A view of the DeepSeek office in China (Long Wei/Getty Images)

Adding Up

The new age of ad-supported AI, brought to you by DeepSeek

If you thought promoted search results were bad, wait till you get a load of ad-supported AI.

2/24/25 11:17AM

DeepSeek’s splash onto the scene with a lower cost and comparable AI model last month caused a moment of reckoning for the industry. While major competitors haven’t yet reversed course on their huge AI capital expenditure outlays, some have made changes to how much money they’re taking in, offering more of their wares to individuals for free.

Baidu reacted by offering its Ernie chatbot free to individual users. OpenAI, which had been losing money on its most expensive $200 per month ChatGPT Pro offering and has 400 million weekly active users, is now promising users of its upcoming GPT-5 unlimited chat access for free. Even before DeepSeek, this had been the trend. Elon Musk’s Grok AI had been only available to paying X users, but in December the company rolled it out free to everyone on the site and it’s now available as a free standalone app for everyone else.

The idea that Amazon or Google or Meta, which are cramming generative AI for free into their existing products, would put up a paywall for regular consumers is more remote than ever.

Meanwhile, tech companies are shelling out hundreds of billions a year to furnish their AI ambitions. And the more people use these services, the more it costs the companies who offer them. Generative AI has yet to afford its creators the shrinking computing costs of earlier web technology.

It seems unlikely that typical consumers would choose to pay for something they can get for free, and currently there’s a variety of comparable options. Though OpenAI’s ChatGPT was the first out of the gate for American consumers, it’s not clear that consumers have particular love for any particular gen-AI model. And every other day a different company says its model is better or at least comparable to the last.

“ChatGPT was significantly better than anything else out there two years ago as far as consumer experience goes,” Arun Sundararajan, a professor at NYU Stern, told Sherwood News. “Now there are plenty of very good alternatives.”

Indeed, in the last month a number of options — DeepSeek, ChatGPT, and now Grok — have traded places at the top of the app stores. Popularity seems to follow whoever has the newest, freest model.

Generative AI, it seems, has become a commodity for typical consumers. (There’s certainly a lot more potential — and money — in enterprise use cases, but even there it seems people are still searching for the killer use case.) As far as consumer generative AI, you have a situation where there’s more and more money going out and now potentially less coming in. Something has got to give.

It’s likely we’re about to see the end of truly free gen-AI chatbots. To help recoup costs from consumer models, ad-supported generative AI is likely coming soon.

“Maybe advertising is on the horizon,” Columbia Business School professor Olivier Toubia said. “Maybe this will become just like search.” He added that perhaps advertising could work in concert with specialized applications and consulting as different types of revenue sources.

Sundararajan said that it’s impossible to know where generative AI and its business model will be in the future, given the uncertainty of what large language models’ performance and price will be even a year from now, but added that advertising has a habit of creeping up on technology.

“Advertising seems to be the fail-safe of the internet era,” he said. “Advertising always comes to the rescue.”

Meta has already signaled that once it hits a billion users of its gen-AI products, it will monetize with ads. Google, which already has an existing user base of people using its sites to find information and which has its own AI model as well as a giant ad business, is also an obvious candidate for launching ad-supported AI.

Of course, there are many technical and ethical implications to consider regarding ad-supported AI. Promoted ads that were clearly labeled were controversial on Google’s Search. What happens when the motivations behind a gen-AI chatbot’s responses or recommendations are even more obscured?

Rani Molla15h

Anthropic’s scramble for compute now includes rival xAI

Another day, another major partnership with an AI rival. This time, Anthropic signed a deal with SpaceX’s xAI to access compute from its Colossus 1 data center to help it improve capacity for its Claude Pro and Claude Max subscribers. Just yesterday, The Information reported that Anthropic planned to spend $200 billion on Google Cloud services over the next five years. As Sherwood News’ Luke Kawa wrote:

“Anthropic has been a victim of its own success: the popularity of Claude Code and Cowork have revealed compute constraints and left users frustrated by caps. In response, the Claude developer has embarked upon a mad scramble for compute, striking or expanding deals with CoreWeave, Amazon, Google, and Broadcom.”

Now, it’s adding xAI to the list — even as the Elon Musk company builds a competing model.

In less terrestrial news, xAI said that as part of the agreement, Anthropic “expressed interest in partnering to develop multiple gigawatts of orbital AI compute capacity.”

xAI — Creators of Grok, the AI Chatbot

“Anthropic has been a victim of its own success: the popularity of Claude Code and Cowork have revealed compute constraints and left users frustrated by caps. In response, the Claude developer has embarked upon a mad scramble for compute, striking or expanding deals with CoreWeave, Amazon, Google, and Broadcom.”

Now, it’s adding xAI to the list — even as the Elon Musk company builds a competing model.

In less terrestrial news, xAI said that as part of the agreement, Anthropic “expressed interest in partnering to develop multiple gigawatts of orbital AI compute capacity.”

17h

TEST TIME!

Meet your new interview partner: Claude

Chris Stokel-Walker

Rani Molla19h

SpaceX and Tesla’s Terafab could cost $119 billion — far more than expected

The initial phase of SpaceX and Tesla’s joint chip production effort, called Terafab, could cost $55 billion, with additional phases adding up to $119 billion in capital investment, Reuters reports, citing a notice posted on a Texas county website. Ultimately the goal of Terafab is to build enough in-house AI chip capacity to supply both companies.

The price tag is also higher than expected. Morgan Stanley had previously estimated Terafab would cost $34 billion to $45 billion.

Fortunately for Tesla, whose capex is expected to skyrocket this year, much of the early spending will sit on SpaceX’s balance sheet.

Here’s Musk on the last earnings call:

“SpaceX is going to take care of like the initial phase of the scaled up Terafab... Any kind of intercompany thing has to be approved by both the SpaceX and Tesla board of directors. It’s got to go through a conflict resolution. It’s going to have, unfortunately, a lot of complexity because we’ve got to make sure Tesla shareholders are served and SpaceX shareholders are served, and strike the right balance there.”

SpaceX files plan for $55 billion Terafab chip facility in Texas

The price tag is also higher than expected. Morgan Stanley had previously estimated Terafab would cost $34 billion to $45 billion.

Fortunately for Tesla, whose capex is expected to skyrocket this year, much of the early spending will sit on SpaceX’s balance sheet.

Here’s Musk on the last earnings call:

“SpaceX is going to take care of like the initial phase of the scaled up Terafab... Any kind of intercompany thing has to be approved by both the SpaceX and Tesla board of directors. It’s got to go through a conflict resolution. It’s going to have, unfortunately, a lot of complexity because we’ve got to make sure Tesla shareholders are served and SpaceX shareholders are served, and strike the right balance there.”

Rani Molla5/5/26

Apple to let users choose between Anthropic, Google, and OpenAI models

Apple has been inching toward letting outside AI power its devices — and now it’s going further.

The company plans to let users choose between rival AI models across iOS 27, due this fall, expanding beyond ChatGPT to include players like Google and Anthropic, Bloomberg reports. The difference this time: deeper integration, with outside models powering features like Siri, writing tools, and image generation across the system.

Currently, Apple’s voice assistant, Siri, gives users the ability to query ChatGPT, but doing so requires a clunky extra step and usage has been poor. Meanwhile, Apple’s own AI tools have fallen short. (Apple has decided to use Google’s Gemini to power Siri in the future.) It’s not clear users care which AI is under the hood — as long as it works.

Apple to Let Users Choose Rival AI Models Across Its iOS 27 Features

Rani Molla5/5/26

FactSet and S&P Global fall after Anthropic releases financial services agents

FactSet and S&P Global are trading lower after Anthropic unveiled a set of AI agents meant to automate financial services work. Both stocks also sold off earlier this year after Anthropic’s Claude introduced financial research tools.

The 10 agents handle tasks like earnings analysis, market research, financial modeling, and auditing — tasks that mirror how analysts use FactSet and S&P Global’s data and research platforms.