Tech Brew Ride Home - Fri. 03/28 – CoreWeave’s Weak IPO

Episode Date: March 28, 2025

Move big moves in the gaming space. But we need to talk about that CoreWeave IPO. It. Didn’t. Go. Well. Facebook has a shocking idea: what if you could see what your friends were up to? Anthropic says it’s getting close to understanding how LLM’s actually work. And, of course, the Weekend Longreads Suggestions. Sponsors: LinkedIn.com/ride Links: ‘Assassin’s Creed’ Maker Ubisoft Jumps After Games Carve Out (Bloomberg) CoreWeave prices IPO at $40 a share, below expected range (CNBC) CoreWeave’s Shaky IPO Signal (The Information) Facebook Returns to Its Roots: Showing Posts From Friends and Family (NYTimes) Thoughts on setting policy for new AI capabilities (Joanne Jang) Anthropic makes a breakthrough in opening AI’s ‘black box’ (Fortune) Weekend Longreads Suggestions: Yahoo Is Still Here—and It Has Big Plans for AI (Wired) Inside arXiv—the Most Transformative Platform in All of Science (Wired) YouTube Video Of My Essay Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to the Tech meme right home for Friday, March 28th, 2025. I'm Brian McCullough today. More big moves in the gaming space. But we need to talk about that CoreWeave IPO. It didn't go well. Facebook has a shocking idea. What if you could see what your friends were up to? Anthropics says it's getting close to understanding how LLMs actually work. And of course, the weekend long read suggestions. Here's what you missed today in the world of tech. The games industry continues its multi-year reshuffling Ubisoft plans to spin off a year. unit, which will include the Assassin's Creed franchise into a around 4 billion euro subsidiary with Tencent investing around $1.16 billion for a 25% stake, according Bloomberg.

The deal represents a vote of confidence from Tencent, which already owns a 10% stake in Ubisoft in the wake of a difficult few years since a pandemic era lockdown easing. The unit's 4 billion euro valuation is higher than the group's current enterprise value. The valuation, quote, confirms that the group was severely undervalued. Charles Louis Plannad, an analyst at Midcap Partners, said in a phone interview, the cash injection of over $1 billion euro is very significant and solves the group's financing problems, he said. Ubisoft shares rose 9% to 14 euro and 9 cents at 9.30 a.m. in Paris trading on Friday

after earlier touching 14 Euro 27, the highest inner-day price since November. It's the outcome of a process that starts. started in January after last year's launch of Star Wars Outlaws disappointed, and the release of the latest Assassin's Creed installment was delayed. The new subsidiary will include the teams working on the titles and will control the back catalog and any new games under development. The deal is expected to close by the end of 2025. Assassin's Creed Shadows launched on March 20th to a broadly positive reception. While the company has not revealed specific sales figures for the new game, in which players inhabit one of two fictional characters in feudal Japan, it said in

a LinkedIn post that the launch surpassed previous installments. Early reviews averaged 82 out of 100, according to aggregator Metacritic, end quote. Okay, we need to talk about CoreWeave. They had their IPO yesterday raising $1.5 billion by selling around 37.5 million shares at 40 bucks a piece. But here's the thing. They had wanted to sell around 49 million shares in the $47 to $55 range. So this was significantly down from what they wanted to do, indicating weakness and demand from the market. And by the way, they only got the price they did, because Nvidia anchored the price a bit by putting in a $250 million buy order. So not great, Bob, not the AI IPO debut we were looking for, quoting CNBC. At the top end of the range, that would have valued CoreWeave at about $26.5 billion based on Class A and Class B shares outstanding.

but at the lower level, Corweaves valuation will be closer than $19 billion, though the market cap will be higher on a fully diluted basis. The IPO is a major test for tech startups and the venture capital market after an extended lull in new offerings dating back to the beginning of 2022, when soaring inflation and rising interest rates pushed investors out of risky assets. Other tech-related companies that have filed to go public in recent weeks include digital health startup hinge health, online lender Klarna, and ticketing marketplace stub hub. Bloomberg reported on Wednesday that chat app maker Discord is working on an IPO. The last venture-backed tech company that raised at least $1 billion for a U.S. IPO was FreshWorks in 2021. Last year, Reddit and Rubrik, each raised about $750 million in their offerings.

After Donald Trump's election victory in November, Goldman Sachs CEO David Solomon said he expected renewed IPO activity. But President Trump's imposition of tariffs in recent weeks added uncertainty to economic forecasts, and led to increased volatility to tech stocks. CoriWeave counts Microsoft as its biggest customer by far. Other clients include meta, IBM, and cohere. Revenue soared more than 700% last year to almost $2 billion, but the company recorded a net loss of $863 million. Coriweave's model is capital-intensive,

requiring hefty purchases of equipment and expenditures on real estate. A week after filing to go public, Coriave announced a contract with OpenAI worth up to $11.9 billion over five years. Open AI agreed to buy $350 million in CoreWeave stock as part of the deal, end quote. More on this from the information, quote, A poor showing on Friday would be both pretty shocking and not surprising at all. The surprising part is due to the whipsaw nature of AI hype. Just a couple months ago, an investment banker was telling me how enthusiasm from investors

was so strong that one of the biggest challenges for bankers was managing the hype. But that hype has begun to deflate on its own, and tech investors seem a bit overstuffed on AI stock stories. Oracle and Nvidia, two public companies investors might compare to Corweave are down 12 and 19% respectively since the start of the year. It's hard to ask investors to pay up for new AI firms when they're worried about their existing portfolio. I've been talking to Corweave's prospective IPO investors for months, and the reason for

the indigestion aren't surprising to anyone who has read our stories. They worry about how much of the business is tied to Microsoft or Nvidia. They worry about CorWeave founders having sold some. so much stock already. They worry about how much cash the company expects to burn. Investor sentiment has clearly swung toward the bears. I found more revealing details about investors' sour feelings in the results of a confidential survey, recently produced by an investment bank not involved in the deal. The bank anonymously surveyed 135 investors, including hedge funds and long-only stock pickers. A whopping 90% of the participants said that they didn't think Corweave had a sustainable moat,

essentially meaning it wasn't really a good long-term investment. One respondent summed up a broader perception about Corweave, quote, it's radioactive, and I think every investor knows that, end quote. The author of that piece I just quoted from Corey Wingingberg told our friend Alex Cantowitz at Big Technology, quote, I think people are just stopping and thinking about how companies are actually going to make money on AI and who's going to be left holding the bag, end quote. The data is adding a Friends tab on Facebook that will show content just from users' friends, with no recommended posts, as it brings back what it calls,

OG Facebook experiences, quoting the times. The feature called the Friends tab will replace a tab in the app that showed new friend requests or suggested friends. Friends tab will instead show a scrolling feed of posts such as videos, video stories, text, birthday notifications, and friend requests. For now, it will be available to Facebook users only in the United States and Canada. This idea of having a central place of what's going on with your friends, that was like the the magic of the early days of social media, said Mr. Allison, who is head of the Facebook app. We're making sure that there's still a place for this stuff on Facebook. It is something that shouldn't get lost in the modern social media mix. The new feed is a sharp departure

from the way social media has evolved over the past decade. The rise of apps like TikTok, habituated users to sing posts in their feeds from influencers and content creators who were often people they had never met. Other companies followed suit. Meta's apps, which also include Instagram, began leaning more heavily on recommended content to keep people engaged for longer periods. In recent years, more of people's feeds on Facebook and Instagram became dominated by creators, businesses, and brands. Recommended content like Reels, Meta's video product led people to spend more time on the apps, the company has said. Meta has no plans to stop adding recommended content to users' feeds, Mr. Allison said in the interview. For now, the company does not expect the Friends tab to be more popular than the home feed of recommended content.

And more changes to Facebook are likely coming. Meta plans to introduce other features and updates to Facebook in the coming year to make social media, Still, quote, feel social, Mr. Allison said. It is, frankly, core to Facebook, he said, end quote. Charlie Blix posted on blue sky, quote, How about a way to split friends and family because I'm still not willing to go back if I have to deal with boomer family members ranting about chemtrails and trans mice, end quote. Sam Altman says, quote, our GPUs are melting due to ChatGPT's viral image generation and OpenAI plans to

temporarily introduce some rate limits because of that. Certainly the socials haven't calmed down with all of this new AI art at all. And again, people are posting images of famous people, things that look obviously like very specific styles like Studio Ghibli. This is all stuff that was previously explicitly verboten. Joanne Zhang, who leads model behavior at OpenAI, says this is maybe because they are, quote, shifting from blanket refusals in sensitive areas to a more precise approach focused on preventing real-world harm. quote. When it comes to launching what feels like a new capability, our perspective has evolved across multiple launches. Trusting user creativity over our own assumptions. AI lab employees

should not be the arbiters of what people should and shouldn't be allowed to create. We're always humbled after launch discovering use cases we never imagined, or even ones that seems so obvious in hindsight, but didn't occur to us from our limited perspectives. Seeing risks clearly, but not losing sight of everyday value to users. It's easy to fixate on potential harms and broad restrictions always feel safest and easiest. We often catch ourselves questioning, do we really need better meme capabilities when the same memes could be used to offend or hurt people? But I think that framing itself is flawed. It implies that subtle everyday benefits must justify themselves against hypothetical worst-case scenarios, which undervalues

how these small moments of delight, humor, and connection genuinely improve people's lives. Valuing unknown, unimaginable possibilities. Maybe due to our cognitive bias against loss aversion, we rarely consider the negative impacts of inaction. Some people refer to it as invisible graveyards, although that's a bit too morbid and extreme. There are second order or indirect impacts unlocked by new capabilities, all the positive interactions, innovations, and ideas from people that never materialized simply because we feared the worst-case scenario. Ultimately, these considerations coupled with our progress toward more precise technical levers led us toward more permissive policies. We recognize this might be misinterpreted as Open AI lowering its safety standards, but personally, I don't think

that does justice to the team's extensive research, thoughtful debates, and genuine love and care for users and society. My colleague Jason Kwan once passed on to me, quote, ships are safest in the harbor. The safest model is the one that refuses everything, but that's not what ships or models are for, end quote. Anthropic says it has created a new tool for deciphering how LLMs think, and you used it to resolve some key questions about how Claude and probably other LLMs work. Quote, the researchers found that although LLMs like Claude are initially trained to just predict the next word in a sentence, in the process, Claude does learn to do some longer-range

planning, at least when it comes to certain kinds of tasks. For instance, when asked to write a poem, Claude finds words that make sense with the poem's topic or theme that it wants to rhyme, and then works backwards to construct sentences that will end with those rhyming words. They also found that Claude, which is trained to be multilingual, doesn't have completely separate components for reasoning in each language. Instead, concepts that are common across languages are embedded in the same set of neurons within the model, and the model seems to reason in this conceptual space, and only then convert the output to the appropriate language. The researchers also discovered that Claude is capable of lying about its chain of thought in order to please a user.

The researchers showed this by asking the model a tough math problem, but then giving the model an incorrect hint about how to solve it. The ability to trace the internal reasoning of LLMs opens new possibilities for auditing AI systems for security and safety concerns. It also may help researchers develop new training methods to improve the guardrails that AI systems have and to reduce hallucinations and other faulty outputs. Batson said that thanks to the kinds of techniques that he and other scientists are developing to probe these alien LLM brains, a field known as mechanistic interpability, rapid progress is being made. I think in another year or two, we're going to know more about how these models think than we do about how people think,

he said, because we can just do all the experiments we want, end quote. For the weekend, long read's suggestions this week, two from Wired. First up, Yahoo, just turned 30 years old. Yes, they're still around. And according to their CEO, they have big AI plans like everybody else. Quote, Lanzone has little patience for exhuming past blunders. I think the story of Yahoo's missed opportunities is tired. he says. It's boring. Instead of crying over lost search glory, Lanzone concentrated on improving what Yahoo did. We didn't have to worry about what we weren't, he says. He got rid of money-losing units like some non-performing ad tech divisions and quietly made some acquisitions to bolster the

best properties like Wager, a sports betting app to bring Yahoo Sports into the gambling age. He also brought in capable executives like former ESPN digital head Ryan Spoon, who now heads Yahoo Sports. He's boosted profits and grown the company's audience to the point where he says that Yahoo has performed the quickest return of any Apollo acquisition. Since Yahoo is private, the actual financials aren't available, but Yahoo's comms team provided me with a lengthy document packed with data to bolster Landzone's claim that Yahoo still has something to yodel about. Comscore, a marketing company that measures traffic, ranks Yahoo number one in news, number one in finance, and number three in sports. It's second only to Gmail in mail. He tells me

that in the U.S. alone, hundreds of millions of people use Yahoo every month. A year after Landzone took the job the entire tech world was turned around by the appearance of chat GPT. In previous transformations like search, social, and mobile, Yahoo has a near-perfect record of botching these moments. Landzone says Yahoo won't be creating its own language models or dropping $100 billion on data centers, but he believes the company will seize the moment nonetheless. I'd like to automate the word AI, so I don't have to say it so much, he says. Yahoo has in-house machine learning talent and draws on outside companies for AI technology. For instance, it partners with the startup Sierra for robot customer service agents. One of Landzone's canniest AI moves was acquiring

artifact, the AI-powered news aggregator created by Instagram co-founders Kevin Sistram and Mike Krieger. When the pair decided it would not become a viable business, they announced its closure and Landzone was among multiple suitors vying for the underlying technology. It became the centerpiece of the homepage that Yahoo! Relaunched earlier this year. Instead of incorporating their technology into our product, we did it the other way, Landzone says. Essentially, Yahoo News is now artifact. System-approached. proves. We partnered with Yahoo because they made a strong offer, but also because they planned on deploying our hard work to many millions of people, he says. Next up, for an AI-driven remake is Yahoo!

Finance, the leader among consumer investment tools, and arguably the company's crown jewel. Lanzone says he's already gotten a boost from product refinement. Yahoo is no longer trying to compete with CNBC on finance news, he says, and is focusing more on data. But a bigger reinvention is in the works. You're going to make more money. You're going to save more money. And we will use AI to do that for you, he says, end quote. And then the other one, again from Wired, the history and inside story of archive, the online science repository, quote, nearly 35 years ago, Genspark created Archive, a digital repository where researchers could share their latest findings before those findings had been systematically reviewed or verified. Visit Archive.org

today, it's pronounced like Archive, even though it has an X in the middle of it, and you'll still see its old-school web 1.0 design featuring a red banner and the seal of Cornell University, the platform's institutional home. But Archives' unassuming facade belies the tectonic reconfiguration it's set off in the scientific community. If Archive were to stop functioning scientists from every corner of the planet would suffer an immediate and profound disruption. Everybody in math and physics uses it, Scott Aronson, a computer scientist at the University of Texas at Austin, told me, I scan it every night. Every industry has certain problems universally acknowledged as broken. Insurance in healthcare, licensing and music, standardized testing

and education, tipping in the restaurant business. In academia, it's publishing. Academic publishing is dominated by for-profit giants like Ellsvier and Springer. Calling their practice a form of thuggery isn't so much an insult as an economic observation. Imagine if a book publisher demanded that authors write books for free and, instead of employing in-house editors, relied on other authors to edit those books also for free. And not only that, the final products was then sold at prohibitively expensive prices to ordinary readers and institutions were forced to pay exorbitant fees for access. The free editing academic publishers facilitate is called peer review, the process by which fellow researchers vet new findings. This can take months, even a year, but with archives, scientists could

post their papers, known at this on vetted stage as preprints, for instant and free access to everyone. One of Archives' great achievements was showing that you could divorce the actual transmission of your results from the process of refereeing, said Paul Fendley, an early archive moderator and now a physicist at All Souls College, Oxford. During crises, like the COVID pandemic, time-sensitive breakthroughs were disseminated quickly, particularly by Bio-Archive and Med Archive, platforms inspired by Archive itself, potentially saving by one study's estimate millions of lives, end quote. weekend bonus episodes for you this week. First up, I told you all that talk of sovereign

tech stacks in Europe maybe pulling back from buying U.S. cloud products led to a bit of a rant that I almost did at the end of an episode earlier this week, but the rant kept growing and growing until I was like, I just need to pull this out and do it as a separate essay. So that's what I did. You can watch the YouTube video of the essay if you want, because that's already live. Link at the very bottom of the show notes today, but I'll have the audio in this feed. tomorrow, Saturday. Then, in about an hour from now, I'm going on Alex Kentowitz's big technology podcast. So, as per usual, I'll share our conversation as we break down the week's news on this feed. I'll post that on Sunday. Talk to you on Monday.