SoftBank Surges 30% in 3 Days on GPT-6 Hype: Bubble or Breakout?
AI chip demand is pulling Asian tech stocks higher, but September OPEX history says watch your back

Ticker Ratings
If you blinked, you missed SoftBank's 30% three-day surge. The OpenAI and GPT-6 demand wave is doing what Masa Son has always dreamed of: making his AI bets look prescient instead of reckless. Bloomberg's Stock Movers segment flagged the move, noting volumes are elevated and options activity suggests traders are still targeting further upside. The catch? SoftBank is still well below its July highs, which means this is either a genuine re-rating or a very enthusiastic dead-cat bounce. You get to decide which one keeps you up at night.
The memory chip side of this trade looks cleaner. SK Hynix is up for a third consecutive session (off its intraday highs of 6%, settling nearer 2%), and Samsung added roughly 1%. Taiwan's Natech gained 2.7% and Winbond rose 4%, all riding the same GPT-6 memory demand thesis. The logic is straightforward: next-gen AI models eat memory bandwidth for breakfast, lunch, and dinner, and these are the companies cooking the meal.
Here's the wrinkle the bulls need to square up with. Japan's broader market is not celebrating. The Topix dropped 1% and the Nikkei finished in the red, weighed down by Bank of Japan rate hike expectations. Japanese nominal wages just posted a 4.7% annual gain in July, the fastest pace in nearly three decades, and markets are now pricing in back-to-back BOJ hikes in September and October. The yen hit a six-month high against the dollar as a result. A stronger yen is good for your Tokyo vacation, not great for export-heavy Japanese tech earnings.
And then there is the September elephant in the room. One trader on YouTube laid out the historical case: September OPEX has produced drops of 10% or greater in midterm years including 2014, 2018, and 2022. We are in a midterm year. Tech momentum has already been soft on a 60-day rolling basis since June highs. The setup looks great on individual names, but the macro calendar is waving a yellow flag.
The Nvidia angle matters here too. Its AI chip market share has slipped from roughly 80% to 60%, with custom silicon grabbing 40% of the market and projected to hit 50% by decade's end. The GPT-6 trade is real, but it is becoming a broader ecosystem bet rather than a one-ticker story. Memory and power infrastructure names may actually be the cleaner way to play it at this point, with less concentration risk and more room to run before the valuation conversation gets uncomfortable.
Thirty percent in three days is a party. The question is whether you arrived before the hosts started running out of ice.
BullApe's AI grades every pick against the S&P 500 - wins and misses published. See the track record →