Google's TurboQuant Can Now Cut RAM Requirements by up to 6x

Google introduced a new breakthrough in the field of artificial intelligence that has the potential to change the global technology industry landscape. Through a compression algorithm called TurboQuant, the technology giant claims to be able to cut the memory (RAM) requirements for AI models up to six times.

This innovation comes amid a surge in RAM needs due to the rapid development of large language models such as ChatGPT and Gemini. So far, advanced AI models require very large memory capacity, triggering a rise in global memory chip prices and creating bottlenecks in various sectors, from data centers to consumer laptops.

TurboQuant: "Digital Courier" that Can Carry 6x More

TurboQuant technology works with a method called quantization, which simplifies data without significantly sacrificing accuracy. One of its main targets is the "Key-Value cache", an important component in AI that stores the context of a conversation so that the model does not have to recalculate from scratch - but so far it is known to be very wasteful of RAM.

With this new approach, Google claims the efficiency achieved is close to the maximum theoretical limit. A simple analogy: it's like putting six times more clothes in a suitcase without adding weight - saving space, but still functional.

This announcement immediately shook the market. Shares of major chip manufacturers such as Samsung Electronics, SK Hynix, and Micron Technology weakened, as investors feared that RAM demand could drop drastically if the technology was widely adopted.

Even so, a number of analysts believe that the global RAM crisis has not really ended. Increased efficiency can actually trigger the development of more complex and "resource-hungry" AI, so that memory requirements remain high in the long term.

For now, TurboQuant is still in the research stage and is not ready for mass use. Large-scale implementation is expected to take time, especially since many chip supply contracts have been locked by major technology companies.

One thing is clear, however: if AI efficiency can be drastically improved through software, the pressure on the global RAM supply could potentially ease faster than expected. The AI world may not be frugal, but at least it is starting to learn to "live more efficiently."