December 9, 2025
GPUs sizzle, SPVs dazzle?
After the Bubble
Pop or not? Internet splits over AI crash, burning chips, and “voodoo” money
TLDR: A veteran warns the AI boom could burst, with pricey, fragile chips and off-books financing leaving little useful infrastructure. Commenters split: some insist there's no bubble and AI kills jobs, others smell 2008-style debt tricks or say at least the buildings and power upgrades will remain.
A fiery thread erupted after a veteran technologist warned the AI boom is a bubble set to burst—this time leaving us with scorched graphics chips and massive power bills instead of bargain internet pipes. One camp came in swinging: “There is no AI bubble,” argued one user, saying AI is already replacing or boosting millions of jobs, not vanishing like a fad. The doubters fired back that the money magic—SPVs (special companies built to keep debt off the books)—smells like 2008 all over again, and dropped the classic [citation needed] meme on rosy profit claims.
The hottest clash: will anything useful be left if the air whooshes out? The original piece says “not much,” because GPUs can burn out and the power costs never stop. But pushback was sharp: one commenter snapped that we’ll still have the buildings, permits, and bigger power lines—aka real-world upgrades you can reuse. Meanwhile, a side-quest debate erupted over how fast companies can write off hardware for taxes, turning into accountants vs. engineers while everyone argued over who’s cooking the books and who’s cooking GPUs.
For context, folks linked to a chat with an economist and VC Krugman & Kedrosky, Matt Levine’s funny take on finance tricks Coffee Pods, and the FT’s bond deep-dive. The vibe: anxious, spicy, and very online.
Key Points
- •The article argues the GenAI surge is a bubble and focuses on GPU fragility/costs and SPV-based financing as core risks.
- •Depreciation is explained, with management discretion affecting financials; a cloud vendor’s shift from 3- to 4-year depreciation affected its share price.
- •LLM training runs GPUs at extreme duty cycles, leading to higher failure rates; indicative data points come from Meta and Google.
- •AI infrastructure may not leave cheap, reusable assets post-bubble due to GPU burnout and high ongoing electricity costs.
- •Big Tech is using SPVs to fund massive data center builds, keeping obligations off balance sheets; referenced analyses by Matt Levine and the Financial Times explain these structures.