[News] Samsung Groq Orders May Jump ~70% to 15K Wafers; Tesla Reportedly Delays Other Foundry Clients

Groq, an AI chip startup that NVIDIA “acquired indirectly,” has reportedly requested increased production from Samsung Electronics Foundry. According to Chosun Biz, sources say Groq recently decided to boost the output of its AI chips produced by Samsung from roughly 9,000 wafers to about 15,000 wafers.

As the report highlights, last year’s production was largely limited to sample chips used to evaluate their suitability for AI inference. This year, however, the company is believed to be entering the early stage of large-scale commercial production. Although the production volume Groq has entrusted to Samsung Electronics remains relatively modest, the report notes that the order could help Samsung’s foundry gain a foothold in the inference AI chip market.

Groq was reportedly “acquired indirectly” by NVIDIA in December last year for about $20 billion. Rather than taking direct management control, NVIDIA said it would collaborate with Groq through a non-exclusive technology licensing agreement. Industry observers widely believe NVIDIA’s indirect acquisition of Groq is intended to expand its ecosystem into the inference segment, complementing its dominance in AI training chips, the report notes.

The report also adds that, besides Groq, processors developed by HyperAccel, a Korean inference AI chip startup, are also entirely produced by Samsung Electronics Foundry using its 4nm process. Market interest in inference AI chips is also rising as NVIDIA is expected to unveil an inference-optimized chip based on Groq’s design at GTC 2026. The chip may use SRAM instead of HBM commonly used in traditional AI chips.

Tesla Delays Reportedly Disrupt Samsung Foundry’s AI Chip Schedule

While Samsung Foundry may benefit from rising production orders from Groq, another report points to delays affecting other projects. According to The Elec, sources say Tesla has postponed multi-project wafer (MPW) production, a move that could delay the mass production of a next-generation NPU from Korean AI chip firm DeepX, as both rely on the same Samsung foundry line. Customers of DeepX reportedly include Samsung Electronics (Exynos 2100), Hyundai, and Intel (some i5 processors). The company has also recently delivered chips to Baidu.

Sources say MPW production for DeepX’s second-generation NPU DX-M2, originally scheduled for April, has been delayed by about six months. MPW allows semiconductor prototypes from multiple customers to be produced on a single wafer, enabling companies to share wafer costs and reduce prototype manufacturing expenses.

As the report highlights, sources say the MPW postponement is linked to Tesla. DeepX’s DX-M2 is the first external customer chip designed for Samsung’s 2nm process, which Tesla is also using to produce its AI6 AI chip. The specific reason for Tesla’s MPW delay has not been disclosed, the report notes. Industry observers estimate that autonomous vehicle and humanoid robot production timelines, along with supercomputer investment schedules, may have collectively contributed.

Meanwhile, the report adds that Tesla recently dispatched a procurement executive to Samsung Electronics to discuss expanding production of its 2nm AI6 chip. The original contract covered 16,000 wafers per month, which could rise to about 40,000 wafers per month if the deal proceeds.

Read more

• [News] Delay at Samsung’s Taylor Reportedly Slips Mass Production to 2027, Raising Concerns for Tesla
• [News] Tesla Reportedly Weighs Doubling AI6 Output, May Add Samsung Orders Ahead of Executive Talks

(Photo credit: Samsung)