Photo-Illustration: Intelligencer; Photo: Getty Images
Start-ups might be in a slump, but the biggest players in tech are still investing incredible sums of money into AI. Microsoft and OpenAI have proposed building a $100 billion supercomputer to train future models. Meta expects to spend upward of $40 billion by the end of 2024, while Google expects to spend even more. Elon Musk’s xAI is spending billions to stand up data centers so large that they need natural-gas turbines to meet their power demands.
At the same time, the AI giants are still scrambling to figure out how they might even begin to make this money back. OpenAI, which is on track to generate around $3 billion of subscription revenue in 2024, is reportedly considering higher-priced subscriptions to help offset its massive operating costs. Google is still figuring out which parts of its AI software portfolio it can charge for, and how much, while both Amazon and Apple are rumored to be working on paid versions of upcoming Alexa and Apple Intelligence features, respectively — even Meta is wondering if it might be able to charge for its AI assistant. Meanwhile, according to the Information, Microsoft, which has been bundling paid AI features into its productivity software for a while now, has been battling a “lukewarm” reception from business customers due to “performance and cost issues.”
The gentle way to describe what’s happening is that the companies spending the most on generative AI are betting that customers will soon come and be willing to pay. A more honest way might be to say that, with potentially hundreds of billions of dollars missing from balance sheets in the near future, they simply need people to pay something for anything, and the best plan they’ve come up with — or borrowed from OpenAI — is subscriptions.
The most obvious challenge here is that most potential customers don’t yet know why they should pay for generative AI tools and chatbots of any sort. It’s a challenge these companies hope to overcome with a combination of aggressive salesmanship and, ideally and eventually, software so self-evidently good and valuable that it sells itself. In the meantime, though, two slightly counterintuitive factors are working against big tech’s build-it-and-they-will-come plan. One is that for customers that already pay for AI — developers buying major capacity from AI providers, for example — competition between AI firms and open-source AI models has driven the price of using AI dramatically down. While companies like OpenAI and Google are brainstorming ways to charge future customers more, they’ve been engaged in a race-to-the-bottom price war to keep the customers that they already have. (OpenAI itself has boasted that its cost-per-token has fallen by 99 percent since 2022.)
For anyone interested in using generative AI, this is great news: Leading models are becoming more efficient and cheaper to use at scale, and there are lots of alternatives for different tasks. For OpenAI, which has been able to reduce the cost of serving its low-paying customers and vastly more numerous free users, it’s a mixed bag. Its customers, who have been conditioned through a dazzling hype cycle to think in terms of exponential progress, are expecting a lot more for a lot less. Customers have been complaining about ChatGPT’s performance and pricing for nearly as long as it’s been around. In theory, AI companies will soon be charging more than ever for next-generation models. In practice, in the actual marketplace, competition is driving prices to the floor.
The other complicating factor in big tech’s plan for AI is about to become more obvious: Users who don’t already pay for AI are becoming accustomed to using it for free. This, too, is something that big AI firms are acutely aware of — they, and their competitors, are the ones rolling out free-to-use features to hundreds of millions of users. There’s a ChatGPT-grade chatbot tucked into Facebook, Instagram, and WhatsApp; some of Google’s AI features, including chatbots and AI-powered search results, are showing up for users whether they ask for them or not, and are built right into newer Android phones; Microsoft includes a limited free version of Copilot, its AI assistant, in Windows. Mostly, these free offerings are part of a current or potential upsell. But they run the risk familiar to companies, like Meta and Google, that have tried to push into subscription services after years of offering services subsidized by ads. It’s hard to get people to pay for things that they’ve grown accustomed to using for free. It doesn’t help that most of these features haven’t been breathtakingly impressive to regular users, despite marketing suggesting that they’re revolutionary. They feel, instead, like routine software updates.
On Monday, Apple announced its new iPhones for late 2024, which will ship with Apple Intelligence features for free (or at least free for now). These features, which include writing help, summarization tools, photo editing, and a more fluent and capable version of Siri are fairly conservative expressions of what’s currently possible with generative AI — they’re not going to make iPhone 16 buyers feel like they’re interacting with a superintelligent machine. But having extensively used most of these new Apple features for a while now (as well as free and cheap paid offerings from Meta, Google, OpenAI, and others), it’s clear that users will, in a very short time, take them for granted. That’s not to say they aren’t useful or occasionally impressive: While Apple’s AI notification summaries are still frequently strange or wrong, they attempt and at least partially solve the platform’s massive notification-spam problem; the new writing-assistant tools are unobtrusive and mostly welcome, more like an extension of autocorrect and spellcheck than an impertinent Clippy; Apple’s new photo-editing tools are useful; and Siri, in the end, might end up being usable for more than setting timers and reading texts. Apple is promising more advanced features to come and will integrate outside chatbots as well, but with the release of the next iPhone and the next version of iOS, Apple is about to get a lot of people used to getting a lot of AI for free.
This isn’t an emergency for Apple, of course, because Apple makes most of its money selling phones and these features may help it do that. But most of the other companies giving away AI don’t sell $1,000 handsets. They’re either companies with legacy advertising businesses, like Meta or Google, or AI firms like OpenAI and xAI that are banking on building subscription businesses, winning the race to achieve AGI, or some combination of both. Altman and Musk might be imagining a future in which employee-like AI agents are priced against human labor, but anything short of that is a world in which generative AI is understood by most people as a set of new tools and incremental features that the public expects access to for next to nothing.
Source link
