$500,000,000 Spent on one AI vendor. In thirty days. By accident.
The CEO of an un-named client of Anthropic right now
An AI consultant told Axios that one of their clients burned half a billion dollars on Claude in a single month. Not over a year. Thirty days. The reason was almost stupid in its simplicity: nobody had capped what employees could spend. Thousands of people got handed an open tab, and a chunk of that money went to autonomous agents grinding through tasks a human could do in seconds. Some of it, reportedly, went to people asking a frontier model to tell them the weather.
Turns out, The AI did not fail. The buyer did, along with the fleet of corporate consultants leading the company on AI strategy. We are in a new frontier, we need to build our own playbook, instead of looking up an answer from Mkinsey, PWC or Deloitte.
"So you're a software company"
"That's right Dave"
"and you spent $500 million on AI tokens in a month?"
THIS ISN'T ONE COMPANY. IT'S A PATTERN.
Uber · burned through its 2026 AI budget by April. The CTO admitted the Claude Code budget was already gone. The COO called it a head-exploding moment and started asking the quiet question: are we spending on tokens what we should be spending on people?
Microsoft · pulled most internal Claude Code licenses. Analysts called it the clearest enterprise-scale AI spending pullback of the year.
Amazon · killed its internal AI usage leaderboard. Employees were gaming it to look productive. They even gave it a name: tokenmaxxing.
Starbucks · scrapped its AI inventory tool after nine months. It kept miscounting milk and syrup and slowing baristas down. A frontier of technology, defeated by a bottle on a shelf.
This is bigger for humanity than Tiger winning the Masters this year
Sit with the Uber one for a second. The company that built the gig economy, the one that turned labor into an on-demand line item, is now the loudest voice in the room arguing that maybe we are spending too much on machines and not enough on humans. I am not sure how I feel about that. But when the inventor of the gig economy is the one sticking up for people, the pricing has clearly gone sideways.
WHY I BUILT A WAY TO MEASURE AI ON PRICE AND PERFORMANCE
I have spent my whole career on the buy side of expensive tools. As a dealership operator, I would never have run a store without knowing my exact cost per sale. You track it daily or you go broke. Yet here is an entire industry spending fortunes on AI with no idea what a single dollar actually buys them.
So we built the instrument. The Freddy Compute Index takes every major model from every major lab and answers the only question a CEO should care about: what do I get for one dollar?
It ranks every model by intelligence per dollar, shows you which one actually wins each job, and tracks the whole thing against the live cost of compute, energy, and Bitcoin. It is honest about what is measured in real time and what is not. No pay-to-rank. The chart works for you, not for the labs.
Use the best tool for the best task
HOW TO ACTUALLY USE IT (CEO OR SOLO)
Price the job, not the logo. The most expensive model is rarely the smartest buy. A mid-tier frontier model can deliver most of the capability of the top one at a fraction of the cost. The board shows you that gap in one number.
Do not rewind a generation. Drop a tier. Older flagships are often the same price as the new one, just weaker. Real savings come from a smaller, faster model for routine work, not last year's premium model.
Match the model to the task. One model leads coding, another leads agent work, another leads factual recall, and image and video are a different world entirely. Stop forcing one model to do everything.
Meter everything before you scale it. The $500M company did not have a model problem. It had a thermostat problem. Caps, role-based access, and a dashboard would have saved it half a billion dollars.
Personal rule: pass the weather test. If a human can do it in five seconds, do not send it to a frontier model. Save the expensive intelligence for the work that actually moves money.
Two of the biggest IPOs in history are lining up, and the rulebook is being rewritten to let them in
I got to spend time with the CEO of Hedgeye at the Indy500
SpaceX, now combined with xAI and trading toward a roughly $1.75 trillion target, is moving even faster, with a listing expected within weeks under the ticker SPCX.
The quieter story is what's being dismantled to make room for them. A widely shared Hedgeye post put it bluntly: index providers are waiving profitability requirements and slashing the seasoning window so passive money has to buy SPCX near its IPO price. The post compresses the details, but the bones are real. Nasdaq's new Fast Entry rule lets a megacap join the Nasdaq 100 just 15 trading days after IPO, down from three months, and dropped its float minimum. FTSE Russell cut its post-IPO window to five days. And S&P Dow Jones has proposed waiving the four-quarters-of-profitability test entirely for megacaps, the exact rule, born from the 2002 dot-com cleanup, that kept Tesla out of the index until late 2020. The S&P change is still a proposal, not yet final, but the direction is set.
Why it matters: more than $30 trillion is benchmarked to these indexes, and when a stock gets added, the funds tracking them have to buy, on schedule, regardless of price. Analysts peg the forced buying at $15B to $30B in conservative scenarios and north of $200B in aggressive ones, aimed at a company floating only 3 to 5% of its shares. SpaceX lost over $4 billion last quarter. It would be the least profitable company ever fast-tracked into the blue chips.
Which is where the Amazon comparison earns its keep. For years Amazon carried a massive market cap on razor-thin or negative GAAP earnings, and the bears called it a bubble every quarter. The thing they missed was the cash. Amazon was a cash-flow beast wearing an unprofitable costume, reinvesting every dollar before it could hit net income. The bet on SpaceX is the same wager: that the losses are a choice, not a sickness, and that the cash engine underneath justifies the price. The difference this time is that you may end up owning it either way, through a retirement account that never asked you.
SpaceX also tore up the other half of the playbook. A normal IPO locks insiders out of selling for about 180 days. SpaceX's prospectus lets investors sell 20% right after its first post-IPO earnings, with another 10% unlocked if the stock runs 30% above its offer price. Musk locked himself out of every early-release provision.
I drive a Rav 4
Freddy Media would like to share our best practices with you
Here are some products we find useful, and would encourage you implement in your company.
Go High Level- CRM and Marketing tool, it CRUSHES Hubspot and is cheaper. They are a day one Freddy Media client and now our business runs on it. This link gets you a 14 day free trial on us.
Check out RFW Training so that your company can learn how to do content for your business. Use Code “Freddy” for 15% your first 3 months.
Freddy Media would love to make your payment processing easier for your car dealership. We have partnered with DealerWorks to create a fee recapture solution compliant in all 50 states and Canada. Click Here to begin saving thousands per month on credit card fees