Today’s premium newsletter is part 1 of my What If...We're In An AI Bubble? series, covering scenarios that could burst the bubble starting with token-based billing's destruction of most AI products, and the slow collapse of data center construction. wheresyoured.at/premium-what-i…
I’ve spent hundreds of thousands of words in the last year talking about the very and scary signs that we are an AI bubble, and the exuberance fueling it. Today is the first time in a long time where I’ve discussed the potential consequences in depth.
I feel like people are very cavalier when discussing the AI bubble, saying stuff like "we're in an OpenAI bubble," not realizing that the collapse of OpenAI would destroy AI investment, AI startups, and kill debt issuance for AI data centers. wheresyoured.at/premium-what-i…
GitHub Copilot moves to token based billing on June 1, and its users got a calculator showing their actual monthly token burn - some $39-a-month users have spent $1500 to $5800 in tokens a month.
AI companies currently subsidize their users $3 to $25 for every dollar of subscription. As a result, users are very sensitive to *any* price or rate limit changes, and have no idea of the actual costs of AI. Token-based billing will be lethal at scale. wheresyoured.at/premium-what-i…
The current API revenues of Anthropic and OpenAI are driven predominantly by AI companies subsidizing their users' compute by 300% to 1000%, obfuscating the actual costs. Once the industry moves away from that, API revenues could face a brutal collapse.
Meanwhile, those currently paying token rates - big enterprises - are blowing through their annual AI budgets in weeks. CFOs are already saying that they need to reduce budgets, as this token spend is experimental - and finance executives are worried.
Enterprise AI token spend is out of control, driven by do-nothing CEOs with AI psychosis. Sources say Stripe is spending $2.8m+ a month, and a major real estate platform $300k-$700k a month, killing margins.
It's experimental and deeply unsustainable.
One scenario that could burst the bubble is if data center construction doesn't keep pace with Anthropic or OpenAI's capacity demands. If the capacity crunch doesn't clear quickly, both Claude and ChatGPT will begin having serious stability issues. wheresyoured.at/premium-what-i…
If data center construction slows or stalls, neoclouds and hyperscalers will suffer. $748bn of Microsoft, Google and Amazon's backlog is entirely dependent on them building gigawatts of capacity they don't have and that isn't getting built with urgency. wheresyoured.at/premium-what-i…
Another question: What happens if hyperscalers have hundreds of thousands of GPUs sitting in warehouses, yet to be installed in data centers?
This isn't a hypothetical question. Google has nearly $100bn in yet-to-be-deployed capex, and Meta over $50bn.
I think that if data center construction doesn't speed up dramatically in the next 6 months, hyperscalers will see massive write-downs on aging Blackwell GPUs as Vera Rubin ships. Either way, depreciation charges are likely to skyrocket in the next year. wheresyoured.at/premium-what-i…
So what if data centers aren't getting built? On the supplier front, customers can (and do!) cancel orders, and if this happens, we’ll witness a repeat of the kind of beatdown that Cisco and other telecoms firms suffered during the dot com bubble. wheresyoured.at/premium-what-i…
The thing you should know is that small events can have outsized impacts in a bubble. Something as simple as a slowdown in demand caused by a shift to token-based billing will have massive repercussions throughout the supply chain.
This was the first of a two-part (or possibly three-part) "What If...We're In An AI Bubble" series. Here's what's up next.
My premium newsletter is $7 a month/$70 a year, and weighing in at 10k+ words a week, making it the best value in tech analysis.
Free newsletter: The ROI debate has come at worst time for Anthropic and OpenAI - two companies that can’t afford to slow down thanks to their $1.1 trillion in compute commitments and a data center buildout that needs a trillion in annual revenue by 2030. wheresyoured.at/ai-is-slowing-…
My appearance on Bloomberg last week ruffled some feathers because I don't have "skin in the game" (IE: money in the market). This is because AI boosters can't imagine somebody with a moral position on an industry run by grifters and sold with lies. wheresyoured.at/ai-is-slowing-…
The AI industry is anxious because it knows it made too many egregious promises. There are 190GW+ in planned AI data centers, and to justify their existence, they'll need $2 trillion+ in AI compute revenue by 2030.
Free newsletter: The dawn of token-based billing has shown that generative AI doesn’t have a return on investment. It's too unpredictable, too unreliable, you can't easily measure the cost of tasks, and organizations are already pulling back. wheresyoured.at/ai-doesnt-have…
Ever since the AI industry started its transition away from subscriptions to token-based billing, users have been hit with sticker shock. Once companies start paying the real cost of AI, they realize there's no measurable ROI to be found.
GitHub Copilot moved users to token-based billing on June 1 from a request-based system that allowed them to burn thousands of dollars for $39 a month.
Users are currently burning their entire monthly balance in a couple of prompts.
Free newsletter: The economics of AI don't make sense. Monthly subscriptions to AI services are scams that represent a product that is impossible for OpenAI and Anthropic to provide long term, and AI data centers are an economic disaster at scale. wheresyoured.at/ais-economics-…
Github Copilot, as I reported last week, is moving towards token-based billing, charging the actual cost of token burn — a move that alienated many of its users who built entire workflows around a request-based product that no longer exists. wheresyoured.at/ais-economics-…
This is the subprime AI crisis — and a crisis of deception. AI products tricked users by offering unsustainable monthly subscriptions built to separate the process of using AI from its real cost. Eventually everybody will go token-based. wheresyoured.at/ais-economics-…
Premium Newsletter: OpenAI needs to raise or make $852 billion through 2030 to pay for its compute costs, and must pay $75bn a year for Oracle's 7.1GW of Stargate data centers, or Oracle will collapse, destroying its share price and Ellison's empire. wheresyoured.at/how-openai-kil…
It’s been little over a year since Sam Altman, Masayoshi Son, and Larry Ellison stood shoulder-to-shoulder with Donald Trump to announce the "$500 billion" Stargate data center project. In truth, no entity was ever formed, and no money was invested. wheresyoured.at/how-openai-kil…
The "$500 Billion" Stargate data center project is really a $340 billion, 7.1GW Oracle-and-debt-funded construction nightmare. Outside of Abilene - which existed before the announcement - these data centers are mostly scaffolding and steel beams. wheresyoured.at/how-openai-kil…
$5 billion TO DATE? This heavily suggests that Anthropic? Over $10 billion in inference and training costs? This heavily suggests that almost every story about Anthropic’s revenue and costs they’ve leaked is wrong and that it’s been misleading the media.
Remember: reports said Anthropic made $4.5bn for all of 2025, and all of this guff about “$19bn in annualized revenue” no longer make sense, especially as that “exceeds $5 billion” line is from a lawsuit filed in *march*. This is a scandal!
The only thing that makes sense here is that Anthropic’s 2025 revenues were much lower than $4.5 billion. $5bn ALL TIME? Through march? Remember this is filed with a court. Suggests Anthropic is a much much smaller business than we believed
Premium: The AI Bubble is a time bomb, burdening hyperscalers with billions of new depreciation a quarter that will soon eat away profits, VCs with dead equity in fallen AI startups, and Oracle with hundreds of billions of unpayable leases. wheresyoured.at/premium-timebo…
A year ago, I wrote that Coreweave was a timebomb — a rickety tower of debt and bad deals that was primed to blow.
I could have made that argument about hyperscalers, NVIDIA and OpenAI. Everybody is waiting for somebody else to make AI a success.
The AI bubble is slowly turning cash-rich, margin-heavy hyperscalers into something far worse. Now these asset-light, high-margin software companies are becoming bloated, filled with tens of billions of dollars in GPUs and data centers. wheresyoured.at/premium-timebo…