A (0:04)

Welcome to the Techbrew Ride Home for Tuesday, August 19, 2025. I'm Brian McCullough. Today Masa San is investing $2 billion into intel and Uncle Sam might join in as well. Small AI models continue to have a moment. Chamath is re entering the arena with a new SPAC. The Texas Attorney General is investigating AI chatbots and does GPT5 prove that we've hit an AI ceiling, even if only temporarily? It here's what you missed today in the world of tech. Over 10 years ago, Mizzen and Main invented and some might say perfected the performance fabric dress shirt. To this day, they continue to embrace that same entrepreneurial spirit by re engineering classic American styles with modern fabrics. The goal is to make it easier for guys to achieve and enjoy their version of success. So whether you're grinding away in an office in San Francisco or on site in Austin, you can rest easy knowing you don't have to sweat about sweating. They're the brand you can turn to if you're looking for a machine washable and wrinkle resistant white dress shirt you can wear every day of the week, high stretch pants that are just as comfortable in the office as they are on the golf course, or a lightweight and breathable wool blazer that makes dressing up feel like a breeze. Go to mizzen and main.com and use promo code brew15 to get 15% off your first purchase. That's mizzen and main.com promo code brew15masa san to the Rescue Softbank has agreed to invest $2 billion in intel, paying $23 per share for Intel's common stock, thereby buying around 2% of its outstanding shares. Quoting CNBC It's a vote of support for intel, which hasn't been able to take advantage of the artificial intelligence boom in advanced semiconductors and has spent he to stand up a manufacturing business that's yet to secure a significant customer. Masa and I have worked closely together for decades and I appreciate the confidence he has placed in intel with this investment, intel CEO Lipp Bu Tan said in a statement, referring to SoftBank founder Masayoshi Son. Intel shares lost 60% of their value last year, their worst performance in the company's more than half a century on the public market. The Stock is up 18% in 2025 as of Monday's close. Intel has been a major topic of discussion in Washington of due to the company's role as the only American company capable of manufacturing the most advanced chips. However, Intel's foundry business, which is designed to manufacture chips for other companies, has yet to secure a major customer, a critical step towards stabilization and expansion. Last month, intel said it would wait to secure orders before committing to certain future investment in its foundry. Tan met with President Donald Trump last week after the president called for the CEO's resignation. The US government is considering taking an equity stake in intel, according to reports. Softbank, meanwhile, has become an increasing large partner in the global chip and AI markets. In 2016, SoftBank acquired chip designer ARM in a deal worth about $32 billion at the time. Today, the company is worth almost $150 billion. ARM based chips are part of Nvidia's systems that go into data centers. In March of this year, Softbank announced plans to acquire another chip designer, ampere computing, for $6.5 billion. The Financial Times is also reporting that Masayoshi San met lip Bhutan to discuss Softbank acquiring Intel's foundry weeks before announcing this new investment. But I guess that didn't pan out back then to those rumors of the US Government maybe taking a stake in intel as well, Quoting Bloomberg the Trump administration is in discussions to take a stake of about 10% in Intel, a move that could see the US become the beleaguered chip maker's largest shareholder. The federal government is considering a potential investment in intel that would involve converting some or all of the company's grants from the CHIPS act into equity, said the people, who asked not to be identified because the information is confidential. Intel has been slated to receive a combined $10.9 billion in such grants for commercial and military production. The company can also draw on up to $11 billion in loans under the 2022 law. The grant money, which was originally designed to be dispersed over time as intel meets project milestones, is roughly enough to pay for the targeted holding at Intel's current market value. A 10% stake in the chipmaker would be worth around $10.5 billion. The exact size of the stake, as well as whether the White House chooses to move ahead with the plan, is still people said. End quote. Nvidia had a bunch of news late yesterday, including plans to roll out RTX 5080 class chips to GeForce now in September, thereby letting users of its $20 per month Ultimate Gaming Tier stream in up to 5K at 120 frames per second. Also, Nvidia Discord and Epic Games are testing game demos on Discord servers, letting users try a game without downloading it or signing up, starting with Fortnite. But I guess the news of the moment since Small AI models are having a moment was that Nvidia debuted the Nemotron Nano 9BV2, a hybrid mama transformer model, saying it achieved scores comparable to or better than Quen38b on reasoning benchmarks, quoting VentureBeat. While the 9 billion parameters are larger than some of the multi million parameter small models VentureBeat has covered recently, Nvidia notes it is a meaningful reduction from its original size of 12 billion parameters and is designed to fit on a single Nvidia A10 GPU. It's based on Nemotron H, a set of hybrid Mamba transformer models that form the foundation for the company's latest offerings. While most popular LLMs are Pure Transformer models which rely entirely on attention layers, they can become costly in memory and compute as sequence lengths grow. Instead, Nematron H models and others using the Mamba architecture developed by researchers at Carnegie Mellon University and Princeton also weave in selective state space models, or SSMs, which can handle very long sequences of information in and out by maintaining state. These layers scale linearly with sequence length and can process contexts much longer than standard self attention without the same memory and compute overhead. A hybrid Mamba transformer reduces those costs by substituting most of the attention with linear time state space layers, achieving up to 2 to 3x higher throughput on long contexts with comparable accuracy. Other AI labs beyond Nvidia, such as AI2, have also released models based on the Mamba architecture. End quote in yet another sign that the boom times might be back for Silicon Valley, Chamath has officially re entered the arena. American exceptionalism acquisition a SPAC, or Special Purpose Acquisition Company led by Chamath Palihapitiya and focused on energy, AI, DeFi and defense, has filed for a US IPO seeking to raise $250 million. Quoting PitchBook, this new vehicle's theme is in line with Silicon Valley's reignited interest in sectors such as defense and domestic infrastructure. Venture capital firms have been increasingly investing in these categories amid the current Trump administration's notable openness to working with Silicon Valley. Palihapitiya came out in support of then presidential candidate Donald Trump last year alongside a growing cadre of fellow startup investors. The SPACS founder shares the equity the vehicle's sponsor earns will notably only vest if the merger business achieves at least a 50% stock price increase, though it will be a hefty stake 30% instead of the common 20%. Sponsors were heavily criticized during the pandemic era SPAC mania for essentially getting sizable stakes in the combined business at nominal prices then generating high returns if the business performed well in the public market. Polihapitiya had previously sponsored 10 SPACs since his first in 2017. According to SPAC research. Four were eventually liquidated, while the rest merged with companies including Virgin Galactic, Open Door sofi and Clover Health. And quoting Cointelegraph, Polihapitiya led several high profile specs during 2020 and 2021, including successful mergers involving Social Capital Suvretta Holdings 1 and Social Capital Hedosofia Holdings 5, now operating as SoFi Technologies. However, other SPACs led by Palihapitiya such as Social Capital Suvretta Holdings 2, 3 and 4, were liquidated, giving him a mixed record. SPACs face challenges because they are bound by strict time limits to find a private company to merge with, and often struggle to identify companies worthy of high valuations and operate under considerable regulatory scrutiny. I said this is a signal that the boom times might be back, though lots of other folks were pointing out this could also be a sign that we've already reached the top of a market cycle. As Inverse Kramer said on X, Chamath just launched a new spac, if anyone needed a new short target. And as UKO Capital said on X, with Chamath launching a SPAC and Benioff buying every company not pinned down, the only question is whether it's January or November of 2021. End Quote Texas Attorney General Ken Paxton is investigating Meta and Character AI over claims they deceptively marketed chatbots as mental health tools, allegedly misleading kids and other vulnerable users, quoting TechCrunch. The probe comes a few days after Senator Josh Hawley announced an investigation into Meta following a report that found its AI chatbots were interacting inappropriately with children, including by flirting. The Texas Attorney General's office has accused Meta and Character AI of creating AI Personas that present as, quote, professional therapeutic tools despite lacking proper medical credentials or oversight. Among the millions of AI Personas available on Character AI, one user created bot called Psychologist has seen high demand among the startup's young users. Meanwhile, Meta doesn't offer therapy bots for kids, but there's nothing stopping children from using the Meta AI chatbot or one of the Personas created by third parties for therapeutic purposes. We clearly label AIs and to help people better understand their limitations, we include a disclaimer that responses are generated by AI, not people, Meta spokesperson Ryan Daniels told TechCrunch. These AIs aren't licensed professionals, and our models are designed to direct users to seek qualified medical or safety professionals when appropriate. However, Techcrunch noted that many children may not understand or simply ignore such disclaimers. We have asked Meta what additional safeguards it takes to protect minors using its chatbots. For its part, Character includes prominent disclaimers in every chat to remind users that a character is not a real person and everything they say should be treated as fiction, according to Character AI spokespeople. The spokesperson noted that the startup adds additional disclaimers when users create characters with the word psychologist, therapist, or doctor to not rely on them for any type of professional advice. In his statement, Attorney General Paxton also observed that though AI chatbots assert confidentiality, their Quote Terms of Service revealed that user interactions are logged, tracked and exploited for targeting, advertising and algorithmic development, raising serious concerns about privacy violations, data abuse and false advertising. According to Meta's privacy policy, Meta does collect prompts, feedback and other interactions with AI chatbots and across Meta services to improve AIs and related technology. The policy doesn't explicitly say anything about advertising, but it does state that information can be shared with third parties like search engines for more personalized outputs. Given Meta's ad based business model, this effectively translates to targeted advertising. Character AI's privacy policy also highlights how the startup logs identifiers, demographics, location information and more information about the user, including browsing behavior and app usage patterns. It tracks users across ads on TikTok, YouTube, Reddit, Facebook, Instagram and Discord, which it may link to a user's account. This information is used to train AI, tailor the service to personal preferences, and provide targeted advertising, including sharing data with advertisers and analytics providers. End quote We've all been there. Too many SaaS tools, not enough visibility at all, and way too much access for you to keep track of. It's the stuff security nightmares are made of. That's where Trelica by1Password comes in. They inventory every app your company uses and create app profiles to help you easily assess risks, manage access, and make sure your password security is locked down tight. With 1Password's extended access management, you can control your company's many, many SaaS tools, securely onboard and offboard your people, and actually hit your compliance goals. I've been telling you about 1Password Extended Access Management all year, and now Trelica comes along to make things even better. Sleep Easy with Trelica by 1Password Learn more@1Password.com Rise that's 1Password.com Ride okay, I know we've spoken about this a lot, but the Financial Times says what is on a lot of people's minds lately that is that GPT5's underwhelming performance on benchmarks might suggest that the current approach to scaling LLMs is starting to reach the limits of available resources for GPT5. People expected to discover something totally new, says Thomas Wolf, co founder and chief executive officer of open source AI startup Hugging Face. And here we didn't really have that. Following hundreds of billions of dollars of investment in generative AI and the computing infrastructure that powers it, the question suddenly sweeping through Silicon Valley is what if this is as good as it gets? Over the past three years, AI researchers, users and investors have become accustomed to a breakneck pace of improvement. Where once OpenAI seemed to have an insurmountable lead, rivals Google, Anthropic, Deep Seq and Elon Musk's XAI have narrowed the gap at the front. Development that intensifying race fueled promises that AGI is imminent, with Sam Altman even predicting it would come during Donald Trump's presidency. A lot of those expectations, which underpin OpenAI's mooted new $500 billion valuation, collided with reality when GPT5 underwhelmed. GPT5 was the central icon of the entire approach of scaling to get to AGI, and it didn't work, says Gary Marcus, a prominent critic of AI and professor emeritus of psychology and neuroscience at New York University. Stuart Russell, professor of computer science at the University of California, Berkeley, was one of the earliest researchers to warn of the dangers of AI's capabilities outrunning humans ability to control them. But now he likens what is happening today to the start of the AI winter in the 1980s, when the innovations of the day failed to deliver on expectations and offer a return on investment. Part of the problem lies in the way companies have been building large language models for the past five years. Companies such as OpenAI and Anthropic have been able to show consistent gains in the performance of their systems by using a simple formula. More data and more computing power equals bigger, better models. Many AI leaders still believe that these scaling laws can continue to hold for years to come. But the approach is starting to reach the limits of available resources. First, AI companies have sucked up all of the available free training data on the Internet. They are now seeking more fuel for their models by making data sharing deals with publishers and copyright holders, but it is unclear if that is enough to push the frontier forward. AI labs are also constrained by computing power. It is very energy intensive to train and run large AI models. Back in 2022, GPT4 was trained on several thousand of Nvidia's chips. Estimates suggest that GPT5 was trained on hundreds of thousands of Nvidia's next generation processors, with even more powerful chips on the way. Altman acknowledged this week that his company is bumping up against some limits. While underlying AI models are still getting better at a rapid rate, he told reporters at a San Francisco dinner, chatbots like CH are quote, not going to get much better. Some AI researchers say that the overwhelming focus on scaling large language models and transformers the architecture underpinning the technology which was created by Google in 2016, has itself had a limiting effect coming at the expense of other approaches. We are entering a phase of diminishing return with pure LLMs trained with text, says Yann LeCun Mehta's chief scientist who is considered one of the godfathers of modern AI. But we are definitely not hitting a ceiling with deep learning based AI systems. Trained to understand the real world through video and other modalities, these so called world models are trained on elements of the physical world beyond language and are able to plan reason and have persistent memory. This new architecture could yet drive forward progress in self driving cars, robotics or even sophisticated AI assistants. There are huge areas for improvement, but we need new strategies to get there, says Joel Pineau, the former Meta AI research lead, now chief AI officer at startup Cohere. Simply continuing to add compute and targeting theoretical AGI won't be enough, and it may not have been OpenAI's intention, but what the launch of GPT5 makes clear is that the nature of the AI race has changed. Instead of merely building shiny bigger models, says Syash Kapoor, a researcher at Princeton University, AI companies are slowly coming to terms with the fact that they are building infrastructure for products. OpenAI and other leading AI companies such as Cohere, Mistral and XAI have started employing so called forward deployed engineers who are embedded into client companies to integrate their models into their client systems. Companies wouldn't do that if they thought they were close to automating all of human work for the rest of time, kapoor says. Many believe there is a huge amount of value yet to be unlocked in the current generation of models. Startups and businesses have not begun to scratch the surface of what they are capable of in business and consumer applications, says Peter Deng, former OpenAI, Uber and Facebook executive, now general partner at venture capital firm Felicis, which has invested in AI coding company Poolside and video generation startup Runway Way. GPT5 may have underwhelmed, but with Silicon Valley running more on vibes than scientific benchmarks, there are few indications that the AI music will stop anytime soon. There's still a lot of cool stuff to build, Wolf of Hugging Face says, even if it's not AGI or Crazy Super Intelligence Hey, I forgot to give you a heads up to the slight ad changes on yesterday's show, because I didn't know they were going to happen yesterday. As I said, one of the reasons I moved the show over to Morning Brew was in search of a slightly higher class of advertisers. And right now we're sticking with the usual 2 ads per show maximum. And you might notice that these ads are shorter than what you're used to. And the ad for ad breaks will usually come in the middle of the show, as per always. But also as you heard yesterday and today, some days the two ads will be split up. One will come after the intro to the show and one in the middle of the show, as per usual. This week is one of those times. That's the only change. Still, two ads and the ads are shorter, but sometimes they're broken up in terms of where they are in the show. No biggie. Just wanted to acknowledge that. Talk to you tomorrow.