AI BriefingAI GPUs Likely Have a Lifespan Longer Than 3 Years
Key point
Some claim that AI GPUs only last around 3 years, but real-world cases suggest they are likely to remain in use for much longer.
Details
Recently, among those concerned about the sustainability of AI infrastructure, a claim has emerged that Inference GPUs have a lifespan of only up to 3 years. This claim leads to the logic that if the AI bubble bursts, there won't be enough funds to replace aging GPUs, causing the cost of AI services to surge.
However, the basis for this '3-year theory' rests on a speculative statement made by an anonymous expert through the interview platform Tegus. Even if it comes from an expert's statement, it's a stretch to consider this figure as reflecting the actual failure rates of real data centers.
Actual cases show the opposite picture.
- Google stated that it is still using 8-year-old TPUs at 100% utilization.
- AWS's CEO mentioned that not a single Nvidia A100 server has been retired yet.
- In academic GPU cluster cases, failure rates have also been reported to remain below 20% over 6 years.
In conclusion, the outlook that AI GPUs have very short lifespans and that infrastructure will rapidly become obsolete may differ from actual operational data.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
AWS Introduces Enhanced Custom Event Buses for Organization-Wide Event Sharing