1. Stock prices soar with the advent of AI
The S&P 500, which tracks America’s 500 largest companies, has tumbled and risen nearly 80% over the past five years. The rally was led by the “Magnificent Seven” of big tech stocks involved in the AI boom: Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla.
Investor concentration in technology is unprecedented, with 41 AI stocks now accounting for nearly half of the S&P 500’s market capitalization, said Jim Bianco of Bianco Research.
Neil Wilson, an analyst at investment platform Saxo UK, says the prospect of a 1970s-style inflation shock, generally high valuations for tech companies and the possibility of a freeze in private credit markets does not bode well for stocks.
“The entire market has become one giant AI conglomeration,” he says. “The danger is a repeat of the dot-com bubble: a big crash and years of lost profits. Depending on how you look at it, valuations aren’t as strong as they were back then, but this looks like an incredibly dangerous market.”
2. Spending is increasing at an alarming rate
From data centers to chips, spending on AI is on track to jump from $765 billion this year to $1.6 trillion by 2031, according to Goldman Sachs. Investment banks acknowledge that commitments of this size can be problematic. What happens if there is a delay in the data center?
“Given the size of the capital committed, even a small delay in execution will require increased scrutiny of the demand assumptions used to underwrite these investments,” Goldman analysts wrote, but added that if spending plans go ahead without a hitch, it could spark a new wave of AI demand. Nevertheless, this spending shows how much of the world’s financial resources and expectations of return are being poured into AI.
3. Businesses and consumers are rapidly adopting AI
Despite mixed reports about the benefits, the majority of companies have started using AI, increasing from 33% in 2023 to nearly 80% today, according to consultancy McKinsey. According to data from Sensor Tower, OpenAI’s ChatGPT now has 1 billion monthly active users and high usage among the general public, a record for any app.
The question now for AI developers is how to monetize this vast public and private customer base. Companies need to be able to demonstrate that AI improves outcomes and reduces costs to justify the claims. That means using it to build an entire workflow (a business term for executing an entire task from start to finish). It has a long way to go.
4. Claude is following ChatGPT
Anthropic began establishing itself with OpenAI late last year. The company’s Claude Code tool caught on primarily among software developers in the San Francisco area, and then spread far more widely. Claude Code represents a shift in the use of large-scale language models, the core technology behind chatbots, ushering in the transition to autonomous AI agents that perform tasks without human intervention, and enabling non-technical people to write software to perform a wide range of tasks.
Although OpenAI still has a much larger overall user base, data from Kentik, an internet analytics company that tracks usage across a number of internet service providers in the US, shows Anthropic is quickly catching up. Claude’s user traffic grew significantly faster than ChatGPT and Google’s Gemini from January to April, and spiked after the Pentagon declared it a supply chain risk in March. Kentik predicts that this growth rate could overtake ChatGPT by the summer. This is another reason why we think Anthropic has an easier path to an IPO than its rivals.
5. The cost of using AI is getting higher and higher
Every time an AI chatbot or agent issues a response, that response is measured in “tokens,” or building blocks of language such as words, punctuation marks, or syllables. (For example, OpenAI says that the phrase “100% of the shots you don’t take, you miss” is worth 11 tokens.) It also uses tokens to measure input, such as prompts you type into ChatGPT.
These costs vary by model. OpenAI is priced at $5 million per input token and $30 million per output token (response to a prompt) for GPT-5.5.
The problem for subscribers is that the cost of tokens has increased significantly, even as companies around the world are encouraging their employees to get serious about “tokenmaxx,” or the use of AI. The problem for AI companies is that they aren’t getting paid enough yet.
The promise inherent in the use of AI is that the money companies spend using these tools will be returned to them in the form of increased productivity, a measure of economic efficiency. Increased productivity means getting more out of each employee. If this trade-off does not occur, the assumptions underlying AI evaluation and policy will be compromised.
“Costs are becoming completely out of control,” says Liam Betsworth, founder of British AI startup Pendra. He said the software developers around him are using agents to code, starting with the cheapest subscriptions and quickly moving to the most expensive packages. They’re not alone – the news site Axios recently reported on an anonymous company that spent $500 million a month licensing Claude Code.
6. Data center construction may not be able to keep up with demand
Data center construction represents the central nervous system of AI products, so increased development and use of AI tools will require more capacity. Failure to do so will result in a compute crunch and higher costs for AI companies and users.
The scale of this sector’s data center ambitions is vast and seems unlikely. Bloomberg estimates that 23GW of capacity is under construction worldwide in 2025 (capacity is measured in electricity, as it constrains the amount of computing a site can run).
US real estate company JLL predicts that 100GW will be added between 2026 and 2030. This is double the current capacity the company estimates is equivalent to 1,200 data centers. JLL said its estimates take into account speculative projects that have not yet started construction.
Where the funding and energy supplies to realize this prediction will come from is an open question. Cecilia Ricup, an associate professor at University College London, says many projects around the world are based on political commitments to expand the grid and provide electricity. But the government may not have the means to make that happen.
“Has the government calculated whether such an expansion is feasible? Does it have the funds to do it? Has it taken into account the environmental damage that would entail?” she asks.
7. What AI models can do is expanding rapidly.
According to METR, a research organization that measures AI capabilities, the capabilities of AI models have improved dramatically since 2023.
METR measurements are based on the ability of an AI model to perform a coding task, quantified by the time it takes a human to perform the coding task. According to this metric, the power of AI models doubles every four months. For example, Anthropic’s Claude Mythos model has been calculated to have a 50% success rate for tasks that would take a human expert between 8 hours and 2 days.
However, so far there has been no commensurate impact on employment. Anthropic’s March report includes research showing that while AI could theoretically perform many jobs, from computing to legal work, it has not yet been able to do so with any significant force.
Buk Klein Teesling, an academic at King’s College London and an expert on the impact of AI on work, says there are bottlenecks to introducing AI in the workplace. For example, how much of the work of CEOs and senior managers can be safely delegated to bots? Can legally sensitive tasks be performed by non-humans? Still, he says, change is coming.
“We are still in the early stages of the AI revolution. There are many people performing tasks that AI can perform. The amount of change we will see will be enormous.”
8. Data centers support US GDP
Despite U.S. government job cuts and mass layoffs across a wide range of industries under the Donald Trump administration, U.S. GDP continues to grow, reaching 2.1% in 2025 and 1.6% in the first quarter of 2026, according to the U.S. Bureau of Economic Analysis. But Harvard economists calculate that without the data center boom, these numbers would be much lower. In other words, “investment in information processing equipment and software” accounted for 92% of US GDP growth in the first half of 2025.
This means the data center and AI booms account for a disproportionate share of U.S. growth, and are a large part of why the world’s largest economy still looks healthy despite major headwinds. A reduction in this spending could have economic and even political consequences.
