Elon Musk’s Terafab Ambitions Spark Backlash From Korean Semiconductor Engineer Over Samsung Dealings
A public dispute surrounding Elon Musk, Samsung, Tesla, SpaceX, and the future of AI chip production has intensified after a South Korean semiconductor engineer criticized Musk’s recent comments about memory pricing and the companies involved in his massive Terafab initiative.
The controversy began after Musk praised Micron for what he described as reasonable future memory pricing. In the same comment, he suggested that other suppliers had been “extortionate,” a remark widely interpreted as a veiled criticism of Samsung. That statement quickly drew attention across the semiconductor industry, especially as Samsung is already connected to Tesla’s next-generation AI chip plans.
The situation became even more sensitive after Musk indicated that TSMC could be involved in the Terafab project. While he ruled out the idea that the Taiwanese chipmaker would own or operate the factory, discussions between TSMC and SpaceX appear to be underway. During a recent earnings call, TSMC also made a pointed comment about not leaving opportunities for competitors, further fueling speculation about its role in Musk’s semiconductor roadmap.
Amid this backdrop, a South Korean semiconductor engineer using the handle @TTHUEONE on X strongly pushed back against Musk’s apparent criticism of Samsung. The engineer claimed that Tesla turned to Samsung for its AI5 and AI6 chips after TSMC was unwilling to meet certain demands. According to the engineer, Tesla allegedly pushed for unusually tough contract terms, requested access to internal fab information, demanded audits, and pressured Samsung on pricing.
The engineer also accused Musk’s semiconductor project of recruiting heavily from Samsung’s local workforce in Texas, particularly around Samsung’s advanced chip facility in Taylor. This claim comes at a time when Terafab is aggressively hiring talent as it prepares for what could become one of the most ambitious chipmaking projects in the United States.
Terafab is envisioned as a major semiconductor venture tied to Tesla and SpaceX, with technological support from Intel. The long-term goal is enormous: to produce up to 1 terawatt, or 1,000 gigawatts, of AI compute capacity per year. If achieved, that would place the project among the most aggressive attempts to reshape AI chip manufacturing anywhere in the world.
For now, Terafab remains in its early stages. The project is expected to begin with a semiconductor complex in Grimes County, Texas, with planned investment reportedly around $16.8 billion. Over time, the facility could expand dramatically, eventually reaching a footprint of around 100 million square feet.
Musk’s broader goal appears to be reducing the complexity and delays that dominate modern semiconductor production. Today, chip manufacturing depends on a highly fragmented network of specialized companies handling different stages of the process, from design and mask generation to lithography, packaging, and testing. Terafab aims to bring many of these processes under one roof.
If successful, this approach could shorten the chip development cycle from the typical six to nine months down to just weeks. That would be a major advantage for Tesla, SpaceX, xAI, and other Musk-led companies that depend heavily on fast-moving AI hardware development.
A key part of this strategy is Tesla’s AI5 chip, which is designed to support advanced artificial intelligence workloads. The chip reportedly uses a half-reticle physical size and includes a large central compute die surrounded by 12 DRAM modules. It is expected to be manufactured using a 2nm process at both TSMC and Samsung’s Taylor facility in Texas.
Performance expectations for the AI5 chip are ambitious. Reports suggest it could deliver between 2,000 and 2,500 TOPS, placing it in the same conversation as some of the most powerful AI accelerators currently available. A single AI5 chip is expected to compete with NVIDIA’s H100-class hardware, while a dual-chip configuration could challenge newer Blackwell-level systems.
Memory capacity has reportedly changed during development. The chip was originally expected to include 144 GB of LPDDR5 memory, but Musk later suggested a reduction, first hinting at 72 GB before settling closer to 96 GB. Trial production is already said to be underway at Samsung’s Taylor facility, with mass production expected in 2027.
Musk has previously claimed that the AI5 chip could deliver roughly three times the performance of a comparable NVIDIA chip while costing only about 10 percent as much. If that target proves realistic, it would give Tesla a significant cost advantage in AI hardware, particularly for autonomous driving, robotics, data centers, and large-scale AI training or inference.
Beyond AI5, Tesla’s AI6 and AI6.5 chips are also expected to rely on Samsung and TSMC as manufacturing partners. These future chips are believed to include large amounts of SRAM, a memory type that can offer much faster bandwidth than conventional DRAM for certain workloads. Musk has previously said that a major portion of the AI compute accelerator design is dedicated to SRAM, which could dramatically improve effective memory bandwidth for calculations that remain inside the cache.
However, the growing tension around memory pricing, contract negotiations, and talent recruitment highlights the difficult balance between Musk’s aggressive ambitions and the realities of global semiconductor manufacturing. Samsung, TSMC, Micron, and Intel all play crucial roles in an ecosystem where advanced chips require enormous investment, specialized expertise, and long-term cooperation.
The engineer’s criticism also underscores a wider concern within the semiconductor industry: as major AI players race to secure supply, build custom chips, and reduce dependence on existing suppliers, relationships between customers and manufacturers are becoming increasingly strained. Companies want faster production, lower prices, deeper technical access, and more control, while chipmakers must protect their margins, intellectual property, and workforce.
For Tesla and SpaceX, Terafab could become a powerful way to gain more control over AI chip production. For Samsung, involvement in Tesla’s AI chip roadmap could strengthen its position in advanced foundry manufacturing, especially in the United States. But the latest exchange suggests that cooperation between the two sides may be more complicated behind the scenes than public announcements imply.
As AI hardware becomes one of the most important battlegrounds in technology, disputes over pricing, production capacity, talent, and control are likely to become even more common. Musk’s Terafab vision may promise a faster, cheaper, and more integrated future for AI chipmaking, but getting there will require navigating some of the most powerful and competitive forces in the global semiconductor industry.
