Musk has set his sights on AI infrastructure again: Tesla is set to sell "computing power building blocks"

Tesla has also set its sights on the AI infrastructure business.

Just now, Tesla submitted a trademark application named “Megapod” to the United States Patent and Trademark Office (USPTO), planning to sell modular AI data center hardware.

According to the trademark description, this is a modular data center hardware system for AI computing, which includes computer servers, AI data processing hardware, network equipment, power distribution units, and cooling systems.

However, less than a year ago, Tesla disbanded the Dojo team and axed its only self-developed AI training supercomputer.

Just sent Dojo off, and now it's registering a new trademark for an AI data center.

Huh? Is Tesla changing its approach and continuing to work on computing power?

What is Megapod?

The current confirmed information comes from the trademark application.

The trademark name submitted by Tesla is MEGAPOD, with an application serial number of 99893717 and an application date of June 18, 2026.

In terms of type, this is a standard character trademark, which means it's reserving the name “MEGAPOD” first. The application basis is intent-to-use, meaning “intending to use,” but the product has not been officially launched yet.

The description in the trademark application is quite specific. The document states that Megapod covers:

A modular data center hardware system for artificial intelligence computing, including computer servers, computer hardware for artificial intelligence data processing, network equipment, power distribution units, and cooling systems.

It also includes a “self-contained modular AI computing hardware system” and “downloadable software for monitoring, managing, and optimizing the above systems.”

Put simply, Megapod is like a plug-and-play AI data center module.

In a single cabinet, servers, network equipment, power supplies, and cooling systems are all installed. Once it's transported to the site and powered on, it can directly run AI training and inference.

This is also where it is in line with Tesla's existing Megapack and Megablock.

Megapack sells large-scale energy storage batteries, and Megablock is a larger-scale and more modular energy storage system.

And Megapod seems to apply the “modular” concept from the power system to the AI computing power system.

Some netizens have directly exclaimed: Tesla has quietly revealed a huge AI infrastructure layout.

Some people have further speculated that this might be related to what Elon Musk mentioned before about “using idle power to run AI,” and even guessed whether Tesla will connect its Supercharger network, battery energy storage, and AI computing nodes in the future to form a distributed AI infrastructure.

However, at present, Megapod is just a trademark application. There is no prototype, no parameters, no price, and no delivery schedule.

So, there is still a long way to go before Megapod truly becomes a product.

However, this trademark itself also shows that Tesla is seriously considering turning AI infrastructure into a salable hardware category.

Is Tesla trying to steal Nvidia's business?

At first glance, Megapod easily makes people think of Nvidia.

After all, the most expensive and core component in today's AI data centers is Nvidia's full-rack computing power system.

For example, GB200 NVL72 has become one of the de facto standards for high-end AI data centers.

A single cabinet integrates GPUs, CPUs, high-speed interconnections, liquid cooling, and networks. After customers purchase it, they can directly deploy large model training and inference. Currently, global cloud providers, AI companies, and sovereign AI projects are basically centered around this system.

That is to say, in the business of “modular AI computing power,” Nvidia is already an absolute core player.

So, is Tesla's Megapod coming to steal Nvidia's business?

In the short term, not necessarily.

Because Tesla itself is also a customer of Nvidia. Tesla needs a large number of GPUs to train its FSD, robot, and in-vehicle AI models; xAI under Elon Musk is also purchasing a large number of Nvidia chips to build training clusters.

Moreover, Tesla's history of self-developing AI chips has been quite bumpy.

Tesla once high-profilely bet on the Dojo supercomputer, hoping to use its self-developed D1 training chips to support FSD model training.

At the 2021 AI Day, Tesla officially unveiled the Dojo and D1 chips, with the logic of accelerating the iteration of autonomous driving models through a self-developed training system.

But in 2025, the Dojo team was disbanded, the person in charge left, some members moved to AI chip startups, and the remaining staff were transferred to other data center and computing projects at Tesla.

Elon Musk then said that the company should not spread its resources to develop two different AI chip designs, and the focus would shift to AI5/AI6 in the future, relying more on external computing power ecosystems such as Nvidia and AMD.

It seems that Tesla's new move may not be to steal Nvidia's GPU business. Instead, it is more likely targeting another aspect of the AI data center business: power, energy storage, cooling, power distribution, and modular deployment.

This is also the current pain point of AI data centers. Large model training and inference consume a huge amount of electricity. New AI data centers not only lack GPUs but also grid connections, power transformation capacity, cooling systems, and quickly deployable infrastructure.

Many projects may not be able to start even after the chips arrive, often getting stuck in power supply, heat dissipation, construction cycles, and grid connection approvals.

And these issues are closer to the capabilities of Tesla's energy business.

Tesla's real money-making AI business may be batteries

In the past few years, when Tesla told AI stories, the outside world was most concerned about FSD, Optimus, and Dojo.

But from a business perspective, the most direct connection between Tesla and AI data centers may be Megapack.

Megapack is Tesla's large-scale energy storage battery product, targeting the power grid, power stations, industries and commerce, and large infrastructure projects.

After an AI data center is connected to the power grid, it will cause very significant fluctuations in electricity consumption. Especially during large-scale GPU cluster training, the load may rise or fall rapidly, placing high demands on the stability of the power grid.

At this time, the energy storage system can act as a buffer.

It charges when the grid has surplus power and discharges when the load of the AI cluster increases; when training tasks cause power fluctuations, the energy storage system can also help smooth the impact.

This is the real entry point for Tesla's energy business into the AI data center.

Previously, documents showed that xAI had purchased approximately $1 billion worth of Tesla Megapacks from 2024 to April 2026, with a single-month purchase amount of $269 million in April 2026.

This shows that the major spending in AI infrastructure is not only on chips and servers. The power system itself is also becoming part of the AI competition.

In terms of naming, there is a clear continuity from Megapack to Megablock and then to Megapod.

It's simply a “mega trilogy”.

Megapack solves energy storage; Megablock solves larger-scale and faster-deployed power modules.

If Megapod becomes a reality, it may further package servers, networks, power supplies, cooling, and software management into an integrated infrastructure product for AI customers.

However, the AI data center hardware market is extremely complex, with a group of established players such as Nvidia, Dell, Supermicro, Vertiv, Schneider, and Eaton.

From full-rack GPUs to server integration, to liquid cooling systems, to power distribution and UPS, each layer has high engineering thresholds and customer certification cycles.

Tesla's advantages lie in modular manufacturing, battery energy storage, power control, and the AI demand within the Musk ecosystem.

But its shortcomings are also obvious: limited experience in enterprise-level data center delivery, an unstable self-developed AI chip route, and it's still unclear whether customers are willing to entrust their key AI infrastructure to Tesla.

However, Elon Musk has already tasted the benefits from the large computing power orders of SpaceX.

It is reported that Google will pay SpaceX $920 million per month to rent approximately 110,000 Nvidia GPUs and related CPUs, memory, and other components for three years.

Before that, Anthropic had also signed a contract to rent all the computing power of SpaceX's Colossus data center at a price of $1.25 billion per month.

Combined, SpaceX can earn approximately $2.17 billion per month just from “renting out computing power.”

What? Are you saying that a rocket-making company can earn more than $2 billion per month from renting out idle GPUs??

To be such a successful landlord, it's really only Elon Musk.

This also makes Megapod more imaginable:

On one hand, SpaceX turns AI computing power into rentable assets, and on the other hand, Tesla uses the “mega trilogy” to enter the fields of power, energy storage, cooling, and modular deployment.

It's imaginable that Megapod may not become a “Tesla version of Nvidia.”

But when all AI companies are short of power, cooling, and deployment speed, this business may be more practical than telling autonomous driving stories~

Reference links:

[1]https://electrek.co/2026/06/21/tesla-megapod-ai-data-center-hardware/

[2]https://x.com/BullTheoryio/status/2068569421971436011

[3]https://techcrunch.com/2026/06/05/google-will-pay-spacex-920m-per-month-for-compute/

This article is from the WeChat official account “Quantum Bit”, author: Tingyu. Republished by 36Kr with permission.