Speaker 1With Anthropic's IPO coming up next week, we are getting numbers. We're finally seeing what valuation we want to go out at. We're getting investors' response to that valuation. We're getting debates on what it says about the AI trade. And all that is happening as Anthropic continues to release new products that rip market cap off of competitors. Today, we're talking about everything we know heading into the IPO and why basically I think that we know nothing at all. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Robots and Pencils, CDATA, and Blitzy. To get an ad-free version of the show, go to patreon.com slash ai-dailybrief, or you can subscribe on Apple Podcasts. And if you want to learn more about sponsoring the show, send us a note at sponsors at ai-dailybrief.ai. And while you're there, catch a link to check out the new How We Got Here podcast. It is a long-form show, so if you like the more big think episodes of the AI Daily Brief, you can subscribe to our podcast. And if you want to learn more about sponsoring the show, you should definitely check it out. Again, it's called How We Got Here with NLW. You can find it linked in the show notes or go to howwegothere.com. Welcome back to the AI Daily Brief Headlines Edition, all the daily AI news you need in around five minutes, and we have a lot to get through today. Google has released, officially, Gemini 4 Carbon, the second iteration of their latest ultra-large model. Carbon, you might remember, was previewed at the end of October, and at the time, it was the top-ranked model on the Artificial Analysis Index. Unfortunately, Google said the government's voluntary safety testing process and wasn't yet ready for public release. And although people were hopeful, nobody was really willing to get all that excited until they got access to it themselves. Now, two weeks later, even if Fable 5.2 has definitively claimed the top spot on the rankings, and is this week's shiny new model, people digging into Carbon are finding lots to like. We have the usual render tests where Carbon stands up remarkably well. In fact, absent some of the mind-blowing capabilities of Fable 5.2, this would likely be the most impressive version of AI generated animations. Ethan Mollick showed off what you can do by combining this with Google's strong multimodal capability, using Carbon to recreate a real-world video clip of a train ride in 3JS. He wrote, It is a very good model that improves on Argon. It's useful to have an AI that thinks differently than Claude and GPT, and is still highly capable but has its own strengths and weaknesses. Still, the coding capabilities might be the most interesting upgrade with Carbon. Google has, of course, struggled mightily with AI coding over the years, but Carbon seems to mark a new moment in their catch-up efforts. Pushmit Kohli, the VP of DeepMind, showcased a formal verification of Signal's security properties, which required Carbon to analyze the machine code underpinning the software. He wrote, More than a research breakthrough, it's a prime example of how frontier AI can be a force multiplier to counter the speed and scale of modern cyber threats. Formal verification offers a higher standard of security. However, it's extremely complex and time-consuming, which is why, in many cases, it's considered to be impractical. Similar efforts could take teams decades of combined effort. We did it in a month's time. Well, Huron took Carbon out for a spin on some more mundane coding problems, running it through his bug hunt test, and he found that Carbon found 57 of the 105 bugs, second only to Fable 5.2 at 58. Perhaps even more impressive was that Carbon completed the run for $31, compared to almost 140 for Fable 5. Huron commented, Google's own benchmarks looked super strong, but nobody could access the model. Turns out, they didn't lie. Nobody should be laughing at Google anymore. Importantly, these cost savings were pretty consistent across a range of companies, one Render test produced similar results to Fable 5.2 at a third of the cost. It is still early, but overall, it's looking like this is an extremely competent model that's competitive with rivals in the ultra-large model class category. Carbon gives those who are locked into the Google ecosystem a very meaningful upgrade and a new level of coding ability, and potentially signals a turn. Ethan Malek again sums up, One of the most significant things about Gemini 4 Carbon is that it is showing that Google is back in the race of frontier models, rapid releases, and rapid code. Next up, we move on to the scandalous. NVIDIA's Chinese spy scandal continues to reverberate across the headlines as Washington, Wall Street, and Silicon Valley ask what can be done about Jensen Huang. On Monday, President Trump held a press conference to discuss it. He said, I know the FBI is looking into it, in terms of something else, and I think he'll be in good shape. I just know Jensen very well. I just think he's a very honorable man. I mean, he built a great American company, and I think he's a very honorable guy. Trump acknowledged that he had knowledge of the affair before he was born. He said, But was vague about the precise timing. One of the key details from the New York Times reporting was that President Trump was fully briefed on the matter last April, but elected to do nothing. Tuesday's news also brought the revelation that as much as $7.5 billion worth of NVIDIA chips may have been funneled to Chinese companies through Megaspeed. That's the neocloud that employed Alice Huang, who is the alleged Chinese spy. The number comes from sources speaking to the information, who claim to know Alice Huang personally. The information's reporting claims a and they seem to have a full recap of how the chips flowed through it. The New York Times, meanwhile, continues to land bombshells, reporting that a Commerce Department investigation into Megaspeed was halted after contact from NVIDIA. The reporting claims that NVIDIA General Counsel Tim Teeter discussed the case directly with investigators and said their superior at the Commerce Department had approved the contact. Sources said the Commerce Department agents were alarmed that their superior was mentioned and were concerned he was, quote, violating government procedure and potentially tainting the case. Sources told the information that the Commerce Department is working on new export controls, including shutting down the remote access loophole for NVIDIA chips. But at the same time, they're delaying their decision to place Megaspeed on the entity list and block all exports to the company that is supposedly run by Chinese spies. Meanwhile, House Republicans are insistent that Jensen comes to Washington to explain himself. If Jensen doesn't accept their invitation for a closed-door meeting next week, House Foreign Affairs Committee Chair Brian Mass said he's prepared to issue a subpoena. And overlaid on top of all of it is the question of what NVIDIA should do about Jensen. Some market commentators have noted that Jensen is just about the only CEO that could become engulfed in an international espionage scandal and keep his job. NVIDIA has a famously flat structure and Jensen juggling 60 direct reports. What's more, the Wall Street Journal reports that NVIDIA has no succession plan, and the board of directors was treated as more of a governance formality than a check on the CEO. Rene Haas, the CEO of Arm, who previously worked for NVIDIA, said in a 2024 interview that Jensen had resisted succession planning for years, commenting, He didn't like the idea. He felt it would create competition. among executives and be divisive. The board is currently investigating Jensen with the help of an outside law firm, but even if they find damning evidence, it's unclear what the correct path will be. Given that the board has a fiduciary responsibility to shareholders, getting rid of a charismatic CEO with no succession plan seems risky to say the least. Even kicking him upstairs to the chairman's role has its risks. The Wall Street Journal is currently suggesting that a reined-in Jensen is NVIDIA's best option. They wrote that NVIDIA's task is to, quote, find a way to keep Huang on while reassuring shareholders, the government, and the general public that it takes national security risks seriously. That seems like a tall order, but the stakes are also extremely high for the nation at large. NVIDIA stock is down 11% since the story broke in an otherwise positive market. The company makes up 7% weighting in the S&P 500 and around 8% in the NASDAQ, meaning most U.S. investors have a ton of NVIDIA exposure. What's more, the company's success has become a matter of national security via the AI race. In short, there is just not a simple solution to the scandal, and it seems that the entire nation, NVIDIA, is experiencing the mother of all key man risks. One small but interesting NVIDIA side story, consulting giant Booz Allen has switched to open-source, adopting NVIDIA's Nemotron. Finally today, a set of OpenAI news. The company's revenue is soaring after a strong reception with GPT-6 Astra. Sources speaking with Bloomberg said that Q3 revenue came in above $10 billion, a record quarter for OpenAI and a 50% increase from quarter two. Operating losses also grew, reportedly coming in between $10.1 and $10.9 billion, and while operating margin did come down from -150 to -109, OpenAI still has a long way to go before turning a profit. Still, this is definitely a good news story for OpenAI, who has been struggling to shift the narrative with regard to their own growth vis-a-vis Anthropic. Q2's quarter-over-quarter growth rate was only 17%, which would be exciting for most companies but a little lacking compared to Anthropic's 144% for Q2. Sources highlighted solid API demand for recent model releases, including GPT-5-6 Sol and GPT-6 Astra as the driving force of the growth, and API demand for OpenAI's image models, which have seemed to be a quiet success for the company. In addition, OpenAI brought in ad revenue at a billion-dollar annualized pace, adding significantly to the bottom line, and overall the numbers have analysts reconsidering where OpenAI sits in the AI race, positively for the first time in a while. OpenAI has also acquired an AI travel agent. Francis Davidson, the founder of Odessia, announced that OpenAI had acquired a startup on Tuesday. Odessia was founded less than a year ago, and it seems it's still pretty small. Last summer, they booked $60 million in travel across the country, and the company is expected to start tens of thousands of users, which, while huge for an early-stage startup, is a drop in the ocean for the multi-trillion-dollar travel industry. Davidson was previously the founder of Sonder, an Airbnb competitor which grew to a thousand-person headcount and won public fee of SPAC in 2022. It filed for bankruptcy in 2025 after a key partnership with Marriott fell apart. This time around, Davidson wrote that he threw out the old playbook and tried to build a lean startup with AI-first process. At some point, I might come back to his blog post which gives a ton of the tips around how he works, including dictating around 1.2 million words a month, which is around $11. But for OpenAI's part, they're staying pretty quiet about their future plans for travel. Odessia shut down their service as part of the acquisition, and it's a little unclear whether Davidson will be working on existing travel features in ChatGPT or launching something new. Lastly, OpenAI has introduced a new business hub that offers more powerful tools to companies advertising on the platform. The business hub allows businesses to register their details with OpenAI and includes a dashboard that tracks advertising metrics similar to Google Analytics. The other big addition is the ability to create what OpenAI is calling business agents. If a brand has the feature enabled, consumers will be able to speak with a business agent after clicking on an ad to ask any questions or explore the brand's range. Businesses can provide their agents with instructions and additional context to ensure they represent the brand appropriately. And if you don't see this as how Metamuse and ChatGPT start competing more directly, I don't know what to tell you. What I think you can assume is that we are about to see a lot more integration and consideration of ChatGPT as a commerce platform, given how many of the answers it provides consumers relate in some way to shopping or commerce. Lots more to explore there, but for now, that is going to do it for today's headlines. Next up, the main episode. If you're leading AI inside an enterprise, you already know that the gap right now isn't capability but execution. That's why KPMG's You Can With AI is back with a new season featuring conversations with leaders like Sorojit Chatterjee of Emma, May Habib of Writer, McKesson CIO Ellery Fisher, and others focused on practical execution. What's working, what's not, and what it actually takes to move from pilots to real scaled impact across strategy, data readiness, governance, workforce, and value. And of course, it's co-hosted by me, Nathaniel Whittemore. Go listen and subscribe at www.kpmg.us slash AI podcasts. That's www.kpmg.us slash AI podcasts. The best teams don't have a single star carrying everyone else. They know their own strengths and each other's weaknesses and play to both. That's the team Robots and Pencils has built on purpose. Nobody there is grinding through busy work to pad a headcount number. People come for the hard problems and they stay because everyone around them is leveling up at the same time. In a market full of companies that are just trying to hire fast, that's worth a look. Check out robotsandpencils.com slash careers. This episode is brought to you by CDATA. Token spend is climbing, security is holding agents back, and companies rolling out AI initiatives might not know how it's truly being used. The root cause, AI, is cut off from the systems and context your business runs on. That's why CDATA built Connect AI Gateway, one control point between your AI and your systems. Connect AI Gateway is not just another static model router. Prompts get intelligently routed to the right model with exactly the tools and data it needs governed down to the record. The result? Accurate answers at a fraction of the token cost with a context engine that learns your business with every interaction. Built on the connectivity layer powering Palantir, Google, Salesforce, and hundreds more. Give it a try and start your free trial at cdata.ai. That's C as in connect, data, dot AI. Here's why most legacy modernization projects fail. The AI doing the work can't understand code bases at scale. It sees a small slice of context, examines syntax, and misses years of decisions distributed across the global application ecosystem. Blitzy solves this the way it solves everything. Grounded in your code before any migration begins, Blitzy's agents reverse engineer the entire legacy system into a persistent knowledge graph. Every dependency, every constraint, every piece of tribal knowledge that used to live in one engineer's head. From that understanding, Blitzy autonomously executes language migrations, framework upgrades, and monolith-to-microservices transformations, all validated end-to-end. One Blitzy customer modernized a $10 million monolithic insurance stack in 16 weeks against a 137-week baseline with coding agents. That's 9x compression. Retire technical debt while accelerating your roadmap. See how at blitzy.com. That's B-L-I-T-Z-Y dot com. Welcome back to the AI Daily Brief. At this point, the speculation has come and gone, and the midterm blue wave that brought a bunch of theoretical data center opponents to power has not deterred Anthropic from going public as some imagined it might. In fact, it appears that the IPO will go off before Thanksgiving, with the company aiming to go public next week. Anthropic kicked off their roadshow and are making their final pitch to the market. Here's an excerpt from the roadshow video today, with Dario making his most consolidated pitch.
Speaker 2Now, interestingly, for a company that has often struggled to find the right
Speaker 1balance of these messages, because careful listeners will know that none of that that Dario just said is a new statement for the company, but the balance that the point, in fact, is the age of miracles, and that the risks are simply what we have to overcome to achieve those miracles, is something that has, I think, in the past, eluded it. And of course, that's not the only communications coming from Anthropic and Dario. Given how much contention there has been between Anthropic and the Department of Defense slash War, many were surprised about the crispness of his answer to MLS. When she asked, in no uncertain terms, if he was comfortable with Claude helping the military kill more people more quickly. Dario reinterpreted the question and said, basically, what you're asking is, do you want this country to be a stronger country? To which he said, of course I do. I'm a patriot. But more than that, he said, the decision about what to do with the technology is not ours to make. You have to trust the military to make those decisions. And whatever I might think as a private citizen, that has no bearing. On the decisions the military makes, and it's not the job of the technology provider to make those decisions, except to provide some very high-level boundaries. Once again, there's nothing there that's necessarily a radical departure from things Dario has said before. Although I don't believe that he's ever been quite so crisp about saying that it is not the technology provider's decision about what the military uses that technology for. But the way that this is being delivered is finally clicking. Still, for the markets, none of that is as important as the question of price. Anthropic is asking between $70 and $78 per share, which puts the implied valuation somewhere between $1.25 trillion and $1.4 trillion. Anthropic said they seek to raise between $57 billion and $64 billion. That said, Anthropic is also raising $26 billion from industry partners including Nvidia, Microsoft, Google, and Broadcom, with Nvidia and Google being the largest participants at up to $10 billion a piece. Combined with the so-called green-shoe option that allows underwriters to buy extra shares to service excess demand over the first month, Anthropic will float a total of around $100 billion in stock. Now, on the one hand, the valuation is lower than previous discussions. In September, investors floated a $2 trillion valuation in the press. And while some took that as a reasonable evaluation for a few weeks, in more recent times, we've seen the valuation estimates tick lower. $1.6 trillion was mentioned, then $1.5, and ultimately last week's reporting had one group of investors saying the IPO might need to price at $1.1 trillion to get sufficient demand. Now, one thing that has driven me absolutely batty, about the coverage of these numbers, is that they've been treated as specific targets coming from Anthropic. In other words, like Anthropic has had to rein in its ambitions because it went out for $2 trillion and was unsuccessful and had to settle for $1.4 trillion. Let's be clear about the fact that that is not what happened. Some journalist got a source from some investor around Anthropic. By the way, we have no idea which investor, how close to Anthropic they are, whether they're actually even invested in Anthropic, or whether they're just an investor on deep background. And the $2 trillion number was so splashy that it stuck. Anthropic, of course, can't comment being in their quiet period, and so the narrative takes hold. Now, there is still valuable signal in understanding how investor expectations have changed over the time. Darcy AGI on X wrote, October really changed the mood around AI valuations, and to the extent that that's what we're drawing from the analysis, I completely agree. But let's not pretend as though this somehow represents some failure of a goal that Anthropic had set. What's more, again, as Mr. Darcy points out, when he says, I think this might be a much healthier starting point, there is evidence that that's the case. Within a day of that lower valuation target, between $1.25 trillion and $1.4 trillion being made public, it already started to seem like it was being pitched too low. The Financial Times reports that Anthropic's investment banking partners plan to raise the share price to reflect a top-end valuation of $1.5 trillion. The article stated the revision was being considered after receiving, quote, strong indications of demand. It also added that Anthropic cares much more about raising cash than valuation, and that it's not a good idea to raise the share price. It's a good idea to reduce the valuation due to their large compute commitments. And basically, what I think we should take from all of the discourse around the IPO price is that investors do not have a clear idea of what to expect, what they think is reasonable, what they think others think is reasonable, what they think others expect, and we are kind of just flailing around until the thing actually happens. There are plenty of bears out there who are promoting the idea that the reduction in valuation is a bearish signal, and who continue to think that Anthropic underperforming on their IPO is a bearish signal. We'll trigger another leg down in the AI trade overall. John Ma from Artemis actually argued that this is increasingly the consensus view, that Anthropic is overvalued at $1.5 trillion, that revenue is flattening, the tokens are being commoditized due to open source, the token prices are falling faster than demand, etc, etc, etc. And yet, as he points out, there is a pretty compelling contrarian case. One part of that is that Anthropic reported that their run rate revenue was up a full $10 billion in the month of October, reaching $81 billion. And that's a pretty compelling contrarian case. One part of that is that by the end of the month. And not only does that number show strong continued growth, that remains primarily driven by the coding agent use case, and obviously reflects Opus 55, but not Fable 5.2. According to Ma, the contrarian case points out that finance, legal, life sciences are all super early, and even that agents are still early, and most businesses still think of AI as chat GPT in the cloud app. The contrarian take also points out that retail is frankly starving for direct exposure to Frontier Labs, and will ape in regardless of the valuation and drive Anthropic to $3 trillion in market cap. For a point of view that simply tries to use proxies and math, iCapital's chief investment strategist Sonali Basik points out that at a $1.4 trillion valuation, that would represent around 40x trailing 12-month revenue. She points out that by comparison, Palantir is trading at roughly 55x, meaning Anthropic would be closer to multiples not of Palantir but of CrowdStrike and CloudFlare. Now, of course, the other thing that some investors are going to be focused on even more than valuation and revenue is future compute commitments. In their updated IPO paperwork, Anthropic disclosed $10 billion in new commitments with Microsoft and Google. The new commitments are stretched over the next seven years. The same document also said that a $35 billion deal with NeoCloud Lambda could be cancelled due to non-performance. Lambda has defaulted on the contract after blowing a deadline, and Anthropic said in their filing that they're within their rights to cancel the deal but does not currently intend to. Now, this reshuffling of cloud deals gives us a little more insight into how Anthropic's compute spend is structured. When their IPO paperwork was first made public, many analysts were alarmed to see hundreds and hundreds of billions of dollars in compute commitments, with much of it being non-cancellable. What would happen if AI demand faltered? And yet a lot of observers, who have not been unsympathetic to these questions, have pointed out that it feels like more of the risk lies with the NeoClouds, who frankly have far more risk of dragging the rest of the industry down with them even should they implode. One of the interesting reversions that some are having as they look at this Lambda deal is that it basically brings up the question, which is more likely to happen first? AI demand reducing such that Anthropic can't meet its commitments, or the NeoClouds to whom that revenue is going to come from? Or is it just going to be a matter of whether or not they're going to hitting problems in either financing or construction that don't allow them to deliver finished data centers on time? Still, wherever you fall on that, this is going to be a continued narrative, and it is absolutely not going to go away after the company goes public. Finally though, as all of this debate was happening, we also did get one more indication of how immense investors view the power of Anthropic to disrupt other companies. Alongside all of their IPO updates, they launched a new productivity suite that many are seeing as a pretty direct shot at Microsoft and perhaps Google as well. The suite includes Claude Sheets, Claude Whiteboards, and Claude Data. Claude Sheets is Anthropic's direct competitor for Excel, with a much greater focus on collaborative work with humans and agents. Claude Whiteboards is a team planning service for organizing projects and ideas. Claude Data, meanwhile, is the evolution of the feature previously known as Dashboards, which allows users to pull in data from CRMs and database platforms to create live data feeds. It competes with Microsoft's Power BI, but adds the ability to use Anthropic's models to generate visual data stories or narrated short videos. These are, in other words, business productivity tools. And while many are trying to read it as a declaration of war against Microsoft, Anthropic framed it as merely the way the world is going. They wrote, "Inside Anthropic, non-technical teams are starting to get as much out of Claude as our engineers. Our legal team worked with Claude to build a Slack-based legal review tool that automatically draws red lines from the team's own playbook. This brought marketing reviews from three days down to 24 hours. And easy fixes are made before a lawyer reviews the draft. Our finance team turned a long business update document for executives into an interactive page. Now the whole team updates it together with Claude. Neither of these teams writes code, but they're experts in their fields. They use Claude to turn that knowledge into tools that save them time and that others can use." In other words, when Claude becomes the shared work surface, you kind of just need this sort of infrastructure around it. A perhaps more concerning way to put it for Microsoft is that the competition for where work happens started the moment ChatGPT launched in November of 2022. And increasingly, the place where work happens is inside the LLM Harnesser app, and this is just the natural extension of that. Matthew Berman wrote, "I keep trying to warn you, if you build software, the labs will compete with you." And it seems that investors agreed, driving Microsoft's stock down 2.2% on the news. Gabrielle pointed out, however, that even if these tools start to pick up market share, Microsoft is very likely to be fine. He writes, "Excel will be fine. No enterprise is ditching Excel for this. Claude Sheets even exports to Excel. Microsoft is on every side of this. They're buying into the IPO next week. Anthropic owes them $45 billion in cloud spend, and they resell the model through Copilot. They're an investor, landlord, reseller, and now the competition. They want Anthropic to do well. It's funny Microsoft dipped on this." But look, ultimately, I think that the dip is yet another indication of the market just trying to work out what this all means, and perhaps even recognizing that it just isn't sure yet. Still, whether the market knows or not, the IPO is happening, and it's happening soon. So, buckle up, buttercup, because the next phase is here. For now, that's going to do it for today's AI Daily Brief. Appreciate you listening or watching. I'll see you next time. Until then, peace.
PodcastTranscript AI