← Tech Brew Ride Home
Tech Brew Ride Home
Is AI About To Get More Expensive
00:20:25
With everyone suddenly chasing Anthropic, are we getting strong signals AI is about to get more expensive? The dream of pairing AI models against each other and asking them to fight. We really are getting datacenters in space startups, aren’t we? And Project Hail Mary is Amazon’s biggest movie win ever.
Loading summary
Transcript8 lines
- [00:01]
A
It's not just something you made, it's the privilege that you get to work with your hands. It's building something that serves a purpose, proof that you have the grit to keep going. At Timberland, we understand you take your craft seriously, and we do too, which is why our products are built to the highest quality. We put in the work so you can perfect yours with purpose, in every detail, and crafted with intention. Timberland Built on craft. Visit timberland.com to shop.
- [00:34]
B
Welcome to the Tech Brew Ride Home for Monday, March 30, 2026 I'm Brian McCullough. Today, with everyone suddenly chasing anthropic are we getting strong signals? AI is about to get more expensive. The dream of pairing AI models against each other and asking them to fight. We really are getting data centers and space startups now, aren't we? And Project Hail Mary is Amazon's biggest movie ever. Here's what you missed today in the tech. So the Wall Street Journal has a big long form piece out dissecting the whole rise and fall of OpenAI's Sora. QUOTE Sam Altman wanted OpenAI to be the AI company that used the technology to reshape popular culture and Entertainment. In early 2025, he asked former Twitter CEO Parag Agrawal to informally consult a team inside OpenAI that was working on a separate social media project similar to X people familiar with the matter, said Altman, working with then Disney CEO Bob Iger on a deal to enable the entertainment giant's fans to bring their favorite characters to life with AI through Sora. OpenAI previewed a version of Sora to its employees in late September before releasing it to the public at the end of that month. The product garnered mixed reviews internally, some employees felt launching a social media app built around engagement would hurt the company's brand. Others had concerns about the safety implications of allowing users to create AI generated videos, even with guardrails. OpenAI researchers are able to track how AI chips are allocated between different groups inside the company through an internal dashboard. Some of them were surprised by the amount of computing resources the company gave to the Sora team, given that video generation tools didn't make much money nor improved the capabilities of its large language models. Sora's work was closely guarded from the rest of the company, leading some former employees to describe the project as a startup within a startup. The worldwide user count for Sora peaked at roughly a million soon after the app's launch, but never reach that level again. In the subsequent months, it dwindled to less than 500,000 according to data from SimilarWeb. Now a lot of that is stuff we've heard before, so I'm adding this analysis from a friend of the show, Andrew Koran, because I find it more interesting and probably closer to the metal in terms of what's actually going on here. Quoting his Twitter post 3 weeks ago There were rumors that one of the labs had completed its largest ever since successful training run, and that the model that emerged from it performed far above both internal expectations and what people assumed the scaling laws would predict. At the time, these were only rumors and no lab was attached to them. But in light of what we now know about Mythos from Anthropic, they look more credible and the lab was probably Anthropic. Around the same time, there were also rumors that one of the Frontier labs had made an architectural breakthrough. If you are in enough group chats, you hear claims like that constantly, but most turn out to be nothing. However, if Anthropic found that training above a certain scale or in a certain way at that scale produces capabilities that sit far above the prior trend line, then that is an architectural breakthrough. I think the leaked blog post was real, but still a draft. Mythos and Capybara were both candidate names for the new tier, though Mythos may now have enough mindshare that they end up keeping that. The specific rumor in early March was that the run produced a model roughly twice as performant as expected. That remains unconfirmed. What is confirmed is that Anthropic told Fortune the new model is a step change. A sudden 2x would fit the definition of a step change. We will find out in April how much of this is true. My own view is that the broad shape of this is correct even if some of the numbers are wrong. And if it is substantially accurate, then it also casts OpenAI's recent restructuring in a new light. If very large training runs are about to become essential to staying in the game, then a lot of their recent decisions like dropping Sora even more sense strategically for the public. This would mean the best models in the world are about to become much more expensive to serve and therefore much more expensive to use. That will put pressure on rate limits, pricing and subscription plans that are already subsidized to some unknown degree. Instead of becoming too cheap to meter, Frontier Intelligence may be about to become too expensive for most of humanity to afford second order effects. Here, compute, memory and energy are about to become much more important than they already are. In the blog they described how the new models are not just an improvement, but having dramatically higher scores than Opus 4.6 in coding and reasoning and as being far ahead of any other current models. If this is the new reality, then scale is about to become king in a whole new way. It would also mean, as usual, that Jensen wins again. Well shot and chaser on that over the weekend, Anthropic adjusted Claude session limits and says users will hit their limits faster during peak hours amid compute strain due to Claude's new popularity. Quoting TechRadar, Anthropic is reducing message limits for even Pro and Max customers during its peak hours in a new effort to cope with demand. To manage growing demand for Claude, we're adjusting our five hour session limits for free Pro Max subs during peak hours, your weekly limits are remain unchanged, said Tariq Shihepar, an engineer who works on Claude code and a post on X. Unlike ChatGPT, which has a daily message limit, Claude operates in five hour windows. Once you've hit your limit in a five hour window, you have to use a less premium model or wait for your next window refresh to use it again. An AI company Changing the rules after you've already signed up is the sort of thing that usually provokes a user backlash, but it seems to happen quite often. OpenAI ditching AI video generator Sura with no notice this week springs to mind as an example. This time, however, it's Anthropic who is pulling the rug out from under its users by changing its rules around message limits. Shira park continues during weekdays between 5am to 11am Pacific Time, 1 to 7pm GMT. You'll move through your five hour session limits faster than before, so if you are in the Pacific time zone, you might want to use Claude mainly from the late morning onwards. If you don't want to burn through your message allowance for users in the Greenwich Mean Time, you're best avoiding the afternoon and early evenings. If you live in another country, you'll have to work out when Claude's peak hours will affect you. According to Shipar, around 7% of users will hit session limits they wouldn't have before, particularly for pro tiers. If you run token intensive background jobs, shifting them to off peak hours will stretch your session limits further. Earlier this month, Claude temporarily doubled usage rates for everyone outside peak hours. In contrast, that announcement was made via the official Claude account on X, even though the increased rates were only temporary. From March 13 through March 28, Claude has always operated a pay as you go system for getting extra usage, provided you are on a paid Claude plan. Pro Max 5X or Max 20X, which allows you to continue working with CLAUDE after reaching your plan's usage limits. Details of how to enable this can be found on the Claude website, but in practical terms, the rule change means you need to start thinking more strategically about when you use your Claude account, especially since the time of day when peak hours occur varies widely by country. End quote. What if you wanted to use two different AI models and then ask AI which of the models was giving you better output? Well, Microsoft has released Copilot cowork to its Frontier program for early stage testing and that includes a new Researcher critique tool using anthropic and OpenAI models. Quoting Thurat, Microsoft's Researcher AI Agent is getting a new critique feature that uses OpenAI's GPT and anthropic cloud models together to improve the quality of its responses. A new Council model in Researcher can also be used to compare responses from different AI models side by side. CLAUDE models became available in Researcher as an alternative to GPT models back in September. With the new Critique model, the AI AI Agent will use both models at the same time and combine their strengths for deep research tasks. GPT drafts CLAUDE reviews for accuracy, completeness and citation integrity before it's delivered. We expect this workflow to be bidirectional in the future, Microsoft said about the new Critique feature in Researcher. The company also pointed out that Researcher with Critique now scores 13.8 higher on the Draco benchmark, which analyzes deep research quality. The new console mode in Researcher, which delivers side by side comparisons across Anthropic and OpenAI models, is also available today from the Model Picker. It will give users two separate reports created with each model, as well as a summary of the key points of agreement and divergence. Microsoft's researcher agent is available to everyone with a Microsoft 365 copilot license. In addition to these Researcher updates, the company also announced today that its Claude powered Copilot Cowork agent, which is designed to complete multi step tasks, is now available in early access via its Frontier program. End quote.
- [10:10]
A
The world moves fast.
- [10:11]
B
Your workday even faster Pitching products, Drafting reports analyzing data Microsoft 365 copilot is your AI assistant for work built Into Word, Excel, PowerPoint and other Microsoft 365
- [10:25]
A
apps you use, helping you quickly write,
- [10:28]
B
analyze, create and summarize so you can
- [10:31]
A
cut through clutter and clear a path to your best work.
- [10:34]
B
Learn more@Microsoft.com M365 Copilot Cap Table Management who needs it? Well, you probably do, but that doesn't mean it should drain your time or derail your budget. Pulley knows there's a better way. That's why they help take the complexity and surprises out of equity management. Pulley's intuitive workflows, built in compliance tools and decision ready reporting are designed to work for you, not against you. Pulley helps you issue, track and manage equity, stay compliant with up to date 409A valuations, complete stock based compensation reporting, and more. Learn more and get started@pulley.com brew that's pulley.com brew some days it just feels like my entire job is keeping up with the paperwork. I've got forms for taxes right now, contracts, scheduling for interviews, sucking up every single spare minute I have. That's why upgrading to Gusto is such a game changer. Gusto is online payroll and benefits software built for small businesses. It's all in one remote, friendly and incredibly easy to use so you can pay, hire onboard and support your team from anywhere. Gusto's got it all. Payroll tax filing, direct deposits, health benefits, commuter benefits, workers comp, even 401ks. They've even got tons of options for pretty much all budget sizes. Try gusto today@gusto.com brew and get three months free when you run your first payroll. That's three months of free payroll@gusto.com brew I guess we are going to start getting these coming down the pike, Space Data center startup StarCloud raised a $170 million Series A round led by Benchmark and EQT at a $1.1 billion valuation and plans to launch its StarCloud 2 later in 20. TechCrunch StarCloud's latest funding round values the Space Compute company at $1.1 billion, making it one of the fastest startups to reach unicorn status after graduating from Y Combinator. The company's Series a, which closed 17 months after its Demo day presentation, was led by Benchmark and EQT Ventures. It's another sign of the interest in the outsourcing data centers to orbit as resource and political obstacles slow their development on Earth. But the business model depends on unproven technology and significant capital expenditure. Starcloud has now raised a total of $200 million and launched its first satellite with an Nvidia H100 GPU in November. The company will launch a more powerful version, StarCloud 2, later this year with multiple GPUs, including an Nvidia Blackwell chip and an AWS server blade, as well as a bitcoin mining computer. The company will also begin developing a data center spacecraft designed to launch from Starship. The reusable heavy lift rocket being built by Elon Musk's SpaceX. StarCloud 3, as the spacecraft is named, will be a 200 kilowatt 3 ton spacecraft that fits the PEZ dispenser system. SpaceX designed to deploy its Starlink satellites from Starship. CEO and founder Philip Johnson said he expects that will be the first orbital data center that is cost competitive with terrestrial data centers, with costs on the order of $0.05 per kilowatt hour of power if commercial launch costs land around $500 per kilogram. The challenge is that Starship isn't flying yet. Johnston says he expects commercial access to open up in 2028 and 2029. That's the reality facing all the big space data center projects. Powerful space computers will be cost prohibitive until a new generation of rockets starts launching at a high operational cadence, something that might not happen until the 2000-30s. If it ends up being delayed, we'll just carry on launching the smaller versions on Falcon 9, Johnston said. We're not going to be competitive on energy costs until Starship is flying frequently. There's two business models, Johnson explains. One is selling processing power to other spacecraft on orbit. The company's first satellite, for example, analyzes data collected by Capella Space's radar spacecraft. Then, in the future, when launch costs go down, more powerful distributed data centers could potentially pull work from their terrestrial counterparts that gets at how this new industry really is. When Nvidia CEO Jensen Huang unveiled the company's Vera Rubin Space one chip modules at his company's annual GPU technology conference last week, he didn't note that none had been produced or shared with the company's development partners. In fact, the number of advanced GPUs on orbit is numbered in the dozens, while Nvidia is estimated to have sold nearly 4 million to terrestrial hyperscalers just this year. Or consider that SpaceX's Starlink communications network, the largest satellite network in orbit with 10,000 spacecraft, produces something around 200 megawatts of energy, while data centers with more than 25 gigawatts of power are currently under construction in the U.S. according to Cushman and Wakef. Johnson argues that his company is well ahead of the competition, though with the first terrestrial GPU deployed in orbit, it was used to train an AI model in orbit a first, according to Starcloud, and run a version of Gemini. Beyond the performance, Johnson says Starcloud now has valuable data about what it takes to run a powerful chip in space, there's a laundry list of technical challenges to be solved, including efficient power generation and cooling the hot running chips. StarCloud 2 will have the largest deployable radiator flown on a private satellite. He expects at least two additional versions of that spacecraft will head to orbit. Then there is the challenge of synchronization. The largest data center workloads, often for training, require hundreds or thousands of GPUs to work in tandem. Doing that in space will either require fantastically large spacecraft or powerful and reliable laser links between spacecraft flying in formation. Most companies working on this technology expect those workloads to come along after simpler inference tasks take place on orbit. Besides StarCloud, Aetherflux, Google's Project Suncatcher and ethereal, which launched Nvidia's first space based Jetson GPU in 2025, are all developing space data center businesses. The elephant in the room is SpaceX itself, which has asked the US government for permission to build and operate a million satellites for distributed compute in space. End Quote. Amazon MGM's project Hail Mary has become its highest grossing film ever, crossing $300 million globally, including $54.1 million at the box office just this past weekend. The movie only cost $200 million to make, but you know they're in the black quoting TechCrunch that's a big price tag for any film, but especially one that's not a sequel or part of an existing franchise. Instead, it's based on a best selling science fiction novel by Andy Weir, whose book the Martian was adapted into a successful film a dec. That's not the only thing that makes Project Hail Mary feel unconventional. For long stretches of the film, Ryan Gosling is the only human actor on screen as the scientist he plays works with a rock like alien to solve the mystery of why multiple stars, including our own, seem to be dimming. This weekend's numbers make Project Hail Mary the biggest hit of 2026 so far, as well as one of the most successful non franchise, non sequel films of the past decade. And it's good news for what's now known as Amazon MGM Studios. The company's cinematic ambitions have evolved over time from distributing smaller critically acclaimed titles like the Big Sick and Manchester by the Sea to more recently acquiring movie studio mgm, leading to a battle for control of the James Bond franchise and declaring its intention to bring 14 movies to theaters every year until Hail Mary. Those movies, including after the Hunt, Mercy and the controversial Melania documentary, seemed to be falling flat with audiences Amazon's head of film, Courtney Valenti, told the New York Times that project Hail Mary's big opening weekend validated the company's strategy of making big, bold, entertaining commercial films. And it has more movies coming to theaters soon, including the Sheep Detective, starring Hugh Jackman in May and a Masters of the Universe reboot in June. End quote. This show is a little late today because my son helpfully gave me a lovely spring cold. My throat has never hurt so much doing a show. I delayed hours hoping that it would feel better, but in the end I just had to power through with shards of glass in my throat. Talk to you tomorrow.