Elon Musk's ambitious "Terafab" plan is facing a sober assessment from Wall Street. According to analysis, Bernstein Research, in a recent report utilizing detailed quantitative calculations, indicates that achieving Musk's envisioned goal of producing 1 terawatt of computing power annually would require capital expenditures reaching $5 to $13 trillion. The necessary wafer capacity would be equivalent to the entire existing global semiconductor production capacity, making this challenge "more difficult than landing on Mars."

Musk announced the launch of the "Terafab" project last weekend, planning to expand human computing power production to 1 terawatt per year. This is approximately 50 times the current global computing power supply (about 20 gigawatts). The plan involves first building an advanced wafer fab in Austin, Texas, encompassing the entire production chain for compute chips, logic chips, memory, packaging, and mask production. A team led by Bernstein analyst Stacy A. Rasgon promptly issued a report providing a systematic quantitative assessment of the plan's feasibility.

Bernstein pointed out that if investors genuinely believe Musk can achieve this goal, semiconductor equipment stocks would be the most direct beneficiaries. However, for the time being, the plan's substantive impact on the industry is "likely not significant, remaining more at the level of hype."

The capital expenditure estimates are astronomical. Bernstein used the current computing architecture (Nvidia racks) as a baseline to roughly estimate the wafer requirements for three core chip types: GPUs, HBM memory, and CPUs. The report shows that producing 1 terawatt of computing power annually would require a starting volume of 7 million to 18 million 300mm wafers per month, with HBM memory demand being the dominant factor.

Converting this demand based on a factory unit with a capacity of 50,000 wafers per month, the requirement equates to building 140 to 360 new factories of this scale. Assuming a capital expenditure of $35 billion per factory, the total investment would reach $5 to $13 trillion. Bernstein explicitly noted in the report that these calculations are "very rough estimates" and do not yet include other semiconductor categories such as networking, optical, analog, and power management chips.

The capacity requirement is akin to recreating the entire global semiconductor industry. Bernstein's calculations reveal the staggering scale of the plan in terms of production capacity. The report states that the wafer capacity needed for 1 terawatt of computing power is roughly equivalent to the total existing global semiconductor capacity (approximately 16 million 300mm equivalent wafers per month).

Focusing solely on "relevant" semiconductors—namely memory plus advanced logic wafers at 4 nanometers and below—the current global installed capacity is about 5 million 300mm equivalent wafers per month. The capacity required for the 1-terawatt target is several times this number. In other words, Musk's plan essentially requires a several-fold expansion of the existing advanced process capacity, highlighting the immense difficulty.

Regarding industry impact, Bernstein believes the near-term substantive effect of the Terafab plan on the semiconductor industry is limited but offers several noteworthy directions from an investment logic perspective. The report indicates that if the market believes Musk can advance this plan, semiconductor equipment stocks would be the most direct investment targets.

On the question of whether Musk building his own chips would threaten existing chip manufacturers, Bernstein holds a relatively optimistic view. They suggest that in an environment with such strong computing power demand, any participant would face potential far exceeding their own capacity to fulfill it, and memory manufacturers are in a similar position.

At the business model level, Bernstein notes that Terafab integrates logic, memory, mask manufacturing, chip design, and packaging into one entity, essentially constituting a "super IDM" model. This model has been proven to be far less efficient than the specialized division of labor system of "foundry + fabless + specialized memory IDM." Therefore, the report concludes that Terafab currently poses a limited threat to foundries like TSMC.

Bernstein does not outright dismiss Musk but acknowledges the challenges are "extremely arduous." Despite the staggering calculations, Bernstein does not completely rule out the possibility of Musk's success. The report states that Musk has previously accomplished things considered impossible by outsiders, and "we wouldn't bet against him lightly." However, the report also clearly states that "a true Terafab feels like a stretch to us, particularly under the current computing paradigm."

Bernstein suggests two possible alternative paths: First, if unable to proceed independently, Musk might seek partnerships with existing chip manufacturers. Second, Musk might have a "more surprising" technological path to break the current paradigm, but the report admits, "we don't know what that would be."