Latest / AI Fire Daily / 🎙️ EP 273: Nvidia's $81B Record Quarter & OpenAI's Historic Erdős Math Proof
Transcript
- 0:00Think about the assumptions we make every single
- 0:02day. Beat, we spent decades thinking AI was just
- 0:06a chat bot. Yeah, just a simple digital box.
- 0:08Right, you type a prompt, you get text back.
- 0:10But what happens when that chat bot finally stands
- 0:12up? It starts acting like a digital employee.
- 0:16Exactly. or when it autonomously solves 80 -year
- 0:19-old math problems. These are puzzles that baffled
- 0:22the absolute brightest human geniuses. It completely
- 0:25changes our relationship with this technology.
- 0:28I mean, the ground is shifting right beneath
- 0:29our feet. Welcome to this deep dive for you today.
- 0:32We are looking at a stack of incredibly fresh
- 0:35sources. Really exciting stuff. They map out
- 0:38the next era of artificial intelligence. We are
- 0:41going to trace the massive hardware shift happening
- 0:44right now. We'll explore the mind bending billions
- 0:47trading hands between top labs. We'll examine
- 0:50the tools turning AI into proactive workers.
- 0:52And finally, we'll look at a historic breakthrough
- 0:54in autonomous reasoning. It is a fascinating
- 0:56blueprint of what comes next. So to truly understand
- 0:59how AI is getting smarter. We have to start physically.
- 1:02Right at the foundation. Yeah. We have to look
- 1:05at the underlying hardware. And we really need
- 1:07to follow the money moving behind it. Right.
- 1:09And the numbers here are absolutely staggering.
- 1:12NVIDIA just posted another record -breaking financial
- 1:15quarter. What are we looking at? They reported
- 1:17$81 .6 billion in revenue. That is up 20 % from
- 1:22the previous quarter. That is a massive leap
- 1:24in just... Three months. Yeah, it really is.
- 1:27Their data center business alone hit $75 .2 billion.
- 1:31Wow. That shows infrastructure demand is still
- 1:34incredibly strong. The physical foundation is
- 1:37still being rapidly built. I want to pause on
- 1:39the financial flexing here. They approved $80
- 1:43billion in share buybacks. What does that signal
- 1:46to the broader tech market? Well, it signals
- 1:49massive internal confidence in their future growth.
- 1:53They are telling the market they have plenty
- 1:55of cash. Right. But what's even more interesting
- 1:57is their venture capital strategy. They hold
- 1:59$43 billion in private startup stakes right now.
- 2:02Exactly. They bought $18 .5 billion this quarter
- 2:06alone. That is wild. They aren't just selling
- 2:09chips to random companies anymore. They are actively
- 2:11funding the exact startups that buy their hardware.
- 2:15It creates this massive self -sustaining financial
- 2:18flywheel. Yeah, it does. That's brilliant, but
- 2:21it can't accelerate forever, right? No, probably
- 2:23not. Forecasts predict $91 billion for the next
- 2:27quarter, but their overall growth rate may slow
- 2:29to 12%. Plus, their China export revenue remains
- 2:34highly uncertain right now. That brings us to
- 2:37the biggest pivot in their strategy. Jensen Huang
- 2:40just announced a new $200 billion market. This
- 2:43is huge. It centers entirely around a new CPU
- 2:46called Vera. This is a fundamental shift in computing
- 2:49architecture. Vera is built specifically for
- 2:52agentic AI workflows. Let's define that term
- 2:55for everyone listening. Agentic AI is AI that
- 2:58takes actions independently to achieve a goal.
- 3:00Perfect definition. Beat. Think of this shift
- 3:03like a busy corporate office. The GPUs are the
- 3:06creative brainstormers in the room. Right. They
- 3:08generate huge ideas and process heavy data. Exactly.
- 3:11But CPUs like Vera are the actual project managers.
- 3:15They're the ones executing the complex daily
- 3:18tasks. That analogy is spot on. NVIDIA isn't
- 3:21just selling GPUs to researchers anymore. They
- 3:24want to power the full AI stack entirely. From
- 3:28massive data centers down to proactive autonomous
- 3:31agents. Huang says Vera has already generated
- 3:34$20 billion in sales. That is just in this current
- 3:38year alone. It proves the market demand. for
- 3:40agentic hardware is real. Future AI agents will
- 3:43deeply need these CPU -like systems. Why is that?
- 3:47Because they need them to use external tools
- 3:49efficiently. They have to run sequential tasks
- 3:52without constant human intervention. How does
- 3:54NVIDIA stepping into CPUs affect the rest of
- 3:56the industry? Well, they aren't just selling
- 3:59individual processing parts anymore. They want
- 4:01to own the entire operational architecture. Right.
- 4:03They are building the infrastructure for autonomous
- 4:06digital workers. So they own the brains and the
- 4:08hands now. Exactly. They are the foundation of
- 4:10the new economy. But if they are supplying the
- 4:12brains and the hands the energy required to power,
- 4:15that must be astronomical. Oh, it is. Who is
- 4:18actually footing that massive computing bill?
- 4:22The operational costs are becoming almost incomprehensible
- 4:25today. Let's look at Excel. for a very clear
- 4:28example. Okay. They burned $6 .4 billion last
- 4:31year. That was on just $3 .2 billion in revenue.
- 4:34The math there is incredibly brutal to look at.
- 4:36It is. And a recent SpaceX IPO filing revealed
- 4:40something even wilder. What was it? It detailed
- 4:43Grok's plan for a multiple trillion parameter
- 4:45model. A trillion parameter model has a trillion
- 4:48adjustable internal data connections. Yeah. That
- 4:51ambitious plan will make the compute bill substantially
- 4:54bigger. But here is the wildest twist in our
- 4:57sources. I know what you're going to say. Anthropic
- 4:59is reportedly paying XAI directly for compute
- 5:02power. They are paying $1 .5 billion to Sex Silence
- 5:06per month. Yeah. That is for access to Colossus
- 5:09Compute through May 2029. Whoa. Imagine scaling
- 5:13to a billion queries. The power required to run
- 5:16an entire data center just for one model. That
- 5:19is the reality of an industrial revolution scale.
- 5:21You are paying for miles of servers and gigawatts
- 5:24of power. It is a staggeringly expensive physical
- 5:28footprint. Yet despite these massive infrastructure
- 5:30costs, Anthropic expects its first profitable
- 5:34quarter very soon. Right. Their revenue is...
- 5:36jumping to around $10 .9 billion. That is a massive
- 5:40milestone for the entire tech industry. It proves
- 5:43the underlying business model can actually work.
- 5:45It dropped right alongside rumors of an open
- 5:47AI IPO. The market is maturing incredibly fast
- 5:51right now. Extremely fast. How does a company
- 5:54possibly survive burning billions a year? It's
- 5:57a massive infrastructure land grab right now.
- 6:00The sheer utility of the resulting models is
- 6:02the price. Right. That utility will eventually
- 6:05outweigh the massive upfront cost. We see this
- 6:08clearly with Anthropic finally hitting profitability.
- 6:11We must spend billions today to make trillions
- 6:13tomorrow. Exactly. The initial capital expenditure
- 6:16is the only way forward. So what incredible utility
- 6:19justifies paying a billion dollars monthly? It
- 6:21is the fundamental shift from generating text
- 6:23to taking action. We are looking at the toolkit
- 6:26for these new digital workers. The ecosystem
- 6:29of agentic tools is exploding right now. And
- 6:32it completely changes how we interact with software.
- 6:36EXA just raised $250 million. That is at a $2
- 6:40.2 billion valuation. Yeah. They are simply building
- 6:44a search engine for AI agents. This is a crucial
- 6:47piece of the puzzle. Agents don't browse websites
- 6:51like human beings do. Right, they can't see the
- 6:53screen. Exactly. They can't navigate flashy pop
- 6:56-ups or read layout designs. They need raw...
- 7:00perfectly structured data to operate effectively.
- 7:02So EGZA provides that exact structured data layer
- 7:05for them. Yeah. And developer demand for this
- 7:07is just huge. Google is also making massive updates
- 7:09to their tool ecosystem. Gemini Omni is generating
- 7:12entire new video scenes now. It's amazing. It
- 7:15uses mixed inputs to create high quality outputs.
- 7:18You can combine images, audio, video, and text
- 7:21seamlessly. Yeah. It is completely changing the
- 7:24creative production pipeline forever. Google
- 7:26AI Studio is taking this even further today.
- 7:28It turns simple text prompts into native Android
- 7:31apps. The traditional barrier to software coding
- 7:33is rapidly disappearing. Then there's anti -gravity
- 7:372 .0 to discuss. It is turning AI into a real
- 7:41developer workspace. This one is wild. They feature
- 7:44parallel agents and background coding tasks seamlessly.
- 7:48We should definitely explain what parallel agents
- 7:50actually do. Parallel agents are multiple AI
- 7:52programs working on tasks simultaneously. Beat,
- 7:56I still wrestle with prompt drift myself. Oh,
- 7:59same here. That's when AI slowly loses focus
- 8:02on your original instructions. So the idea of
- 8:05parallel agents managing whole workflows feels
- 8:08wild. It is a lot for human beings to adapt to.
- 8:11We're moving from prompting to managing a digital
- 8:14staff. That is probably why Andrew Nehring released
- 8:17a free course. It is a two and a half hour masterclass
- 8:20on AI prompting. Very timely. It is designed
- 8:22to help humans keep up with these tools. Prompting
- 8:25is quickly becoming a necessary professional
- 8:27management skill. We also have to talk about
- 8:29proactive actions. Gemini Spark handles proactive
- 8:31actions across Google products natively. Right.
- 8:34It can manage tasks under your direction entirely
- 8:36in the background. But having proactive agents
- 8:39doing things. automatically raises massive concerns.
- 8:43You need serious security frameworks before you
- 8:45deploy them. Absolutely. OpenAI is adopting Google's
- 8:48SynthID watermarking for generated images. They
- 8:52are also building a public verification checker
- 8:54very soon. We really need to know what's real
- 8:56and what's AI. We are seeing movement on the
- 8:59regulatory front, too. The European Commission
- 9:01released draft guidance on high -risk AI. It
- 9:04clarifies exactly which systems fall under stricter
- 9:07rules. Now we have impending U .S. executive
- 9:09orders coming soon. OpenAI and Anthropic are
- 9:12discussing voluntary frameworks as well. They
- 9:15are talking about model sharing directly with
- 9:17the government. It's all moving so fast. So how
- 9:19do we prevent chaos if agents are acting proactively?
- 9:24We have to rely on strict verification systems.
- 9:26The strict necessity of Google SynthEye is a
- 9:29vital start. Yeah. We also need those incoming
- 9:31EU and US regulatory frameworks. Trust requires
- 9:35built -in digital watermarks and strict guardrails.
- 9:37Right. And without those guardrails, the... enterprise
- 9:39utility drops to zero. We've talked extensively
- 9:42about the physical hardware and the money. We
- 9:45explored the powerful new agentic tools being
- 9:48built. But the ultimate test of this era is pure
- 9:52reasoning. Can these machine learning models
- 9:55actually think deeply? This is the most profound
- 9:57question in the field today. OpenAI just announced
- 10:01a truly historic mathematical breakthrough. GPT
- 10:04-5 .5 autonomously created a completely original
- 10:07geometric proof. It's incredible. It disproves
- 10:10a famous geometry conjecture by Paul Erdos. He
- 10:14proposed this specific riddle back in 1946. This
- 10:18mathematical problem stood unsolved for nearly
- 10:2180 years. Mathematicians strongly believed the
- 10:23best solution would involve square grids. But
- 10:26the AI model found an entirely new geom... construction.
- 10:29Right. Crucially, this wasn't a specialized math
- 10:31tool at all. It was a general purpose reasoning
- 10:34model exploring the logic. It worked through
- 10:36the complex logical steps completely on its own.
- 10:39Wait, hold on. I have to push back a little bit.
- 10:41Okay, go ahead. Didn't this exact same thing
- 10:43happen late last year? OpenAI faced intense public
- 10:46skepticism back then. They did. People found
- 10:49the models were just surfacing existing human
- 10:51work. They weren't actually solving brand new
- 10:54geometric problems. Why should we believe this
- 10:57time is any different? Yeah, and that skepticism
- 11:00was completely justified back then. But this
- 11:03is exactly why this specific moment matters.
- 11:06How so? OpenAI came incredibly prepared for the
- 11:09public pushback this time. They brought the receipts?
- 11:12They absolutely did. They didn't just drop a
- 11:14press release. They subjected the model's output.
- 11:18to rigorous human peer review. They included
- 11:20supporting remarks from highly respected mathematicians
- 11:23today. People like Noga, Alon, and Melanie Wood
- 11:26reviewed the work. And Thomas Bloom, who runs
- 11:29the Erdin's Problems website itself. These experts
- 11:32validated that the logical steps were genuinely
- 11:35novel. This proves the model generated completely
- 11:37new mathematical logic. Wow. It didn't just memorize
- 11:40old data from the internet. Does this mean AI
- 11:42is now better at math than humans? It's not really
- 11:46about being quote unquote better than us. It
- 11:48is about AI being capable of genuine logical
- 11:50discovery. It works right alongside human experts
- 11:53to find objective truth. AI is a peer now, not
- 11:57just a calculator. Exactly. It fundamentally
- 11:59changes how scientific discovery will happen.
- 12:02Sponsor. Let's bring all these incredible pieces
- 12:04together for you. The era of the simple chatbot
- 12:07is truly over today. From NVIDIA's Verichips
- 12:10to billion -dollar compute infrastructure deals,
- 12:14the entire tech world is fundamentally re -architecting
- 12:18for proactive agents. We are arming general -purpose
- 12:21reasoning engines with massive hardware capabilities.
- 12:24They are smart enough to solve 80 -year -old
- 12:26math puzzles autonomously, and we are giving
- 12:29them tools to operate natively inside workflows.
- 12:31Right. They are quickly becoming our digital
- 12:33colleagues and active project managers. It is
- 12:35a profound shift in how humans interact with
- 12:38machines. I want to leave you with a final provocative
- 12:41thought today. If a general purpose AI model
- 12:44can look at an 80 -year -old geometry problem
- 12:46and find a completely new angle that human geniuses
- 12:50missed for decades, what longstanding assumptions
- 12:53in your own industry is an AI agent about to
- 12:56completely rewrite? It is a thrilling, slightly
- 12:59terrifying question to consider. Thank you for
- 13:01taking this deep dive with us today. ODTRO Music.