Nvidia expands its AI footprint, Groq’s Samsung AI chip orders may increase by 70% to 15,000 units.

```

Demand for AI inference chips is heating up, reshaping the order landscape for Samsung's wafer foundry.

On March 10, according to sources cited by Chosun Ilbo, AI chip startup Groq has requested Samsung Electronics to increase its AI chip production from about 9,000 wafers to 15,000 wafers, an increase of about 70%.

According to reports, Groq was "indirectly acquired" by Nvidia for about $20 billion last December. Nvidia did not directly take over management control, but stated it would cooperate through a non-exclusive technology license agreement. The industry generally believes that the move is intended to extend Nvidia's ecosystem into the inference domain, complementing its dominance in the AI training chip market.

Notably, Nvidia is expected to launch a chip based on Groq's design, specifically optimized for inference, at GTC 2026. The report cites sources saying this chip may use SRAM instead of the HBM commonly used in traditional AI chips, and this architecture choice is expected to significantly lower inference latency and power consumption.

Meanwhile, according to Korean tech media The Elec, Tesla has delayed the production plans for multiple wafer projects, forcing South Korean AI chip company DeepX to postpone its next-generation NPU mass production schedule by about six months.

These developments reflect the complex situation currently faced by Samsung's wafer foundry: on one hand, the rapid expansion of demand for AI inference chips brings new customers and incremental orders; on the other, major customer Tesla's changes in production tempo have already had a tangible impact on other clients sharing the same production line.

Groq expands production, Samsung foundry enters the inference chip market

According to reports citing industry insiders, AI chip startup Groq's production at Samsung Electronics is moving from sampling to mass production. Previously, Groq's production at Samsung mainly focused on sample chips used to assess applicability in AI inference scenarios, while this year is seen as the formal start of large-scale commercialization in its early phase.

Even though Groq's production orders at Samsung remain relatively limited in absolute scale, this order is expected to help Samsung's wafer foundry establish a foothold in the inference AI chip market.

Additionally, Korean inference AI chip startup HyperAccel’s processors are also fully produced by Samsung’s foundry using a 4nm process, further highlighting Samsung’s deployment in this niche market.

Tesla delays impact DeepX, 2nm production line faces scheduling conflicts

Samsung’s 2nm wafer foundry line is facing order disruptions from Tesla. According to The Elec citing sources, Tesla has delayed its multi-project wafer (MPW) production plans, directly affecting Korean AI chip company DeepX, which also relies on this production line.

MPW is a model that allows customers to share wafers to split the costs of prototype manufacturing. DeepX originally planned to start MPW production for its second-generation NPU chip DX-M2 in April this year and has now postponed it by about six months. The report points out that DX-M2 is the first externally-customized chip designed for Samsung’s 2nm process, and Tesla is also using this process for its AI6 chip production.

The specific reasons for Tesla’s MPW delay are yet to be disclosed. According to industry sources cited in the report, shifts in production schedules for autonomous vehicles and humanoid robots, as well as changes in supercomputer investment plans, could have contributed to the delay.

DeepX's clients reportedly include Samsung Electronics, Hyundai Motor, Intel, and the company recently delivered chips to Baidu.

Tesla may greatly expand production of AI6 chips, Samsung’s 2nm capacity draws attention

Despite recent order disruptions, cooperation between Tesla and Samsung continues to deepen. According to The Elec, Tesla has recently sent a procurement executive to Samsung Electronics to discuss the expansion of its 2nm AI6 chip production.

The original contract stipulated production of 16,000 wafers per month, but if the new agreement is reached, monthly production may increase to about 40,000 wafers, more than doubling.

This potential expansion plan suggests that Tesla’s long-term dependence on Samsung’s 2nm process may significantly intensify, which also means Samsung’s 2nm foundry capacity will face even greater scheduling pressure. How to coordinate capacity allocation among customers like Tesla and DeepX will become a key operational challenge for Samsung in the near future.

Risk Disclosure and DisclaimerThe market involves risks, and investments need to be made cautiously. This article does not constitute personal investment advice and does not take into account specific investment goals, financial situation, or needs of any particular user. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their specific circumstances. Investing accordingly is at your own risk. ```