Meta* is poised to significantly disrupt the AI market as it is likely planning to purchase AI accelerators for billions of dollars not from Nvidia, AMD, or even Intel. Meta is in talks with Google.
Google, as it is known, develops its own TPU chips for AI and recently unveiled a new generation, Ironwood, with a performance of 42.5 ExaFLOPS per module (each module houses 9216 processors).
The Information resource reports that the companies are already negotiating the supply of chips worth billions of dollars, starting in 2027. Moreover, Meta may rent the power of Google’s chips as early as next year.
Unlike Nvidia and AMD accelerators, Google’s chips are primarily not GPUs but ASICs.
Although it is already the seventh generation, Google has never sold its TPUs to other companies before, and such changes could significantly impact the market.
Recently, the AI landscape has been dominated by Nvidia’s graphics processing units, which are well-known for their versatility and widespread adoption across various sectors due to their robust GPU architecture. Google’s Ironwood TPU, centered around Application-Specific Integrated Circuit (ASIC) architecture, offers a streamlined approach specifically tailored for AI. This specialization could yield better performance in dedicated AI tasks but might lack the adaptability of GPU solutions, a factor that could play into Meta’s decision for focused, high-performance AI applications.
Meta’s strategic move to consider Google’s TPUs highlights a shift towards diversifying technological partnerships, potentially decentralizing the dependence on Nvidia. This choice also suggests a stronger collaboration within the tech giant circle to encourage innovation beyond existing industry standards.
* Meta (Facebook and Instagram) recognized as extremist in Russia and banned
