
- 0:30reality. Right. And we've got a lot to cover
- 0:32today. We're looking at Anthropix's massive new
- 0:35climate commitments, for one. And then there's
- 0:38this intense global political fight over an American
- 0:41AI off switch. Plus, Alibaba's new breakthrough
- 0:45in giving AI physical bodies. Before AI can walk,
- 0:49it has to eat. Beat. And right now it is eating
- 0:51the power grid. Oh, absolutely. I mean, training
- 0:53these frontier models requires just immense uninterrupted
- 0:56electricity. Which brings us to Anthropic. They
- 0:59actually just became the first AI startup to
- 1:01join the Frontier Carbon Coalition. I want to
- 1:03break down what this coalition actually does.
- 1:06Beat. Because it was originally founded by major
- 1:08tech players. Companies like Stripe, Google,
- 1:10and Shopify. Yeah, exactly. And it basically
- 1:13acts as a massive vetting machine for... Carbon
- 1:16removal credits. And we're not talking about
- 1:19like planting a few trees here. Anthropic is
- 1:22contributing to a new $915 million funding tranche.
- 1:26Wow. That is a significant amount of capital.
- 1:30It really is. That fresh money nearly doubles
- 1:32Frontier's total pledges to $1 .8 billion. And
- 1:37to date, the coalition has contracted about $700
- 1:39million to remove 1 .8 million tons of carbon
- 1:43across 50 different projects. That feels like
- 1:45a major philosophical shift for Anthropic. They
- 1:48previously leaned on an all -of -the -above energy
- 1:50approach, which naturally implies utilizing some
- 1:53polluting power sources to get that constant
- 1:56baseload power. Right. Because data centers,
- 1:58they just kind of can't wait for the wind to
- 1:59blow. They need 247 power. So this is Anthropic's
- 2:03first major climate deal. And Frontier itself
- 2:05is fundamentally shifting its strategy right
- 2:08now. They are moving away from funding smaller
- 2:10experiments. Yeah, exactly. They are moving away
- 2:13from dozens of small projects. They want fewer
- 2:15massive projects now. The explicit goal is removing
- 2:18a gigaton of CO2 annually on these 8 - to 10
- 2:20-year contracts. A gigaton. Beat. That is one
- 2:24billion metric tons. It is like shifting from
- 2:28planting a few trees in the backyard to building
- 2:30a global industrial scale carbon recycling plant.
- 2:33That is a great analogy. You need direct air
- 2:36capture facilities like giant industrial fans
- 2:39sucking in the atmosphere, chemically binding
- 2:41the CO2 and burying it underground. But to reach
- 2:44that gigaton scale, they're trying to guarantee
- 2:46a market exists so these massive factories can
- 2:48actually get built. Hold on. Beat. There is a
- 2:51strict new rule here that gives me pause. Frontier
- 2:53requires these projects to show a clear path
- 2:55to government subsidy by 2040. Why the strict
- 2:58requirement to hand the ultimate bill to the
- 3:00public sector? Well, that's the critical tension,
- 3:02right? Private tech money can really only kickstart
- 3:04the market. Removing a gigaton of carbon annually
- 3:06requires infrastructure on the scale of, like,
- 3:09the global highway system. Tech companies can
- 3:12de -risk the early technology, sure. But maintaining
- 3:15utility -scale public works over decades. That
- 3:18structurally requires government -level funding.
- 3:21So tech funds the start, but taxpayers must eventually
- 3:24take the wheel. That is the exact gamble they
- 3:27are making. It's a multi -decade baton pass.
- 3:29If governments don't step in, the infrastructure
- 3:31just eventually collapses under its own operating
- 3:34costs. But building these gigaton -scale power
- 3:36grids introduces a new vulnerability. That massive
- 3:40infrastructure is not just about energy. It is
- 3:43driving billions of dollars and creating incredibly
- 3:46high stakes global politics. The geopolitical
- 3:49panic is so palpable right now. I mean, look
- 3:51at the sudden Fable 5 ban overseas. A new letter
- 3:54from the U .S. Commerce Secretary to Dario Amadei
- 3:56just surfaced, giving hidden context on why access
- 3:59to those specific frontier models was frozen
- 4:02so suddenly for international users. And we are
- 4:05seeing immediate pushback on the global stage.
- 4:09At the G7 summit, Emmanuel Macron and Narendra
- 4:12Modi explicitly pushed back against these U .S.
- 4:15restrictions, the same restrictions that froze
- 4:17Claude Fabel and mythos. Right. World leaders
- 4:20are waking up to this harsh reality. They desperately
- 4:22want American A .I. capabilities to drive their
- 4:25economies, but they explicitly do not want an
- 4:28American off switch. They're looking at their
- 4:31critical infrastructure like banking, health
- 4:33care, logistics, running on intelligence hosted
- 4:36in data centers controlled by Washington. We
- 4:38should look at this impartially. The QBES clearly
- 4:40has profound national security concerns regarding
- 4:43these frontier models. Conversely, the G7 partners
- 4:46have entirely valid national sovereignty concerns.
- 4:49They fear sudden economic blackouts. Both sides
- 4:52are operating on fundamentally different yet
- 4:54entirely rational priorities. Two sec silence.
- 4:57Absolutely. And that geopolitical bottleneck
- 4:59is, well, it's forcing a massive hardware pivot.
- 5:02If nations can't rely on cloud APIs, they need
- 5:05local compute. I mean, look at China's response.
- 5:07DeepSeek just raised $7 .4 billion at a $50 billion
- 5:10valuation. Making them China's most valuable
- 5:13AI startup. Beat. And the founder backed it with
- 5:16$3 billion of his own money. Yeah. Liying Wenfeng.
- 5:20It's immense conviction to secure independent
- 5:22compute power. This bottleneck isn't just global.
- 5:26It's hitting everyday consumer devices right
- 5:28now. Apple is facing what industry insiders are
- 5:31calling Ramageddon. Ramageddon, Pete. That is
- 5:34a wonderfully vivid term. It is less wonderful
- 5:36for consumers, though. Apple indicates a massive
- 5:39memory price increase is essentially unavoidable.
- 5:42It could force a $270 price hike across future
- 5:46iPhones, Macs, and iPads. That is a staggering
- 5:49jump for consumer hardware. Why is Apple suddenly
- 5:52facing Remageddon right now? It comes down to
- 5:54how a neural network physically operates. Traditional
- 5:57apps, they just sit memory. You load a photo
- 5:59and you close it. But running a frontier AI model
- 6:01locally means storing billions of parameters.
- 6:04The entire neural network basically has to sit
- 6:06in active, high bandwidth memory simultaneously
- 6:09to generate a single word. Bigger AI models simply
- 6:12demand more memory to run on our personal devices.
- 6:15Exactly. You can't just page AI weights to a
- 6:18slow hard drive. But while hardware gets expensive,
- 6:21the everyday integration of these models is accelerating
- 6:24fast. Have you seen the viral index from artificial
- 6:27analysis? I have. BEAT. It maps top models by
- 6:30capability versus their actual running cost.
- 6:33Yeah, and it proves a vital point. Not all frontier
- 6:35models are expensive. And not all expensive models
- 6:38are necessarily the best. The deployment is getting
- 6:41incredibly smooth. Beat. Anthropic's Applied
- 6:44AI team showed how cloud -managed agents are
- 6:47now running in production. Right. They are no
- 6:49longer theoretical. Yeah. And the competition
- 6:51is driving massive value. Copilot Cowork is now
- 6:54generally available. Half of the Fortune 500
- 6:56already use it. And it claims to be 30 % to 40
- 6:59% cheaper than Cloud Cowork. Beat. It is embedding
- 7:02into everything. Grok is now a free add -in for
- 7:05PowerPoint. Yeah, it takes a rough outline and
- 7:08just generates a research deck. And Google just
- 7:10launched a new $99 home speaker that handles
- 7:13completely messy commands. Like, you can stutter,
- 7:15change your mind mid -sentence, and it just follows
- 7:17the chaotic human logic perfectly. I have a vulnerable
- 7:20admission to make here. Beat. I still wrestle
- 7:23with prompt drift myself. Seeing how smoothly
- 7:25these new tools handle messy, real -world human
- 7:29workflows is inspiring. Oh, me too. It really
- 7:32shows we're moving past the era of rigid prompt
- 7:34engineering. We aren't learning to speak to the
- 7:36machine anymore. The machine is finally learning
- 7:39to speak to us. But handling these massive local
- 7:41models requires totally new infrastructure. Beat.
- 7:45Developers are building a totally new software
- 7:47layer purely for AI agents. Yes, the tooling
- 7:50ecosystem is just exploding right now. Look at
- 7:52the new Minimax M3. It's an open -weight model
- 7:55that combines frontier -level coding, a 1 million
- 7:58token context window, and multimodal support.
- 8:01For the listener who isn't a developer, beat,
- 8:03what does an open -weight model actually mean?
- 8:06Think of a closed model like a baked cake. You
- 8:09can eat it, but you can't unbake it to change
- 8:11the recipe. An open -weight model gives you the
- 8:13exact ratio of flour, sugar, and eggs. It gives
- 8:17you the neural network's parameters so you can
- 8:18tweak the recipe for your own needs. Anyone can
- 8:21download and modify the core model architecture
- 8:23for their own specific needs. Perfectly. And
- 8:26that customizability is empowering developers
- 8:28to build wild new platforms. Take Cursor's Origin.
- 8:32It is a Git hosting platform built specifically
- 8:35for AI agents. Code management designed for bots.
- 8:39Beat. The documentation says it allows automated
- 8:42conflict resolution. How does that actually work
- 8:44without a human? Well, if Agent A changes the
- 8:47database schema and Agent B builds a new website
- 8:50button, Origin understands the purpose of both.
- 8:53It automatically rewrites the button's code to
- 8:56match the new database schema without a human
- 8:58ever mediating. It's true agent -to -agent collaboration.
- 9:02That is fascinating. Beat. Then there's Hermes
- 9:04Agent. It adds official Stripe skills directly
- 9:06into the model. Yeah, to let agents make real
- 9:08financial payments, they can launch their own
- 9:10SaaS product safely. And look at Crilly. It's
- 9:12finding creators based on real -time habits and
- 9:15moving conversations into structured deal workflows
- 9:17automatically. Giving agents the ability to negotiate
- 9:20legal deals or swipe a credit card. Beat. It
- 9:23feels like a massive threshold crossed. It definitely
- 9:25raises the stakes. I mean, we are building an
- 9:27autonomous economic layer. So agents finally
- 9:30understand the purpose of our code and our money.
- 9:33They do. And the thresholds just keep falling.
- 9:35We're crossing the biggest one right now. Midroll
- 9:38sponsor read. We have seen AI consume global
- 9:41power grids. Beat. It has disrupted international
- 9:44geopolitics. It is managing our code and our
- 9:47money. Now, beat. Alibaba is finally giving it
- 9:50a physical body. This is where the science fiction
- 9:52becomes our daily reality. Alibaba just unveiled
- 9:56the Quinn Robot Suite to achieve physical world
- 9:58intelligence. They broke this massive problem
- 10:00down into three specialized foundation models.
- 10:03Beat. First, there is Quinn Robot Manip. Right,
- 10:06basically the hands. It's trained on over 38
- 10:09,000 hours of pure robotics data to translate
- 10:12visual inputs and natural language directly into
- 10:14precise physical manipulation. It learns how
- 10:17to grasp, turn, and pull. Next is Quinn Robotnov.
- 10:20Beat. These are the legs. Exactly. It handles
- 10:23complex spatial reasoning to navigate unseen
- 10:25environments. It avoids obstacles it has never
- 10:27encountered before. Finally, there's Quinn Robot
- 10:30World. Beat. This is the simulator, a video world
- 10:34model. And this is the real breakthrough. Traditional
- 10:37robots learn by physical trial and error, which
- 10:40means they fall over constantly. The simulator
- 10:42model uses real -time visual data to predict
- 10:45future trajectories in a digital space. How does
- 10:48the simulator model change the way robots learn?
- 10:51It allows the AI to test physics digitally. The
- 10:54AI watches a video of a cup falling, learns the
- 10:57implicit rule of gravity, and simulates the physics
- 11:00in its mind. It runs countless virtual iterations
- 11:03before moving a single physical servo. It lets
- 11:05robots safely practice physical tasks in a digital
- 11:08dream world first. Yes, a digital dream. And
- 11:11according to Alibaba, This framework excels at
- 11:14out -of -distribution generalization. Adapting
- 11:17to totally new, unexpected situations without
- 11:20explicit training. Right. If you drop it in a
- 11:22kitchen, it is never seen. It can still open
- 11:25a drawer it has never encountered. It's not just
- 11:27executing a hard -coded script, it's exhibiting
- 11:29genuine spatial exploration. Whoa. P -Land. Imagine
- 11:33scaling to a billion queries. Those queries are
- 11:35robots making physical decisions. Two secs, silence.
- 11:38It shifts our entire relationship with technology.
- 11:41AI is stepping out of the browser and into the
- 11:43living room. Let's pull all these threads together.
- 11:46The big takeaway today is the physicalization
- 11:48of AI. Beat. This technology started as simple
- 11:52chatbots. Today, we saw the entire physical pipeline
- 11:55emerge. It requires immense physical energy,
- 11:58driving 1 .8 million tons of carbon removal through
- 12:02the Frontier Coalition just to power the data
- 12:04centers. Then it hits our physical hardware.
- 12:07Beat. It triggers a ramageddon on our personal
- 12:09phones, potentially forcing a $270 hardware tax.
- 12:13Which escalates into physical geopolitics. G7
- 12:16leaders pushing back against American off switches
- 12:18because their local compute is constrained. And
- 12:20ultimately, it steps directly into reality. Beat,
- 12:24Alibaba's Quinn Foundation models are driving
- 12:26robotic hands and legs. AI is no longer a software
- 12:30abstraction. It is sharing our physical space
- 12:32now. I want to leave you with a new question
- 12:34to mull over. Beat. If an autonomous robot with
- 12:38physical world intelligence causes damage in
- 12:40a foreign country, but its actions were suddenly
- 12:43frozen by an American off switch built into its
- 12:46underlying frontier model, who ultimately holds
- 12:48the legal liability to sex violence? That is
- 12:51the multi -trillion dollar question we have to
- 12:53answer next. Thank you for joining us on this
- 12:55deep dive. Beat, keep questioning everything
- 12:57you see. We will see you next time. Outro music.