AI is a bubble.

That’s not our opinion. That’s the opinion of OpenAI CEO Sam Altman, who has been saying so since at least August. And the opinion of Jeff Bezos, who said AI is in an “industrial bubble” earlier this month. And it’s the opinion of 54% of the respondents in Bank of America’s October Global Fund Manager Survey, released last week, who cite an AI bubble as the biggest tail risk facing the global economy today.

The move from “denial” to “casual acceptance” comes just as the AI-powered bull market officially turns three years old, as of Monday last week (perhaps not-so-coincidentally, ChatGPT turns three next month). In other words, we may just be living in the most talked-about bubble in financial history. So why is everyone singing “Happy Birthday” to the largely AI-fueled bull run with such a happy face?

Chips Don’t Lie

There are about 800 billion reasons some see the AI market as a bubble. That’s the amount in dollars, roughly, that global corporations have put toward AI-related capital expenditures in 2023, 2024 and 2025 combined (and conservative forecasts see AI capex reaching $1 trillion per year by 2030). That titanic level of spending has been the primary driver of US GDP growth this year, according to Harvard economists, as well as the vast majority of gains on the S&P 500, according to Morgan Stanley analysts.

In nominal dollars, it’s the biggest capex boom in history. Venture capitalist Paul Kedrosky recently noted that, as a percentage of GDP, the AI infrastructure buildout is about on par with the telecom infrastructure buildout of the 1990s and could even approach the roughly 6% of GDP that the investment in railroads hit in the 1880s. The railroad buildout led to multiple bubbles and collapses of its own (most prominently, the Panic of 1893). The telecom buildout preceded the dot-com bubble. But both also led to an enduring industry that generated enormous amounts of wealth. That provides some context as to why tech leaders seem to be shrugging off the bubble talk, and perhaps some clues as to why they shouldn’t be:

• The vast majority of AI-related capex has gone toward building massive data centers for expensive AI-powering GPU servers, according to a recent McKinsey report. But unlike relatively long-lasting railway and internet infrastructure, the average data center GPU may have a lifespan as short as one to three years.
• It’s why AI capex isn’t slowing down anytime soon. It’s also why a Bain Capital report published last month argued that AI companies will need to generate $2 trillion in annual revenue by 2030 to be profitable; as The Wall Street Journal recently noted, that’s more than the combined 2024 revenue of Microsoft, Meta, Alphabet, Amazon, Apple and Nvidia, and five times more than the global subscription software market.

The WheAIt from the ChAIff: Most estimates leave the industry at least $1 trillion short of that mark. According to a Citigroup note earlier this year, AI revenue is expected to reach $43 billion in 2025 and grow to $780 billion by 2030. The AI version of Pets.com won’t make it that long.

Imagine you’re a CFO and, suddenly, you’re promoted to CEO. A dream come true for C-suiters who have measured the drapes on the office down the hall with just that in mind.

But in a capitalistic version of “The Monkey’s Paw,” you get the top job at the precise moment your company is caught in the vise grip of the US-China trade war, with just weeks to satisfy American officials or be cut off from goods produced in the world’s largest economy. This is the task before Stefan Tilger, the newly appointed interim CEO of the Netherlands-based semiconductor firm Nexperia.

Auto Production Disruption

You’ve likely heard of Nexperia by now. The chip manufacturer, headquartered in the Dutch city of Nijmegen, was acquired in 2019 by state-backed Chinese semiconductor firm Wingtech. Nexperia makes specialized, though relatively basic, chips for consumer electronics, industrial use and, most crucially, the auto industry. Last year, the US government put Wingtech on its “entity list,” the Commerce Department tally of foreign businesses viewed as national security concerns. To keep those worries from impacting Nexperia as well, plans were put in place to fence off the subsidiary’s European operations from its parent.

But court records released in Amsterdam last week show the US told the Netherlands in June that things were moving too slowly. Then, last month, the Commerce Department said it would extend the entity list designation to Nexperia, which would bar US firms from exporting American-made goods to the company. The Dutch government sprang into action, invoking a never-before-used Cold War-era law to seize control of Nexperia on September 30. Then came the squeeze: China’s Ministry of Commerce retaliated days later by banning the export of finished Nexperia products. Because most of the chips that its European operation manufactures are shipped to China to be finished and assembled as part of larger goods, this could prove disastrous for Nexperia, and the company found itself stuck in the middle of a US-China trade standoff. All the while, lobbying groups representing virtually all major US and European automakers warned of production slowdowns or stoppages as soon as next month if the situation isn’t resolved:

• On Friday, the Dutch Ministry of Economic Affairs said it was “in discussions about resolving this matter with the Chinese authorities,” while Nexperia said it’s negotiating with the US and China for relief. The Dutch government said its seizure of Nexperia was not related to the US, but rather its own concerns that the company could transfer technology to its Chinese parent.
• As well as national security and technology concerns, Nexperia will likely need to address alleged mismanagement to earn the confidence of Dutch and US officials: The Amsterdam court found Nexperia’s now-suspended CEO Zhang Xuezheng made excessive orders totalling $200 million from a supplier called WSS, which is another Wingtech subsidiary, in order to keep it afloat.

Change the Channel: One need only look across the North Sea to see Nexperia isn’t new to controversy. After the company bought an 86% stake in the UK’s largest chip plant in 2021, officials forced it to sell, citing concerns China could undermine Britain’s ability to manufacture semiconductors.

Crypto watchers got another taste of just how different digital and traditional finance are last week when stablecoin issuer Paxos accidentally minted $300 trillion of PayPal’s PYUSD stablecoin before detecting the error and destroying the funds.

That all happened within just 30 minutes, and yes, the word trillion is correct. For perspective, $300 trillion is almost three times the world’s total estimated GDP.

“This was an internal technical error. There is no security breach. Customer funds are safe,” Paxos said in a post on X following the event. The error was part of an internal transfer, Paxos explained, and it immediately identified the mistake and “burned” the excess PYUSD it had created.

Fat Fingers

With the use of stablecoins growing dramatically in recent years due to their low cost and ability to speed up transactions, particularly for cross-border payments, the issue sparked concerns for investors, especially those less knowledgeable about their mechanics. PYUSD is the world’s sixth-largest stablecoin with a current market cap of $2.7 billion, and PayPal says each token is fully backed by US dollar deposits, US treasuries and similar cash equivalents. Therefore, each token should always be redeemable on a 1:1 basis for US dollars.

Despite the massive headline number and the resulting anxiety, however:

• No funds in the Paxos incident were actually transferred to customer wallets or redeemed by anyone.
• Additionally, the error was visible and quickly identified on Etherscan, the public “block explorer” for the Ethereum blockchain that lists all transactions occurring on the network. Similar “fat finger” errors in the world of traditional finance might go undiscovered or get revealed only when they are reported to regulators:

“It’s more like a little bit of egg on the face for Paxos,” said Carlos Guzman, research analyst at crypto trading firm GSR. “But it was just an internal transfer, so the actual backing of the assets of the token were never at any risk.”

Indeed, there are only about $2.3 trillion US dollars in circulation in the whole world, so there’s no way $300 trillion in PYUSD could have been officially created in the first place.

Code Red: The incident, however, does highlight how digital asset issuers, and stablecoin issuers in particular, are able to both create and destroy billions and even trillions of dollars instantly, wielding only the power of code. And so better internal controls are likely needed to prevent incidents like this from occurring again in the future. In 2019, for example, Tether, which issues the world’s largest dollar-pegged stablecoin USDT, accidentally minted and then quickly destroyed $5 billion in USDT. While such cases reveal the unique perils and possibilities of crypto, they simultaneously underscore its similarities to traditional markets, which also come with unforeseen risks. In other words, caveat emptor.