Tech Newsround – Going Bananas

AI Ecosystem: Gemini – No. 1 Banana

  • The battle for the consumer AI Ecosystem is just beginning, and Google has taken some of the momentum away from OpenAI with the hilariously named Nano Banana image function within Gemini, demonstrating that there is still everything to play for.
  • Gemini is currently the No. 1 free app on Apple App Store in the US, pushing ChatGPT into second place, and has risen to 2nd place in the UK and Canada.
  • Key to this success is not the Gemini chatbot but the image-generating and image-function known as Nano Banana.
  • Since August 26th, when Nano Banana launched, the Gemini app has added 23m new users and 500m images have been edited in that period.
  • Computational photography is something that Google has been good at for more than 10 years, and whereas OpenAI, Anthropic, et al have only been at this for a couple of years.
  • Hence, it does not surprise me that Gemini is better at image manipulation than OpenAI, although when it comes to image creation from a text prompt, I would give the edge to OpenAI at the moment.
  • The idea here is that if Google can generate traction within the Gemini app for image manipulation, then its chances of enticing users to use Gemini for other functions improve meaningfully.
  • This is the key to taking share from ChatGPT, which is still far and away the leader in terms of user numbers and usage.
  • This would not be the first time that the early leader ends up losing out, and so in the same way that Chrome overturned Internet Explorer, Google may end up winning out.
  • This would be a disaster for OpenAI, whose valuation and outlook depend on becoming the largest consumer AI ecosystem, which is why I expect that it will be looking at these numbers with consternation.

Memory – HBM4: Ready to go

  • The development of the next generation of high-bandwidth memory is complete, with all of the major players having shipped samples, meaning it is now down to who Nvidia qualifies for its upcoming AI systems based on Rubin and Rubin Ultra.
  • SK Hynix has announced that its development of HBM4 (next generation) is now complete and that it is now ready to begin mass production of its product.
  • SK Hynix has already shipped samples of its 12-layer product, and it is clearly very keen to maintain the lead that it won in HBM3 as a result of Samsung making a horrible mess of its HBM3 and failing to qualify.
  • Samsung shipped samples in July, while Micron shipped its samples one month before in June.
  • SK Hynix has set a high bar with performance improved by 65%+ and its power efficiency improved by 40%.
  • Samsung is rumoured to have also improved power efficiency by 40% over HBM3, but I would expect much, much more given that power efficiency was one of the reasons why its HBM3 failed to qualify at Nvidia.
  • Hence, the improvement figures it puts up should be much higher than anyone else’s, given the last generation was so bad.
  • With the three main players pretty much ready to go, it all now comes down to qualification at Nvidia, which is likely to take place in Q1 2026.
  • Here, I expect that everyone will qualify with Nvidia, which will mean better prices for Nvidia and also a chance for Samsung to regain its crown as the king of memory.
  • This is the catalyst I am looking for to complete the rally in Samsung and to return the shares at least to the levels that they enjoyed prior to the HMB3 debacle.
  • This would put the shares at somewhere between KRW90,000 and KRW100,000, at which point I will begin thinking about taking profits on the London-listed GDR that I hold.

RICHARD WINDSOR

Richard is founder, owner of research company, Radio Free Mobile. He has 16 years of experience working in sell side equity research. During his 11 year tenure at Nomura Securities, he focused on the equity coverage of the Global Technology sector.