Meritz Securities reports that OpenAI's new GPT-6 Astra expands the scope of AI automation by enabling the model to handle complex tasks lasting up to 24 hours. The model achieved a 99.9% score on the ARC-AGI-3 benchmark, a leap from the 7.8% recorded by the previous Sol model, which allows it to infer rules and execute results directly in unfamiliar environments. This technological shift triggered a rebound in semiconductor stocks on September 4, with the DRAM ETF rising 6.6% while the broader US market fell on rising interest rates.

The increase in workload offsets declining token prices, potentially ending a pair trade of long software and short semiconductors that began in early July. Meritz advises against exiting core AI memory positions despite a conservative overall market outlook through mid-October influenced by Korean won appreciation. The analyst suggests the launch marks a structural growth point for AI infrastructure as GPU rental prices rise in response to the model's agent capabilities.

Sign in to suggest edits

Key sources

  1. SOURCE@jukan05“the barrier to entry for AI agents has been lowered, meaning more work can be handed over”x.com
  2. SOURCE@jensenhuang“400K GPUs coming online next”x.com
  3. SOURCE@gdb“we're now moving into the AGI era (whether you view it as this model, the last one, or the next one)”x.com
  4. SUPPORT@thetranscript_“first run that we’ve trained on more than 100,000 GPUs”x.com
  5. SUPPORT@stocksavvyshay“compute required to push the frontier is still scaling”x.com
Markdown