The latest developments in AI and economics
A March 2025 scan
How is AI impacting the economy and society?
Today I’d like to share results from my ongoing AI horizon scanning work. Here we’ll focus on the economic dimension, from the bubble debate to new business structures and labor implications.
(If you’re new to this newsletter, some of its issues scan the horizon for current and recent developments which look likely to have an influence on the future. These are not speculative reports, but instead are grounded in evidence and documentation. Each issue scans for one particular domain, such as politics, education, technology, or economics. I’ve organized things this way because there’s a *lot* going on in the AI world, especially in terms of how we respond to it. Each domain impacts the others, and all of us.
Other issues are more speculative or analytical.)
Are we in an AI bubble?
Answers are divided on this, as they have been for some time. On the “this looks like a bubble” side we can find evidence and analyses. OpenAI admitted that its ChatGPT Pro service is losing, not making money. On the stock market front, a bunch of AI stocks have declined over the past month. Nvidia’s price has really dropped, despite impressive earnings.
Ed Zitron has another long, passionate post arguing that generative AI is insanely overvalued and about to collapse. Specifically, he finds evidence that Microsoft is backing away from data center expansion. Given the huge needs of large language models for memory and processing, and also given that MS has been investing heavily in multiple AI projects, this story could point to the tide turning, or the bubble starting to pop.
On the other hand, there are metrics and stories showing generative AI’s value to still be massive and rising. On the customer demand side, OpenAI reported 400 million active users in February, up from 300 million in December, along with 2 million paying ones. Investment is still roaring in, as the Financial Times estimates more than $320 billion on deck for this year. Investors poured into $3.5 billion into Anthropic, based on a valuation of more than $60 billion. One upcoming South Korean project would cost $45 billion and also use 3 gigawatts of power, far more than the under 1 the biggest ones today tend to use. A Canadian investor is prepping to send $20 billion to build up French data centers.
At a smaller scale, young AI firms are winning investment as well. Some investors put $175 million into Sierra, “founded by ex-Salesforce co-CEO Bret Taylor and former Google executive Clay Bavor,” based on a valuation of $4.5 billion. A Californian AI military startup won $200 million to get working on its $5 billion valuation.
In China, the publication of a group of new AI tools powered the nation’s stock market to new growth:
Alibaba Group Holding Ltd. paced the gains on Thursday after unveiling its latest open-sourced AI model. The platform marked a big leap over the previous version, using just a fraction of the data DeepSeek’s R1 employs.
The flurry of AI announcements also included Tencent’s new open-source video model Hunyuan and a similar product from short-video platform Kuaishou Technology. Earlier this week, Manus AI launched what it called a general AI agent, or a bot that can perform tasks, saying its model is performing better than OpenAI’s DeepResearch on some fronts.
In response, Goldman Sachs thought a bunch of emerging markets would grow. And n contrast, the American financial markers took hits, leading one wag to say the AI excitement has moved from the USA to China.
Emerging AI business models and practices
An important economic feature of the AI revolution is the steady growth of new ways to put LLMs to work. For one example, last month Anthropic and car share company Lyft announced they would collaborate on new and expanded services. The official statement mentions Lyft already used Claude (backed by Amazon’s Bedrock service) as a customer service chatbot. Now they’ll try out more stuff, using AI to “test… new products and capabilities to ensure they align with the needs of riders and drivers, enabling faster integration of proven solutions.” Also, “Anthropic will provide specialized training to strengthen Lyft's engineering teams, accelerating their ability to build innovative AI-powered features.”
In another AI partnership, Google and the Associated Press will expand their collaboration. “AP will now deliver a feed of real-time information to help further enhance the usefulness of results displayed in the Gemini app.”
In an, ah, stronger alliance mood, Grammarly purchased productivity site Coda. The goal of the combined enterprise? “The acquisition will help turn Grammarly’s AI assistant into an “AI productivity platform” thanks to the addition of Coda’s AI tools and products, the company says. The deal will give Grammarly customers access to new features, such as generative AI chat and a productivity suite, to help them work more efficiently.”
On the international side, Biden-era chip restrictions to China still exist, but Chinese firms are circumventing them by purchasing through third parties. “Since 2022, Washington has imposed export controls to curb China’s access to semiconductors for training and powering state-of-the-art AI, but an underground network of brokers has sprung up to get around the controls.”
Meanwhile, OpenAI continues to try becoming more of a for-profit enterprise, as opposed to the sort-of non-profit yet also commercial entity it has become. Elon Musk sued to stop the transformation, but a judge blocked his effort.
Implications for the world of work
A Harvard Business School paper studies the impact of AI adoption on software development and finds that rather than coders coding less, they reduce project management work: “when software developers leverage AI more, they reallocate their efforts towards technical coding activities and away from auxiliary project management activities that involve social interactions with other developers." This is especially true for the highest-performing workers. One reason for this is that programmers are able to work more independently: “generative AI tools reduce (or even eliminate) much of the cognitive and communicative friction inherent in distributed work, enabling workers to tackle complex tasks autonomously.” AI-assisted workers also do more exploratory work.
Along complementary lines, IT sector unemployment rose in 2025’s first quarter, above that of the rest of the workforce. First, “‘Jobs are being eliminated within the IT function which are routine and mundane, such as reporting, clerical administration,’ [Victor Janulaitis, chief executive of Janco Associates] said.” Further, “As they start looking at AI, they’re also looking at reducing the number of programmers, systems designers, hoping that AI is going to be able to provide them some value and have a good rate of return.” So: “Rather than hiring new workers for tasks that can be more easily automated, some businesses are letting AI take on that work—and reaping potential savings.”
One San Francisco AI company trolled that city with ads urging viewers not to hire humans any longer, like this one:
The firm summed up the ad campaign’s results like so:
When we decided to launch our own outdoor campaign at Artisan, we knew we needed something different. Something that would stand out. Something… provocative. What we didn't know was that our controversial approach would lead to 10s of millions of impressions, 1000s of death threats, 100s of articles and our biggest growth months ever.
Was it worth it? Yes.
As an example of this, the CEO of Salesforce says AI makes his engineers more productive, therefore they don’t need to hire as many of them. “Our engineering productivity has increased by 30%, which means we don’t currently require more software engineers. AI has fundamentally changed how we work.”
For another example, Bloomberg Intelligence reports (via Yahoo, nonpaywalled) that Wall Street is looking to cut jobs, but also to modify the remaining ones. So we see certain jobs targeted for reduction:
Back office, middle office and operations are likely to be most at risk, according to Tomasz Noetzel, the BI senior analyst who wrote the report. Customer services could see changes as bots manage client functions, while know-your-customer duties would also be vulnerable. “Any jobs involving routine, repetitive tasks are at risk,” he said.
BI estimates Wall Street cutting up to 200,000 people over the next three to five years. Then there are more general comments about “improving” and “augmenting” work.
Another study estimated AI as having a serious impact on musical creatives by 2028:
In an unchanged regulatory framework, creators will not benefit from the Gen AI revolution, but will actually suffer losses on two fronts:
It is estimated that by 2028, 24% of music creators' and 21% of audiovisual creators’ revenues will be at risk, resulting in a cumulative loss of €22 billion for creators in these sectors over 5 years.
■ The loss of revenues due to the unauthorised use of their works by Gen AI models without remuneration
■ The “cannibalisation” or replacement of their traditional revenue streams due to the substitution effect of AI-generated outputs, competing against human-made works
[emphases in original]
In particular, “[t]he potential impact will be strong on digital collections (up to 30% cannibalisation), TV & radio and Background (c. 22% of cannibalisation).” The market for AI-generated sounds will grow exponentially, according to the authors.
Note the leading line of that first quote, concerning regulation. Calls for government action to protect workers from AI are rising. There are also continued signs of popular resistance to the technology, as when Polish listeners drove Radio Krakow to end the use of AI announcers and reporters.
That’s where we’ll pause for now.
Summing up, the bubble model is still in play as the AI industry grows while doubts persist. More AI business models are appearing, notably new partnerships and combinations. AI is showing signs of impacting the labor market, both in reducing jobs as well as transforming them.
One thought: if the AI investment peaks then declines as a bubble (as opposed to a milder market correction), how far would the ripples go? Might China be able to maintain its AI sector while the US and Europe see theirs collapse?
Coming up: I have notes for scans in technology, culture, and education, plus notes for other posts on utopia and dystopia.