- Google is new Gemma 3 The models, as reported by Deirdre Bosa de CNBC, effectively work on a single chip – H100 from Nvidia or TPU de Google – Overpacing Deepseek’s R1 (34 H100S) and Meta’s Llama 3 (16), highlighting a profitable edge in the AI inference phase.
- The announcement underlines the advantage of TPU of Google, contesting the domination of Nvidia as Hyperscalers like Amazon (Trainium) and Meta, as well as startups like the brains, push internal fleas, while the stock of Nvidia, down 15% per year, increased by almost 6% today.
- With the IA passing to inference, Google’s efficiency game – supported by the use of TPU of Anthropic and the massive expenditure of the data center like Stargate – raises questions on the margins of Nvidia, with its GTC next week pivot to its counter -back.
Google again makes waves in the AI world, as reported by Deirdre Bosa de CNBC, with its new Gemma 3 The family of open -source models which can operate effectively on a single chip – whether it is a GPU, such as the H100 of Nvidia, or the personalized TPU of Google – highlighting a profitable advantage over competitors like the R1 of Deepseek, which requires 34 h100s, or Meta’s Llama 3, requiring 16. lightened, a bosa. By the wider market, offering companies a cheaper way to draw on advanced technology. While the NVIDIA (NVDA) accumulates 7% today, but down 14% for the year, Google’s announcement doubles a flexion of its TPU advantage, signaling a change in the AI race of training models to perform them effectively, a phase called inference where competition is warmed up.
The technological landscape buzzes while major players like Google (Goog, Googl), Amazon (AMZN) and Meta (Meta) push their own AI chips to reduce costs and challenge Nvidia’s grip on the market, where its chips have long adjusted training and inference. Bosa stressed that Anthropic is based strongly on Google Cloud’s TPUs for inference, while Amazon’s trainium fleas and Meta’s internal design gain ground, and even startups like brains feed models for companies such as perplexity and mistral. The CEO of NVIDIA, Jensen Huang, insists that his ecosystem remains high level for an inference, but with hyperscalers pouring money in alternatives – think of hundreds of billions of projects like Stargate – economic economic, which raises questions on Nvidia’s skins margins, as the more effective options and at lower cost.
It should be noted that the latest version of Google does not concern technology – it is a strategic game in a crowded field where efficiency could redefine which directs the next chapter of the AI. Bosa pointed out that the NVIDIA GTC event next week will be a great moment, investors wishing to see how Nvidia thwarts this thrust in the inference phase in the middle of its annual drop of 14%, despite the 7% rebound today. Google’s ability to execute Gemma 3 on a single chip against 34 or Deepseek Meta is not only a statistic – it is a signal that personalized fleas like TPUs could shake the game, in particular as Optai and others run to build their own solutions. This shock of giants and arrivals shows that AI is no longer a question of power – it is an intelligent and affordable delivery, and the last google stage could push the market in a new direction.
Wallstreetpit does not provide investment advice. All rights reserved.