Tech Brew Ride Home - Fri. 02/14 – Arm Upends The Entire Semiconductor Industry

Episode Date: February 14, 2025

TikTok is back in the app stores. But the biggest news is that Arm is going to make its own chips, thereby upending how the entire semiconductor industry has been constituted. Why Reddit has been killing it lately. And in the Weekend Longreads Suggestions, what if TikTok, but for Wikipedia? Sponsors: Tonal.com promocode RIDE for $200 off Links: Apple, Google Restore TikTok App After Assurances From Trump (Bloomberg) Arm to launch its own chip in move that could upend semiconductor industry (Financial Times) Shein IPO plans hit by Trump’s low-cost parcels crackdown (Financial Times) AI Licensing Deals With Google and OpenAI Make Up 10% of Reddit's Revenue (AdWeek) Weekend Longreads Suggestions: Forget DeepSeek. Large language models are getting cheaper still (The Economist) How a resurgent Walmart saw off the Amazon threat (Financial Times) Developer creates endless Wikipedia feed to fight algorithm addiction (Ars Technica) 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 Texas. me right home for Valentine's Day 2025. I'm Brian McCullough today. TikTok is back in the app stores, but the biggest news is that Arm is going to make its own chips, thereby upending how the entire

semiconductor industry has been constituted. Why Reddit has been killing it lately? And in the weekend, long read suggestions, what if TikTok, but for Wikipedia? Here's what you miss today in the world of tech. Well, it's back. Apple and Google have restored TikTok to their U.S. app stores after Trump administration assurances that a ban won't immediately be enforced. Quoting Bloomberg. The two companies had removed TikTok in the U.S. last month to comply with a law passed in 2024. In a January 20th executive order, Trump said he instructed the Attorney General, quote, not to take any action to enforce the act for a period of 75 days from today to allow my administration an opportunity to determine the appropriate course forward. By Thursday evening,

the software had returned to the Apple App Store and Google Play Store. store. Bloomberg News was first to report on the move. Trump previously supported a ban, but has changed his position. I guess I have a warm spot for TikTok that I didn't have originally, he said, when signing the executive order. If he doesn't negotiate a deal by early April to address the national security concerns around TikTok's current ownership, the app could be shut down once again. Bite Dance has maintained that TikTok is not for sale, end quote. Quoting Alex Heath on X, have confirmed Apple is in the process of putting TikTok back in the U.S. app store following a letter from Attorney General Pam Bondi assuring it won't face fines for doing so.

TikTok is now available in the Google Play U.S. store for Android users too, end quote. The Financial Times is reporting that Arm is making a wild new change to its business model. It's not just going to license chip designs for others to produce. It's going to launch its own chip, expected to be a CPU for servers and large data centers. This could happen as soon as this year after securing meta as one of its first customers. And on top of that, there's a Johnny Ive angle, quoting the FT. The move from designing the basic building blocks of a chip to making its own complete processor could also upend the balance of power in the $700 billion semiconductor industry,

putting Arm into competition with some of its biggest customers. Softbank founder Masayoshi-San has put Arm at the center of his plans to build a vast infrastructure network for artificial intelligence. The launch of Arm's own chip is considered just one step in his larger plans to move into AI chip production. say people familiar with the plans. Last month, Son unveiled his Stargate initiative in which he and OpenAI plan to spend a purported $500 billion building AI infrastructure with Abu Dhabi State Fund, MGX, and Oracle,

also providing funding for the U.S.-based project. RM is a key technology partner for Stargate, along with Microsoft and Nvidia, end quote. Again, the thinking here is that this new CPU will target data center applications with manufacturing to be handled by contract producers such as TSMC. Simultaneously, Arm's parent company SoftBank is nearing a deal to acquire server-chip designer Amper Computing, valued at approximately $6.5 billion and a move sources say is crucial to arms chip development initiative. The push into server chips

marks arms the latest success in winning business from major technology companies as Meta joins other firms moving away from traditional Intel and AMD processors. Meta's chief financial officer Susan Lee indicated last month that the company would expand its custom silicon development to include artificial intelligence training, seeking to optimize performance for specific computing requirements. In other words, this is where all that KAPEX spending is probably going. Sources also indicated that an arm design chip could eventually be incorporated into a new AI-focused personal device being developed through a collaboration between Love From, the design firm of former Apple executive Johnny Ive, Open AI, CEO Sam Altman, and SoftBank.

Under CEO René Haas, who took the helm in 2022, Arm has sought to increase revenue from its intellectual property by expanding its chip design offerings. But again, this represents a significant strategic shift that could strain relationships with key customers, including Qualcomm, which is currently engaged in licensing litigation with Arm and Invidia, the semiconductor industry's most valuable company. But also, Arm has long been seen as the neutral Switzerland of the tech industry, producing chips for everything from iPhones to AI. If they were to strike out on their own, could that possibly make existing customers think harder about alternatives? sources say Sheehan's UK IPO may be delayed to the second half of 2025 after President Trump's

recent tariff crackdown. Sheean had told investors a London IPO could happen by Easter, quoting the Financial Times. But an initial public offering is now likely to be pushed into the second half of this year following Trump's move to close the so-called de minimis rule, according to three people familiar with the process. The company, which was valued at $66 billion during its most recent funding round in 2023, has never published. publicly confirmed a timeline or plans for an IPO, which would lend a much-needed Philip to London's lackluster capital markets. The group founded in China and based in Singapore filed confidential papers in June last year with UK regulators for a proposed IPO and is still

waiting for regulatory nods in the UK and China. Plans by Sheehan, whose major markets include the U.S. and the U.K. to publicly list a portion of its shares have been dogged by geopolitics over the past 18 months. The U.S. crackdown affects Chinese e-commerce businesses such as Xi'an and Temu, the U.S. President announced earlier this month that the de minimis rule or exemption of tariffs on goods under $800 in value would be scrapped, and an additional 10% of tariffs on all Chinese goods would apply. Trump has temporarily paused measures to close the loophole, quote, until adequate systems are in place to fully and expediently process and collect tariff revenue after packages piled up at the border. The uncertainty over its impact and

timing is weighing on Sheehan's IPO timetable. The people familiar with its plans said the crackdown has pushed Sheean's focus on to its supply chain, although the group has not stopped work on its IPO and is still pushing for UK approval, according to one of the people familiar with its plans. Sheehan had initially targeted New York as an IPO venue, but shifted to London after being rebuffed by U.S. regulators. In October, its reclusive billionaire co-founder Skyjou met investors in the U.K. and the U.S. in anticipation of a flotation, end quote. Another one from my all-time high stock screener file. You know who has been near all-time highs for months now?

Reddit. Why? Maybe because Reddit now says content licensing deals with Google and OpenAI, among others, read as we'll let you train your AI on our content, now account for around 10% of its revenue, though its primary focus is still on ad revenue, which grew 60% year. on year in Q4, quoting Ad Week. The social platform, which on Wednesday reported a 71% year-over-year

lift in fourth quarter revenue has been, quote, very thoughtful about the AI developers it chooses to work with. Wong said, to date, the company has inked two content licensing deals, one with Google for a reported $60 million and one with chat GPT parent open AI. Reddit has elected to work only with partners who can agree to, quote, specific terms that are really important to us. These terms include user privacy protections and conditions regarding how Reddit is represented, Wong said. While licensing agreements with AI firms offer a valuable business opportunity for Reddit, advertising remains the company's core revenue driver. Much of Reddit's $4.7.7 million Q4 revenues were generated by the ongoing expansion of its advertising business,

and its ad revenue as a whole grew 60% year-on-year underscoring the platform's growing appeal to brands. The company, which IPOed last March, has focused on expanding its ad, ads business. In January, it debuted AMA ads and pro trends to Wu mainstream brands. AMA ads lets brands host Q&A sessions with built-in RSVP and reminder features, while ProTrens enables businesses to monitor brand mentions and viral discussions on Reddit. Reddit has also invested significant resources into machine learning and AI to enhance ad targeting and user engagement. It acquired memorable AI in August, a startup that predicts user engagement with ad creative to bolster ad effectiveness on Reddit. The efforts are paying.

off. The company has doubled its click volume and its conversion volume year on year and recorded significant growth across channels, geographical regions, and verticals in 2024. A particular bright spot has been small to medium-sized advertisers which drove outsized growth during Q4, according to Wong, end quote. Time for the weekend long read suggestions. Remember that S-1 AI model that researchers claimed they trained for just $6 made Deep Seek look positively expensive? I kind of didn't understand how they did it, but then I read this economist article which summed it up so succinctly, even a dumb, dumb like me, could understand it. Quote, S-1 was fine-tuned on the pre-existing Quen 2.5 LLM, produced by Alibaba, China's other top-tier AI lab. Before S-1's training began,

in other words, the model could already write, ask questions, and produce code. Piggibagicking of this kind can lead to savings but can't cut costs down to single digits on its own. To do that, the American team had to break free of the dominant paradigm in AI research, wherein the amount of data and computing power available to train a language model is thought to improve its performance. They instead hypothesized that a smaller amount of data of high enough quality could do the job just as well. To test that proposition, they gathered a selection of 59,000 questions covering everything from standardized English tests to graduate level problems and probability, with the intention of narrowing them down to the most effective training set possible.

To work out how to do that, the questions on their own aren't enough. Answers are needed too. So the team asked another AI model, Google's Gemini, to tackle the questions using what is known as a reasoning approach, in which the model's thought process is shared alongside the answer. That gave them three datasets to use to train S1. 59,000 questions, the accompanying answers, and the chains of thought used to connect the two. They then threw almost all of that away.

As S-1 was based on Alibaba's Quen AI, anything that model could already solve was unnecessary. Anything poorly formatted was also tossed, as well as anything that Google's model had solved without needing to think too hard. If a given problem didn't add to the overall diversity of the training set, it was out too. The end result was a streamlined thousand questions, that the researchers proved could train a model just as high-performing as one trained on all 59,000, and for a fraction of the cost. Such tricks abound, like all reasoning models, S-1 thinks, before answering, working through the problem before announcing it has finished and presenting a final answer. But lots of reasoning models give better answers if they're allowed to think for longer, an approach called test-time compute. And so the researchers hit upon the simplest possible approach to get the model to carry on reasoning.

When it announces that it is finished thinking, just delete that message and add in the word wait instead. The tricks also work. Thinking four times as long allows the model to score over 20 percentage points higher on math tests as well as scientific ones. Being forced to think for 16 times as long takes the model from being unable to earn a single mark on a hard math exam to getting a score of 60%. Thinking harder is more expensive, of course, and the inference costs increase with each extra weight. but with training available so cheaply, the added expense may be worth it, end quote. Then yet another one from the stock screener, Walmart, recently hit all-time highs. Why? Well, basically, they finally figured out how to compete with and basically clone Amazon's success.

Essentially, what if you already had this massive commerce company, and then you finally tacked on a successful e-commerce operation? boom, quoting the Financial Times. Less than a decade ago, investors feared for the group's future as e-commerce sales grew rapidly. In 2015, Amazon overtook Walmart's market capitalization. Its slick delivery services making huge stores on the edges of town seem anachronistic. Many expected Amazon's 2017 acquisition of Whole Foods to presage an assault on the U.S. grocery market in the year to January 2019. Walmart reported its lowest net income since fiscal 2002. Today, the company founded by Sam Walton's 63 years ago is resurgent. Analysts expect it to report a record $681 billion of revenue when it releases full-year results on February 20th,

maintaining its status as the world's largest company by sales. E-commerce sales have been expanding by more than 20% a year. Group-wide, 18% of revenue is now generated online, and its marketplace lists more than 700 million items from third-party merchants. Walmart shares have outperformed the market, and some analysts predict it will become the world's first $1 trillion retailer, though Amazon is still worth more. Walmart is rapidly transforming itself into a tech company akin to Amazon, said Nikki Baird, vice president of strategy and product at Aptos, a retail software company. I view those two in a category by themselves, and then everybody else is retail. The recovery is part of a wider pattern in U.S. retail.

The big are getting bigger. It's just as simple as that, said Jody Love, a portfolio manager. and investment firm Tiro Price, half of the recent growth in U.S. retail sales has been absorbed by just three companies, Walmart, Amazon, and the warehouse club chain Costco, according to Morgan Stanley, end quote. And finally, a weekend thing for you to check out, quoting Ars Technica. On Wednesday, a New York-based app developer named Isaac Gemmell debuted a new site called Wikitok, where users can vertically swipe through an endless stream of Wikipedia article stubs in a manner similar to the interface for video sharing app TikTok. It's a neat way to stumble

upon interesting information randomly, learn new things, and spend spare moments of boredom without reaching for an algorithmically addictive social media app. Although to be fair, Wikitok is addictive in its own way, but without an invasive algorithm tracking you and pushing you toward the lowest common denominator content. It's also thrilling because you never know what's going to pop up next. WikiTalk, which works through mobile and desktop browsers, feeds visitors a random list of Wikipedia articles called from the Wikipedia API into a vertically scrolling interface. Despite the name that harkens to TikTok, there are currently no videos involved. Each entry is accompanied by an image pulled from the corresponding article.

If you see something you like, you can tap read more, and the full Wikipedia page on the topic will open in your browser. The original idea for Wikitokok originated from developer Tyler Angert on Monday evening when he tweeted, insane project idea, all of Wikipedia on a single scrollable page. Bloomberg Beta V.C. James Chamm replied, even better an infinitely scrolling Wikipedia page based on whatever you are interested in next, and Engert coined Wikitok in a follow-up post. Early the next morning at 1228 a.m., writer Grant Slatton, quote, tweeted the Wikitok talk discussion, and that's where Gamal came in. I saw it from Slatton's quote retweet, he told ours. I immediately thought, wow, I can build an MVP, minimum viable product, and this could take off. Gamel started his project at 1230 a.m.

and with help from AI coding tools like Anthropics, Claude and cursor, he finished a prototype by 2 a.m. and posted the results on X. Someone later announced Wikidot on Y Combinators Hacker News, where it topped the site's list of daily news items. The entire thing is only several hundred lines of code, and Claude wrote the vast majority of it, Gemmell told ours, AI helped me ship really, really fast and just capitalize on the initial vertical tweet asking for Wikipedia with scrolling, end quote. Actually, about eight years ago before I started this podcast, I was in very, very limited talks with the Wikipedia Foundation to do a daily podcast like this, one Wikipedia article a day as a podcast, maybe 15 minutes long. I don't know if that

gave me the inspiration for this show. Obviously, the talks did not go anywhere, but I still kind of feel like I wish I could do that podcast too. Okay, I'm going to try the test again today to do a live stream call-in thing. I failed yesterday, but now I know why. If you want to help me test it again, check my socials around 1130 or so Eastern time. I'm going to just leave a live stream up for about a half an hour or more. See the link on my Twitter or Blue Sky account and join while I put together the Omnibus episode for the weekend. You have to be on desktop to do it. Mobile phones won't work.

And what I learned is I can't solicit people coming on stage. You have to actively ask me to come on stage and ask what they call in question. That's what confused me yesterday. I couldn't see who was in the room so I couldn't get people to come on stage. But what you do is, once you're in the live stream, go into the chat and ask a question in chat. and then I believe you should see an option to request a live call-in. They have you select your camera and mic and all that stuff, and at that point I should be prompted to bring you on stage.

Should be, I should stress. That's what we need to test out. Anyway, as I said, check my socials for about an hour, maybe between the 1130 and 1230 Eastern Time window. I'll just leave it running and hope folks swing by to work out the kinks with me. Thanks to those who tried to help yesterday, and thanks to anyone who helps today. talk to you soon or all of you on Monday. Yes, I am currently planning to do a show on Monday despite the holiday.