"It's been over a year since DeepSeek caused a bit of a panic across the AI stocks, as its distillation techniques caused many to question whether the frontier innovators needed all that hardware firepower. As it turned out, DeepSeek R1 wasn't the disruptor that was meant to derail the AI trade for good. Fast forward over a year, and AI demand is hot, and the need for infrastructure has caused big tech to spend colossal sums."
"With a memory chip shortage and a potential supercycle brewing in AI infrastructure, it feels like all is forgotten about DeepSeek and the threat it posed to the American AI innovators. Of course, distillation, the technique DeepSeek leveraged, is a crafty technique when training a model. But, as it turned out, it didn't put a top to the "scaling" for the firms at the very frontier of AI."
DeepSeek's distillation techniques initially prompted fears that frontier AI firms might need far less hardware, but DeepSeek R1 did not derail the AI trade. AI demand remains strong and major technology firms are spending heavily on infrastructure. A memory chip shortage and a potential AI infrastructure supercycle could raise costs and make scaling pricier in the near term. Distillation did not stop the scaling imperative for leading AI firms, though future breakthroughs enabling much cheaper training could again disrupt investor expectations. Firms continue to scale using GPUs and custom ASICs while watching memory markets closely.
Read at 24/7 Wall St.
Unable to calculate read time
Collection
[
|
...
]