Tesla emerges as surprising rival to AMD and Nvidia in quest to grab next-gen HBM4 memory for AI and supercomputers
Posted by:
John Harrison
Tesla is set to advance its AI capabilities with the introduction of HBM4 chips. The company is currently in discussions with Samsung and SK Hynix, two leading South Korean memory chip manufacturers, to obtain samples of their latest HBM4 chips for evaluation. These chips are expected to power Tesla’s customized Dojo supercomputer, a key component in the company’s AI initiatives, particularly in self-driving vehicle technology.
The introduction of HBM4 chips marks a significant upgrade from the older HBM2e chips, promising enhanced speed, power efficiency, and overall performance. Both Samsung and SK Hynix are racing to develop prototypes of HBM4 chips for Tesla, as well as other tech giants like Microsoft and Google. SK Hynix touts its HBM4 chip as delivering 1.4 times the bandwidth of its predecessor while consuming 30% less power, making it ideal for training massive AI models on Tesla’s Dojo supercomputer.
Samsung and SK Hynix are expected to fast-track their HBM4 development schedules, with SK Hynix aiming to release the chips to consumers by late 2025. Samsung, leveraging its advanced 4-nanometer (nm) foundry process, seeks to gain a competitive advantage in the global HBM market. This rivalry is poised to shape the future landscape of high-bandwidth memory technology, impacting various industries beyond automotive and AI.
Recent Posts1
Watch out, your office phone could be hijacked into a Mirai botnet
February 5, 2025
