AI Companies Are Trying to Hide a Staggering Amount of Debt
TL;DR Highlight
Nikkei: five big tech firms hold $1.65T in off-balance-sheet debt, more than reported debt
Who Should Read
LLM app developer worried about AI infrastructure over-investment and the price and supply stability of LLM APIs
Core Mechanics
- $1.65T of debt sits off the books
- Nikkei Asia says Alphabet, Microsoft, Amazon, Meta and Oracle hold about $1.65T in debt not on balance sheets.
- That exceeds the $1.35T the same five officially reported for the latest quarter.
- Meta alone has about $420B
- Per Nikkei, Meta's off-balance-sheet debt is around $420B.
- The article uses it to show how precarious AI spending has become.
- The structure resembles Enron
- The article says they use special purpose vehicles and legally distinct subsidiaries.
- Enron collapsed in 2001 after hiding debt behind shell companies.
- Raising money via new shares too
- Nikkei reports they are also selling new shares.
- The article mentions possible equity dilution and lower investor confidence.
- Accounting expert warns
Community Reactions
- Disclosure is legitimate, a rebuttal
- Some said the debt is disclosed in financial statement notes, so a "hide" headline is overblown.
- A commenter quoted Nikkei saying the practice is legal but may make risks hard for retail investors to see.
- Normal relative to scale, one view
- A commenter argued $420B of debt against $200B revenue and $100B EBITDA would be normal in other industries.
- It only looks large because tech was used to huge cash piles.
- Big tech conflated with AI startups
- A commenter said the article slides from cash-rich megacorps to bubble talk about startups like OpenAI and Anthropic.
- Another said only Oracle is in real danger.
- Depreciation may be the bigger issue
- A commenter said stretching useful lives of datacenters and GPUs could overstate profits by tens of percent.
- They noted it only pulls earnings forward.
How to Apply
- Build a LLM vendor dependency checklist
- Spend 30 minutes today listing your model API vendors and possible replacements.
- Note switch time if prices rise or service fails.
- Move model calls behind an abstraction
- Put model names and endpoints in config so swapping vendors is a one-line change.
- Register an open-weight model as a fallback.
- Regression-test key prompts on a backup model
- Pick 20 key prompts and run them on two candidate models with identical inputs.
- Define pass criteria beforehand.
Try It
# Paste into a Claude or ChatGPT chat (the source is financial reporting with no code, so this is a simple checklist prompt)
Our service uses LLM APIs as follows.
- Providers and models: {{provider and model}}
- Monthly call volume and cost: {{volume and cost}}
- Core features: {{core features}}
Create a vendor lock-in risk checklist for price hikes or service outages.
1) List of single points of failure
2) Code and prompts to change when switching to a backup model
3) Three mitigation tasks doable in one week
Next: implement one of the three tasks right away.Terminology
Related Papers
Mercury 3: Diffusion at scale
Scaling diffusion LLM 40x gives Mercury 3: 4–8x faster than speed-tier models, similar quality
Timestral: Mistral's time-series foundation models
A 209M time-series model ranks first on three benchmarks and runs on CPU.
DeepSeek V4.1 Flash 1017 (Max)
DeepSeek V4.1 Flash 1017 Max scores 45 on the Intelligence Index at 211 tok/s and $0.30 input.
If you don't run your own LLM proxy, how do you know whats going on?
Running your own LLM proxy shows which keys, models, and cache misses drive your usage.
DeepSeek: DeepSeek v4.1 Flash 1017
DeepSeek V4.1 Flash 1017: $0.05/M input, 1M context, built for agents
DeepSeek-v4.1-Flash-1017
DeepSeek V4.1 Flash minor update lifts coding and agent benchmarks while cutting run time and output length
Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware