Can Tesla's supercomputer chip Dojo 2 disrupt the AI Computing Power competition? Musk: Renting computing power in the future is not ruled out.

Tesla's self-developed supercomputer Dojo 2 is said to enter mass production by the end of 2025, with computing power rivaling or even surpassing Nvidia's B200, revealing a key turning point in the rewriting of AI infrastructure landscape. (Background summary: Musk drops a bomb! Tesla launches a new model "Model Y L", targeting six-seat family vehicles to hit the market this autumn) (Background information: Musk's AI girlfriend meme coin Ani's market capitalization surpasses $50 million, setting a new high, but the risqué dance in black stockings leads to calls for delisting) Tesla once again throws out a shockwave; while the outside world is busy digesting the new indicators of Nvidia's Blackwell B200, Musk has quietly laid the groundwork for the next round of computing power competition. The second-generation chip Dojo 2 developed by Tesla is reported to be mass-produced by the end of 2025, achieving 1000 TOPS of computing power and 36 TB/s bandwidth, surpassing Nvidia's strongest B200. This is not just a hardware upgrade; it is crucial for Tesla to keep FSD (Full Self-Driving) training capabilities in its own hands. Dojo 2 challenges Nvidia's dominance. According to the latest external information, Dojo 2's computing power is about 1000 TOPS, while B200 is about 900 TOPS. A report from EE Times China indicates that the bandwidth between Dojo 2 chips reaches 36 TB/s, which is 3.6 times that of NVLink 5. The larger the bandwidth and the faster the data transfer, the shorter the time required for model iterations. Tesla's internal testing further emphasizes: "The overall performance of Dojo 2 can meet the mass video training needs of FSD, and there is a possibility of opening up computing power to the outside in the future." This statement from Tesla not only demonstrates technical strength but also hints at a new business model taking shape that could sell computing power externally. TSMC's packaging technology is a key driver. The advancement of Dojo 2 is inseparable from TSMC. With the support of InFO-SoW, CoWoS, and SoIC packaging technologies, multiple chips can be directly soldered onto the same wafer, reducing data transmission latency and improving heat dissipation efficiency. TSPA Semiconductor analysis indicates that this technology allows for stable operation even at thousands of watts of high power consumption. TSMC also predicts that packaging capacity will double by the end of 2025, providing a safety net for Dojo 2's mass production. For Tesla, mastering a stable supply chain is paramount. Tesla's new business model vision. Starting from 2024, Tesla has invested $500 million in New York to establish a Dojo cluster, currently handling about 5% to 10% of the driving assistance training work. When Dojo 2 goes online, this proportion is bound to soar. Once the training speed is no longer limited by external chips, the advancement curve of FSD will accelerate. Musk does not rule out renting out Dojo's computing power, and electric vehicle manufacturers may instantly transform into AI infrastructure service providers, opening up a new model for revenue structure. The layout of Dojo 2 further illustrates that whoever achieves vertical integration first will hold the dominant power in AI innovation. Taiwan occupies an important role because of TSMC; Chinese car manufacturers and large tech firms are also forced to consider how to reduce "external dependence." Even more daunting, the next-generation Dojo 3 has already been named by Musk to debut in 2026, boasting 40 times the performance of Dojo 2. What will the global computing power look like then? It's very hard to say. Related reports: Musk drops a bomb! Tesla launches a new model "Model Y L", targeting six-seat family vehicles to hit the market this autumn. Tesla Robotaxi opens global registration: priority for trial rides ( teaching ), when will Taiwan achieve full autonomous driving? 〈Can Tesla's supercomputer chip Dojo 2 disrupt the AI computing power competition? Musk: Renting out computing power in the future is not ruled out〉 This article was first published in BlockTempo, the most influential blockchain news media.