On the scheduling chart of semiconductor manufacturing, a "last-minute detour" by a major client often means a "passive withdrawal" for other competitors. According to a report on March 10, due to the ripple effect of Tesla's production plan, the South Korean AI chip leader DeepX had to delay its mass production plan originally scheduled for the second quarter of 2026 by half a year. Its highly anticipated next-generation NPU chip, DX-M2, is now expected to begin quality testing only after the third quarter of 2026.

The first "external customer" of Samsung's 2nm process runs into Tesla's "queue-jumping"

DeepX's DX-M2 had once attracted attention as the first external customer chip for Samsung Electronics's 2nm process. However, another of Samsung's top players - Tesla - is also using the 2nm process to develop its next-generation AI chip AI6.

Due to recent adjustments in Tesla's supercomputer investment plan and the production schedule for its humanoid robot chip, the delay of its multi-project wafer (MPW) has caused a butterfly effect, forcing the production schedule of DeepX behind it to be postponed overall.

"Noise-reduction" power in data centers: DX-M2 shows strong performance

Despite the delay, the parameters of DeepX's DX-M2 chip remain impressive. As an accelerator specifically designed for generative AI data centers, it has the following core advantages:

  • Powerful inference: It can handle complex models with up to 100 billion parameters.

  • Ultra-efficient energy use: It offers 80 TOPS of computing power but consumes a maximum of only 5 watts.

  • Top-tier standards: It fully supports the low-power LPDDR5X memory standard.

Pain before monetization: Orders still need to wait for mass production

For DeepX, the delay means its revenue contribution will be pushed back again. The company has not yet received formal orders for DX-M2, and according to industry practice, customers usually place formal orders after the mass production starts in 2026 and the sample tests are passed.

However, DeepX's technical capabilities have already been endorsed by industry giants. Its customer list includes big names such as Hyundai Motor, Intel, and Samsung Electronics. In addition, DeepX recently successfully supplied 40,000 DX-M1 chips and modules to Baidu, and this transaction is expected to account for 5% to 24% of its revenue forecast this year.

Amid the extreme scarcity of global AI chip capacity, even a dark horse like DeepX has to accept the test of time when facing the capacity competition from a "money-guzzling giant" like Tesla.