Bubble or not?
The evolving economics of AI, fall 2024 edition
Looking into the future of AI requires tracking the technology’s economics. After all, much of generative AI comes from businesses, each of whom must make financial sense of the stuff. So today I’ll update you all on what I’ve seen in my scanning. The key points: we haven’t had a bubble crisis (yet) and new business arrangements are emerging.
NB: I’m talking economics but not labor today. I have so much on the business side and so much on the workers’ side that I’m splitting them in two.
(If you’re new to my Substack, welcome! This issue is an example of my environmental scanning research. Each one focuses on one particular slice of the world of AI and the future: politics, education, economics, culture, etc.)
Double toil and trouble
In August and September there was a flurry of commentary calling out an imminent AI bubble. Companies missing analysts’ expectations, changes to stock value, and above all a lot of vibe detection suggested the AI boom was falling apart. (for example) (for example) (a little earlier) (the famous Goldman letter) A bunch of generative AI opponents predicted on social media that the bubble was popping. The arguments were similar: AI expensive to operate, revenues not pouring in at profitable levels, hype turning to skepticism. CNBC observed pithily: “spending is too high, and returns are too low.” Alphabet and Meta CEOs each warned investors and the world that AI earnings might not arrive soon.
And yet… clearly, there was no collapse. NVIDIA, the major AI chip maker, seems to be doing all right, despite disappointing the market by making a lot of money, just not the astonishing pile they hoped for:
Microsoft is also trundling along, although below July’s peak:
Alphabet, owner of Google, is similar, having enjoyed a July peak and not yet returned to it:
So we experienced a partial market correction in August, but the big firms did not see their values collapse. Nvidia and Alphabet are enjoying some growth over the past month. Taiwan’s world-leading chip-making company, TSMC, saw immense profits in 2024’s third quarter. Microsoft is experiencing a downturn, but its stock value is still very high. And now Goldman Sachs says no, it’s not a bubble. Some user numbers are growing. OpenAI keeps winning new investments and even a giant credit line.
Moreover, some costs are dropping. Check out this curve for token fees, which have been falling steeply:
So could the AI market collapse? Of course. Any market might, and we’re still dealing with unsolved problems for generative AI - namely, there isn’t a good business model so far (yes, Brent, I’m still working on my video response to your rebuttal.). For one example, here’s a recent New York Times analysis:
OpenAI’s monthly revenue hit $300 million in August, up 1,700 percent since the beginning of 2023, and the company expects about $3.7 billion in annual sales this year, according to financial documents reviewed by The New York Times. OpenAI estimates that its revenue will balloon to $11.6 billion next year.
But it expects to lose roughly $5 billion this year after paying for costs related to running its services and other expenses like employee salaries and office rent, according to an analysis by a financial professional who has also reviewed the documents. Those numbers do not include paying out equity-based compensation to employees, among several large expenses not fully explained in the documents.
I think what we’re seeing now is the financial sector disciplining the technologists, as Carlotta Perez described, and as I keep telling you. Banks, venture capital, hedge funds, et al have contributed lavishly and now want to see returns. They could break some projects, which might end up looking like a bubble popping. But today I lean more towards the continued financial restructuring mode, and less the pop.
In short, lots of economic scrambling going on. Speaking of which, the tech firms and other players are busily working on that very topic:
New business model developments
What kind of changes and new projects are appearing?
Switching from non-profit to for-profit - OpenAI is trying to become a fully for-profit entity. Now, this is an outlier, because OpenAI’s structure is very strange. But there might be others.
Partnerships - these are booming, as AI firms openly license stuff from, or provide services to, content owners. OpenAI licensed Condé Nast content. OpenAI also partnered with Hearst: “This collaboration spans over 20 magazine brands and 40+ newspapers, enriching ChatGPT’s 200 million weekly users with a vast array of lifestyle content,” says the press release. Alibaba partnered with NVidia to use LLMs for autonomous cars. Runway partnered with Lionsgate to make AI-generated videos.
A different kind of partnership saw SAG-AFTRA agree to work with AI startup Narrativ to make human voices available. This is more of an agreement to make agreements:
The firm’s new agreement with SAG-AFTRA promises to give the union’s 160,000 members the opportunity to add themselves to a database that connects voice talent to advertisers. The individual members will have the ability to negotiate fees for the use of their voice on a project by project basis, so long as the fee isn’t lower than SAG-AFTRA’s minimum per its most recent commercials contract with advertisers.
And yet OpenAI’s partnership with Microsoft seems to be running into problems. Microsoft hired another AI startup’s CEO, while OpenAI wants to work with other companies and pay less to Redmond.
There is also this… interesting clause in the OpenAI-MS relationship:
The contract contains a clause that says that if OpenAI builds artificial general intelligence, or A.G.I. — roughly speaking, a machine that matches the power of the human brain — Microsoft loses access to OpenAI’s technologies.
So, extra incentive for OpenAI to really beat the Turing Test!
Higher fees - OpenAI is apparently considering setting subscription rates as high as $2,000 per month. Right now, “[r]oughly 10 million ChatGPT users pay the company a $20 monthly fee,” according to the New York Times, and that doesn’t seem to be enough. Perhaps we should expect a much wider spread of prices and services. Meanwhile, digital multitool Canva raised prices by 300% (!), justifying the rise by saying it was due to adding AI.
Ads - Perplexity is starting to sell ads, which would presumably run within its interface. I would expect some ad agencies to use AI to generate such intra-AI ad content. We could imagine some established business models playing out here, such as cheaper services with ads, while more expensive versions are ad free.
Outsourcing - Amazon set aside its in-house tech in order to outsource its upcoming Alexa upgrade to Anthropic’s Claude. Let’s see how many technology firms choose to not use in-house AI in favor of best of breed externals.
Chinese competition - leading Chinese AI firms and projects recently showed off video and other efforts, drawing close to American offerings. And perhaps racing ahead:
“My American friends still can’t use Sora,” Mr. Qu said. “But we already have better solutions here.”
China has, of course, immense efforts invested in AI.
Around these efforts, consultancies seem to be growing, as all kinds of people and organizations (companies, governments, schools) feel that they need help grappling with AI. This summer the New York Times profiled some consultants. I see no evidence that that secondary business is slowing down.
What does all of this mean for us?
Practically, we have to prepare for ever growing ranges of services, prices, and models. Some of us, content creators and/or owners, may also partner with AI businesses. I’m struck by the growth in such partnerships between AI firms and content providers. We might see more of this, even to nonprofits (libraries, colleges, museums).
We should also prepare for some services to change or shut down, as business models keep developing under financial guidance.
…that’s all from the economics scanner for now. More coming up!
(thanks to Todd Bryant and Tom Lairson for good links and discussion)