Tech Brew Ride Home - Thu. 07/17 – AI Pricing

Episode Date: July 17, 2025

TSMC earnings suggest the AI buildout is continuing apace. Stablecoin regulation clears a major hurdle. Is Anthropic doing well growing revenue? Maybe not OpenAI well, but well enough? More signs Microsoft is struggling to sell its own branded AI. And how AI might be about to change how we pay for everything… forever. Links: TSMC Raises 2025 Outlook in a Big Boost for AI Demand Hopes (Bloomberg) US House agrees to consider crypto legislation in big win for the digital asset industry (Reuters) Investors Float Deal Valuing Anthropic at More Than $100 Billion (The Information) Microsoft's Copilot Is Getting Lapped by 900 Million ChatGPT Downloads (Bloomberg) Delta moves toward eliminating set prices in favor of AI that determines how much you personally will pay for a ticket (Fortune) 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 TechMean right home for Thursday, July 17th, 2025. I'm Brian McCullough today. TSM earnings suggests the AI buildout is continuing apace. Stable coin regulation clears a major hurdle.

Is Anthropic doing well growing revenue, maybe not open AI well, but well enough? More signs Microsoft is struggling to sell its own branded AI and how AI might be about to change how we pay for everything forever. Here's what you miss today in the world of tech. We don't typically discuss TSM earnings on this show, but as a bit of further data suggesting the AI buildout is still going strong, I'd like to make note of the fact that TSM reported Q2 net income up 61% year-on-year to $13.5 billion, meaning they have now beat estimates every quarter they've reported since 2021, and the company raised its 2025 sales growth guidance from mid-20% to 30% so 30% growth. Quoting Bloomberg, the world's biggest contract chipmaker on

Thursday forecast sales growth of about 30% in US dollar terms this year up from mid-20% previously. That reinforced expectations that tech firms from meta to Google will keep spending to build the data centers essential to artificial intelligence development. NASDAQ stock index futures swung to gains, while top supplier ASML holding gained more than 2%. TSM's move underscores resilient demand for high-end chips from the likes of Nvidia and AMD, which is outpacing its production capacity. Chief Executive Officer CCY confirmed on Thursday that AI orders still run hot,

seeking to dispel persistent speculation that tech firms may curtail spending, while he stressed that underlying AI demand is strengthening the uncertainty around the Trump administration's tariffs merited caution. This is, quote, supporting the AI value chain and AI optimism still holds, said Billy Long, investment strategist at Global X ETFs in Sydney. For investors, TSM results again ease fears of an AI slowdown. Margins hold, demand outlook, good, generally reinforces the AI buildout is still well underway, end quote. Related, TSMC CEO Y says the company is speeding up construction of its second and third

Arizona plants by several quarters to meet chip demand from U.S. clients. I almost did a segment yesterday about how several crypto regulation bills backed by the Trump administration had failed to clear a key hurdle in the U.S. House of Representatives after 13 Republicans voted with Democrats to block the legislation moving forward. But glad I waited because I can now tell you that the House voted to advance those crypto bills following over nine hours of private talks on July 16th, a day after President Trump intervened to save the bills. Quoting Reuters, a bill to establish a federal framework for stablecoins is likely to be the first to be passed in what would be a watershed victory for the crypto industry. It has already been

approved by the Senate, and if approved by the House, it would go to Trump for his signature. Stablecoins, a type of cryptocurrency designed to maintain a consistent value, usually a one-to-one dollar peg, are commonly used by crypto traders to move funds. They have gained much momentum. them in recent years, offering faster and cheaper transaction costs than moving money through a bank. In addition to stablecoins, the House is set to consider a bill to establish market structure rules for crypto products, including defining when the products are a commodity and not subject to oversight from the Securities and Exchange Commission. The Senate has yet to take up a similar measure. The third bill, strongly backed by conservatives, would prohibit the Federal Reserve from

issuing a digital currency of its own. Some Republicans argue a Fed digital currency could give the government too much control over Americans' finances, current Fed leaders have said they are not considering such an initiative, end quote. The information has sources who say that Anthropics financial performance has prompted some investors to indicate interest in additional funding of the startup at a greater than $100 billion valuation. As I said on the socials last night, this is interesting to me because there had been some whispering that, while Open AI was reportedly blowing the doors off the place in terms of revenue growth, maybe some of their peers were not. The idea was Open AI might be winning the branding race. But this news suggests otherwise,

at least for Anthropic, quote, as Anthropics' sale of artificial intelligence has surged, the company has told investors some of its profit-related metrics are improving, even as it burns a tremendous amount of cash. That financial performance has prompted some investors to indicate interest in funding Anthropic at a valuation of more than $100 billion if the company decides to pursue a deal, compared to a $58 billion valuation in a financing announced four months ago, according to two people involved in the discussions. The company raised $3.5 billion in equity financing in March, but previously told some investors it planned to raise $5.5 billion in total this year,

and it recently let some investors see part of its financial performance, one of these people said. In another piece of good news, Anthropic hired back two former leaders of its coding product who had joined any sphere, a rival that operates the cursor coding app, just two weeks ago. Anthropic recently told investors its gross profit margin from selling its AI models and Claude Chapot directly to customers was roughly 60% and is moving towards 70% according to two people with knowledge of the financials. The margin figure typically refers to gross profit as a percentage of revenue after accounting for the cost of the servers and customer support required to power the company's revenue-generating products. Those gross margins can fluctuate depending on how efficiently the company plans and uses its computing resources including AI chips that power its technology, the person says.

The 60% gross margin figure doesn't tell the whole story, though. Anthropic also sells cloud models through Amazon Web Services and Google Cloud, and that business was generating a gross margin of negative 30% earlier this year, these people said. This could be because Amazon and Google, which both have also invested billions into Anthropic, take a significant cut of revenue when they sell Anthropic models to cloud customers. Anthropic sales through cloud providers is likely a minority of its revenue. As of the end of 2024, about 70% of Anthropics' revenue came from direct sales, rather than through the cloud providers, though it previously projected that in the coming years, cloud providers would account for much of the revenue it generates from selling models to other

businesses. It isn't clear what percentage of Anthropic revenue comes from direct sales today, but the latest disclosures suggest Antropics' overall gross margin may not have improved since the end of 2023 when the figure was between 50 and 55%. In comparison, Open AI earlier this year projected a gross profit margin of 48% in 2025. Open AI projected steady improvements in the coming years en route to an eventual 70% gross profit margin by 2029. It isn't clear whether the two firms calculated their margins the same way. Of course, Anthropic and OpenAIs gross margins don't reflect the billions of dollars a year they spend to develop their AI and to cover operating expenses such as salaries, all of which are higher than traditional software businesses. Overall,

Open AI had been operating more efficiently than Anthropic. It has been burning significantly less cash, despite generating several times more revenue, according to its financial disclosures. as part of its fundraising process in the first quarter. Anthropic told investors it will burn $3 billion this year after burning $5.6 billion last year, and it projected rapid growth to as high as $12 billion in revenue by 2026 on the strength of its market-leading AI for coding-related tasks. Anthropics revenue increased four times in the first half of the year to more than $4 billion in annualized revenue calculated as last month's revenue multiplied by 12. The Claude Chatbot now includes Claude, a coding assistant powered by Anthropic models, which the company says is growing

quickly after becoming broadly available to customers in May. For instance, weekly downloads of Claude Code increased sixfold to $3 million from June, the company has told investors. The company has also told some investors that it attributes more than $200 million in annualized revenue or more than $16.7 million per month to Claude Code, which can be accessed as part of a Cloud Chatbot subscription or through Anthropics Application Programming Interface, end quote. in a chart accompanying this piece, OpenAI is reportedly projecting $50 billion in revenue by 27. As comparison, while Anthropics' optimistic projection for that same year is around $35 billion in revenue, the base, which I assume is the conservative projection, is in the neighborhood

of $12 billion by $27, I guess if things don't go as well as they hope. I swear it's like the dot-com days these days in terms of everything. being horse race news. Who's up, who's down, who's getting traction and raising a new round, who's maybe not getting traction and maybe burning money. Bloomberg has a piece positing that Microsoft's co-pilot is struggling to make headway against rival AI assistance. Sensor Tower says copilot's mobile app has around 79 million downloads far below chat GPTs greater than 900 million. What was I just saying about Open AI may be winning the branding war? quote, Microsoft shares have surged about 20% so far this year, based largely on Wall Street's

expectations that the company's AI bet will help secure its future, but some investors are starting to get impatient. They have to win this, said Gil Luria, an analyst with DA Davidson. If they don't, someone else will. Microsoft is staking its future on three co-pilot-branded products, a coding assistant for developers, a workplace helper embedded in the likes of Outlook and Word, and a personal assistant built to help people navigate and understand the world. At an all-hands meeting in May, Chief Executive Officer Sachin Adela told employees the goal is to get hundreds of millions of people using Microsoft's family of AI apps, Bloomberg previously reported. The company started baking AI into its products two years ago.

The Bing search engine was restyled as an AI companion for the web. Windows users were told to get ready for a chatbot that would personalize and navigate your PC. But behind the scenes, engineers were struggling to create the new world executives were pushing for. They had access to the same raw material, the large language models built, OpenAI, but mostly came up with slightly different spins on how chatbots might improve the lives of users searching the web or writing an email. Whatever advantage Microsoft had, thanks to its close ties with OpenAI, wasn't translating into hoped for market share gains in products like Bing. Nadella eventually tired of the halting progress and recruited Mustafa Suleiman 15 months ago to run

Microsoft's consumer AI operation. Depending on who you talk to, Suleiman was either an inspired or a risky hire. A British founder of two well-regarded AI startups, deep mind and inflection, he's widely considered a brilliant recruiter and motivator of engineers. He also acknowledged missteps while managing large teams at Alphabet's Google, including setting, quote, pretty unreasonable expectations. Besides leading the teams working on the consumer-focused co-pilot, he's responsible for a bunch of existing products, the edge browser, the MSN News and landing page, Bing Search, that boasts millions of users but little cultural cachet. Suleiman tends to wax philosophical on the subject of artificial intelligence, thinking aloud

in LinkedIn posts and frequent podcast appearances on what it means to be human and the nature of computer intelligence. Translated, he's essentially saying that he wants to build AI assistance that keep humans in the loop and help better themselves. He professes zero desire to create machines smarter than people for the sake of a milestone. Artificial general intelligence is not our mission, Suleiman said in an interview earlier this year. Products are our mission and we are singularly focused on, is it useful, does it help? Is it supportive? Am I happy? End quote. Soleiman has said that AI will eventually remake graphical interfaces like Windows, but for now, Microsoft executives are wary of alienating users by forcing them to learn new habits and

tend to bolt AI innovations onto existing tools. When the company started rolling out an AI agent to help manage PCs last month, it wound up in settings, not the co-pilot app. There are technical challenges, too. The operating system only gets a few major updates a year and isn't set up to receive frequent tweaks of the sort the co-pilot team are rolling out. That's our big challenge, Suleiman said. There's a kind of annual rhythm to it, and there's also just a degree of freedom that is restricted. So by default, co-pilot is a smartphone app. That's a problem because Alphabet's Android and Apple's iOS power virtually all of the world's mobile devices. Both are also weaving their own artificial intelligence into their mobile operating systems. There's no precedent for Microsoft building a must-have smartphone app from scratch. It's incredibly difficult, especially when the owners of these devices are trying to do the same. thing, said Matthew Quinlan, a former Microsoft manager who tried with little luck to get people using a Cortana smartphone app a decade ago. Microsoft struggles with its consumer co-pilot echoed the challenges the company is experiencing with the version it created for corporations.

Bloomberg has reported that many office workers prefer chat GPT and have been pressuring their bosses to let them use it. Some companies are testing both co-pilot and chat GPT and awaiting employee feedback before deciding whether to use one or the other or both. Microsoft's long-standing relationship with corporate clients gives it leverage in the workplace. Should corporate IT managers deem co-pilot the superior option they can simply tell employees to use it? Microsoft, aside from the nudges it can drop into Windows, has no such way with consumers. Executives say they aren't sweating the user chasm between ChatGPT and Copilot, confident that when the time is right, they can get the product in front of consumers. They see Copilot's emphasis on being a personable companion as a

potential advantage with a younger cohort that tends to use AI tools as sounding boards rather than replacements for web search. A trial advertising campaign this spring catapulted co-pilot up the charts on Apple's app store, the company says. Co-pilot's monthly active users have increased 76% between April and June to 23 million, Sensor Tower says, but the app's growth rate over the last year has trailed its major rivals according to market intelligence firms, end quote. Finally today, yeah, yeah, Brian, AI everything, AI horse races, AI land grabs, maybe AI's coming for my job, maybe not, but so far, Brian, AI hasn't really touched my life directly. Well, buckle up. Fortune says Delta Airlines plans for 20% of its fares to be individually

determined using AI by the end of this year, up from 3% of fares now, with the goal of completely eliminating static pricing entirely. Quoting Fortune, fresh off a victory lap after a better-than-expected earnings report, Delta Airlines is leaning into AI as a way to boost its profit margins further by maximizing what individual passengers pay for fares. Over time, the goal is to do away with static pricing altogether. Delta President Glenn Hohenstein explained during the company's Investor Day in November, this is a full re-engineering of how we price and how we will be pricing in the future, he said. eventually we will have a price that's available on that flight, on that time to you the individual.

He compared AI to a super analyst who is working 24 hours a day, seven days a week, and trying to simulate in real time, what should the price points be? While the rollout would be a multi-year process, he said, initial results, quote, show amazingly favorable unit revenues. Delta accomplishes this pricing through a partnership with Fetcher, a six-year-old Israeli company that also counts a Jules WestJet Virgin Atlanta, and Viva Aero Bus as clients, and it has its sights set beyond flying. Once we will be established in the airline industry, we will move to hospitality, car rentals, cruises, whatever co-founder Robbie Nissan said at a travel conference in 2022. While Delta is unusually open about its use of AI, other carriers are likely to follow. Already, United Airlines uses generative AI to contact passengers about cancellations,

while American Airlines uses it to predict who will miss their flight. personalized pricing has been an airline goal for the past decade and a half, Gary Leff, a travel industry authority who first noted Delta's AI strategy told Fortune. Delta is the first major airline to speak so publicly about its use of AI pricing to tout it for its potential upside at its investor day in the fall and to offer concrete metrics around its use in its recent earnings call, end quote. And indeed, to be sure, airlines have long offered different prices to different people, even for the same route based on factors like Hal Traveler's book, for example, directly,

or via a comparison shopping site or a travel agent, or how far in advance they shop. As far back as a decade ago, travel websites showed different prices for precisely the same itinerary based on details like which browser a purchaser was using to search for fares. But the use of AI supercharges this type of price discrimination and puts airlines into a legal gray area. AI isn't just optimizing business operations but fundamentally rewriting the rules of commerce and consumer experience, Matt Britton, author of Generation AI told Fortune. For consumers, this means the era of, quote, fair pricing is over. The price you see is the price the algorithm thinks you'll accept, not a universal rate, end quote. So I read this, and in the shower,

my mind started spinning. When we talk on this show about how, when corporate America is coming out suddenly and talking about how AI is boosting efficiency, what they're really talking about is money. And yeah, one version of that is, you know, big law firms don't need as many junior associates to turn through documents and discovery. AI can do that now and do it in hours instead of weeks. But also, we're probably talking about this. Pricing is probably about to become a thing of the past, at least how we've understood it. Airlines are a famously low margin business, so if they can get folks to pay even two to three percent additional, that is huge for them. You know, what's another famously low margin business? Groceries. How much are you really willing to

spend for that box of cereal? You might. be willing to pay more than I would for that particular brand. And if that brand knows that, but Brian, how would they achieve differential pricing if you're not buying online, if you're actually in the grocery store? Oh, I don't know. They stop posting prices. The little price stickers go away forever. But then, Brian, how would I know the price? Well, you'd go through the store, basically shopping with your store app on your phone. So you want to know the price, you scan the barcode, and you see a price that is your price. It might be a price. It might be might be a higher price because they know you like that brand more, but I would see a lower price

because they know I've never tried it before and they want to hook me. It's all sort of related to zero marginal cost thinking. The plane's already scheduled to take off, right? It's going to fly anyway, but what if they can get you to pay $100 more than me for the seat next to me? That cereal box has already been manufactured, shipped to the store, or think about a sneaker. Nike probably knows I can afford to pay $300 for a pair of sneakers, whereas a 14-year-old kid might have only been able to save up about $100 to pay for a pair of sneakers. Could there be some sort of scenario where it's like the inverse of the whole idea of graduated taxing as a system, where less well-off people pay less than taxes in theory,

and rich people pay more in theory? Would I be eventually subsidizing that 14-year-old kid's sneaker purchase by paying more for my pair than they pay for theirs? These are all in the shower, jack-handy, deep thoughts, so not extensively thought through, so feel free to poke holes in what I just said. And I, as always, don't really have an opinion on this either way. I'm just speculating. But I do think pricing like this is inevitably coming, and it's probably coming sooner than we think. Talk to you tomorrow.