The American car manufacturer Tesla (NASDAQ: TSLA) has unveiled on Thursday a custom chip that is set to train Artificial-Intelligence networks in its data centers. The scientific breakthrough was showcased at the automaker’s live-streamed AI Day demonstrating the efforts of the company in the vertical integration field.

The chip called D1 is a part of Tesla’s Dojo supercomputer system. The chip relies on a 7-nanometer manufacturing process with 362 teraflops of processing power. According to Ganesh Venkataramanan, senior director of Autopilot hardware, Tesla placed 25 D1 chips on a “training tile” and of the tiles, 120 came together across several server cabinets, making an exaflop of power, said, Venkataraman. 

 “We are assembling our first cabinets pretty soon,” said Venkataramanan, who previously worked at chipmaker AMD. The engineer added that the company’s chip will be the fastest AI-training computer. Models can rely on the chips to recognize a variety of items from video feeds that Tesla’s cameras recorded. The process requires extensive computing work. 

 “We should have Dojo operational next year,” CEO Elon Musk said.

Tesla's D1 chip abilities

The D1 chip work followed Tesla’s car production using AI chips inside its vehicles. The chips that the company manufactured 2 years ago help the car’s on-board software to decide quickly in response to whatever is facing it on the road.  

Tesla is now offering a “Full Self-Driving Capability” add-on for new vehicles. The $10,000 package enables Tesla cars to move into parking spots, navigate on highways and even change lanes. According to the company’s website, the package will also enable Tesla buyers to enjoy the automatic steer on city streets. Nonetheless, Tesla has already promised that this will come in 2019. 

Tesla’s website says Full-Self Driving elements “require active driver supervision and do not make the vehicle autonomous.” Earlier this year Tesla introduced a $199 monthly subscription for Tesla owners who wish to access Full-Self Driving.