The 2008 subprime crisis caused the U.S. to rescue financial institutions deemed too big to fail. The cost was socialized on taxpayers. A lookalike crisis is brewing in Mississippi. Its cost may fall on Entergy’s residential customers.
The conventional explanation for the subprime crash is that home prices fell in 2006 and borrowers defaulted. It’s wrong. Home prices rose in 2006 — but not as fast as they had been rising. Speculators who had borrowed to flip houses couldn’t flip them in the softer market. They defaulted. The effect snowballed and triggered the crash a year later.
Michael Lewis told that story in The Big Short. Andrew Ross Sorkin’s sequel was Too Big to Fail. Mississippi’s story may be Too Big to Default.
A July 2 essay at Groundbreaker argues the AI data-center boom is a subprime lookalike. If it is, the Entergy-Amazon secret deal brokered by the Governor and the Mississippi Development Authority rests on a false premise: Amazon is too big to default.
Entergy’s uncontrolled expenses for Amazon may be socialized on residential customers. A study this spring found that’s already happening. The average Entergy Mississippi household already pays about $10.60 a month for the Amazon build-out — roughly $38 million by March, on track for $74 million by year-end.
Data centers are financed with debt based on long-term contracts with AI customers. Amazon’s AWS has the most data centers. Its two biggest AI customers are Anthropic (maker of Claude), contracted to pay Amazon more than $100 billion, and OpenAI (maker of ChatGPT), contracted for $38 billion with $100 billion optional. Amazon is also Anthropic’s biggest investor. If Anthropic falters, Amazon takes a double hit. If Anthropic defaults, Amazon’s data-center value falls. And its ability to pay for electricity falls too.
OpenAI and Anthropic both lose money. OpenAI loses roughly $5 billion a year on $13 billion in revenue. Anthropic losses are similar. Both stay in business by raising more money — the way home flippers did by refinancing before the 2007 reset. That’s getting harder for them to do. A year ago each new fundraising round was valued at roughly twice the last one. Now it’s closer to one-and-a-quarter times. The market is getting cautious.
The market still likes their high equity valuations. But those valuations are rising at a slower rate every quarter. That’s a concern. That’s what triggered the subprime bubble burst.
In 2024, the Mississippi Legislature passed Senate Bill 2001 for Amazon. That law sidelined the Public Service Commission, waived the four-percent cap on residential rate increases, and granted automatic approvals for whatever Entergy spent — with a guaranteed return. It put Entergy’s other customers on the hook for spending Entergy said was for them. Does Amazon have a take-or-pay obligation under Senate Bill 2001. If it doesn’t or can’t honor it, do Entergy’s residential customers pay more? How much wiggle room does Amazon have?
The AI world has changed dramatically since Senate Bill 2001 passed in January, 2024.
A ChatGPT-level query paid twenty dollars per million words then. It pays seven cents now — a 280-fold drop. There were no serious Chinese frontier AI models then. Today they do half the searches from the largest independent AI routing platform. They work cheap. Their low prices depress data center prices and values.
When SB 2001 passed, CoreWeave was a private little-known company. Today it is a $42 billion AWS competitor. It offers contract features AWS doesn’t. It just signed a $21 billion contract with Meta for data-center capacity through 2032 — a contract Amazon would probably have won in 2024.
When SB 2001 passed, data center developers were buying and developing. Now many are selling. The Wall Street Journal reported this week that they are lining up to sell tens of billions of their data center holdings. Smart money leaving the party while the music is still playing?
In 2024 it seems that Mississippi’s central planners planned for the future to be like 2024. It’s not. AI technology has changed. The market has changed. Amazon has changed. Its free cash flow decreased 95% in the first quarter this year as capital spending hit $44 billion. It’s borrowing like its AI customers. Borrowing is getting more expensive.
Goldman Sachs projects Amazon and its peers will issue more than $500 billion of new debt in 2026 to fund data-center buildouts. Their borrowing cost has increased. Oracle is one of Amazon’s data-center peers. Its debt default insurance has doubled to 2009 levels. Its investment-grade bonds trade like junk. Same thing happened before the subprime crash.
It’s déjà vu all over again. In 2007 the cost of insuring subprime bonds tripled while housing prices were still setting records. Wall Street’s largest bond investors bought protection against a collapse the equity market said couldn’t happen. Rating agencies said the debt was AAA investment quality. But it turned to junk. Investors who read credit signals and sold early were Big Short winners. Late sellers were losers.
Too Big to Fail is a metaphor about ignoring or misjudging credit risks. In 2024, the Governor and the Legislature decided Amazon was too big to fail. They picked a winner, wrote a secret deal into statute, and waived PSC rules that could have limited residential customers’ exposure. Hindsight is 20-20. You can’t think of everything. Central Planning Economic Developers didn’t. They didn’t have skin in the game either. They may have been blinded by jobs.
I’m not predicting Amazon’s default. But it’s possible. What’s Plan B if it happens? Can the state clean up the mess that will be? Can it do anything now?
One thing is don’t do another secret deal. There is a better way for Mississippi to play the data-center game and attract more investments. It does not involve central planners or secret deals. It does not saddle monopoly-utility customers with higher rates. It takes advantage of Mississippi’s competitive advantages — and lets the market work.
Stay tuned.
Kelley Williams, a Northsider, is chairman of Bigger Pie, a Jackson-based think tank promoting free markets and government efficiency.
