Reduce the Dojo project, core talent flows to DensityAI, Tesla's AI strategy faces a major adjustment?
Tesla has been reported to disband its Dojo supercomputer team, with its leaders and core members joining a mysterious AI startup called DensityAI, founded by former Tesla executives. This dramatic personnel change and project contraction not only raises questions about the sustainability of its in-house chip strategy but also suggests that the company may be undergoing a profound reshaping of its AI strategy through "lightening the load" and strengthening external collaborations
Tesla's internal AI landscape is undergoing a dramatic shake-up, as its once high-profile Dojo supercomputer project seems to be coming to an end.
On August 12, according to news from Chasing Wind Trading Desk, the analyst team at Morgan Stanley pointed out in their latest research report that Tesla has disbanded the Dojo supercomputer team, which is not merely a simple project contraction, but a proactive adjustment of AI strategic focus centered on enhancing capital efficiency. This may be a pragmatic adjustment aimed at reducing costs and increasing efficiency, as the market generally believes it consumed a significant amount of capital expenditure and operating expenses.
At the same time, this marks Tesla's shift from a high-cost self-developed hardware model to a greater reliance on external partners such as NVIDIA and AMD, in order to concentrate valuable resources on areas closer to application implementation, such as robot commercialization and vehicle "distributed inference cloud." This move may also provide a rationale for deeper collaboration between Tesla and Musk's other company, xAI.
Previously, Bloomberg cited informed sources stating that Tesla is disbanding its Dojo supercomputer team, and the project leader Peter Bannon is also set to leave. Tesla CEO Elon Musk has personally ordered the termination of the project, and the company plans to increase its reliance on external technology partners like NVIDIA and Advanced Micro Devices. Meanwhile, about 20 former members of the Dojo team have joined a startup called DensityAI, triggering widespread attention and speculation in the market regarding Tesla's future AI strategy.
The original goal of the Dojo project was to create a dedicated supercomputer to process the massive video data collected by Tesla vehicles, thereby training the machine learning models behind its Full Self-Driving (FSD) and Optimus robots.
Reducing Costs and Increasing Efficiency
Morgan Stanley believes that cutting the Dojo project is a strategic adjustment focused on reducing costs and increasing efficiency. Although Tesla has never disclosed the specific expenses of Dojo, the report believes that the project consumed a significant amount of capital and operating expenses.
The report specifically mentions that Tesla pointed out in its second-quarter financial report that the increase in operating expenses was driven by AI projects and the expanded AI training scale at the Texas Gigafactory. The company's previous guidance indicated that capital expenditures for fiscal year 2025 would exceed $9 billion, primarily consisting of "AI-related capital expenditures."
Therefore, cutting the Dojo project will directly help control this massive expenditure.
Dojo chip video screenshot
DensityAI Attracts Key Talent
Accompanying the contraction of the Dojo project is the loss of core talent. The report emphasizes that a startup called DensityAI is attracting key talent from Dojo.
This company is led by Tesla's former AI head Ganesh Venkataramanan and is currently still in "stealth mode." Reports indicate that about 20 former members of the Dojo project have joined DensityAI, including Tesla's former head of investor relations, Martin Viecha The company aims to build computing infrastructure for training robots and AI agents in the physical world, with unconfirmed reports suggesting it is seeking "hundreds of millions" in financing.
Strategic Restructuring: Strengthening xAI Synergy, Focusing on Robotics and Edge Inference
Analysts have observed that xAI, another company under Musk, is taking on more responsibility in the development of AI "brains." xAI utilizes data from the social media platform X and some real-world visual data from Tesla for model training.
The scaling back of the Dojo project may pave the way for deeper collaboration between Tesla and xAI.
At the same time, Tesla's strategic focus seems to be shifting towards its vast vehicle network, viewing it as a "large-scale distributed inference cloud."
With the commercialization of the Optimus humanoid robot on the horizon, it is expected that Tesla will allocate more incremental capital expenditures and R&D resources to reduce the bill of materials (BOM) costs and enhance manufacturability, while continuing to strengthen its edge (vehicle-side) data collection and inference capabilities.
Musk Video Screenshot
Mature External Computing Environment, Urgency for Self-Developed Chips Decreases
The external environment for making this adjustment has also changed. About two years ago, Musk publicly voiced concerns multiple times about the shortage of GPU resources.
Now, although Tesla remains an important customer of NVIDIA, the shortage of GPU resources is not as severe as it was two years ago, providing external conditions for Tesla's strategic adjustment.
From the perspective of return on technology investment, abandoning Dojo also makes sense. Morgan Stanley's U.S. semiconductor analyst Joe Moore pointed out:
As NVIDIA's annual R&D investment exceeds $15 billion, its investment scope has expanded to rack connectivity, software, and services, making it increasingly difficult for other companies to create application-specific integrated circuits (ASICs) that surpass NVIDIA in mainstream workloads.
He believes that self-developed chips only make sense when the workloads are completely different from NVIDIA's focus (such as general generative AI). Additionally, AMD may offer rack-scale solutions by 2026, providing another viable alternative for the market.
The above exciting content comes from [Chasing Wind Trading Platform](https://mp.weixin.qq.com/s/uua05g5qk-N2J7h91pyqxQ).
For more detailed interpretations, including real-time interpretations and frontline research, please join the【 [Chasing Wind Trading Platform ▪ Annual Membership](https://wallstreetcn.com/shop/item/1000309)】
[](https://wallstreetcn.com/shop/item/1000309)