Tesla's Bold Step into Data Centers: Key Insights

Recently, Tesla announced that it will be constructing a data centre that the world has never seen before, and while some are excited by the news, others are somewhat sceptical, thinking that this claim will be like most of Elon Musk’s other ventures. What data challenges does AI in vehicles present, what has Tesla claimed, and why is it likely to be utter nonsense?

This claim is substantiated by a recent job posting by Tesla, where they are actively seeking a "Sr. Engineering Program Manager, Data Center" to lead the design and engineering of these innovative data centres.

What data challenges dos AI in vehicles present?

Despite what many claim, self-driving vehicles are still nowhere to be seen, and those that claim to have it are usually just advanced forms of cruise control, capable of staying in lane and watching out for dangers. Even though some vehicle companies have demonstrated autonomous vehicles, they either require a human operator to watch out for mistakes or follow pre-defined routes.

Undoubtedly, one of the primary reasons why self-driving vehicles are still a fantasy comes down to the inability of current AI models to properly understand their environment and make safe driving decisions. In order for an AI to provide better results, it not only needs to have access to a large amount of data during operation but also for learning.

To increase the amount of data available to an AI, the addition of more sensors on a vehicle is generally needed and can include radar, LiDAR, and visual cameras. Furthermore, increasing the resolution of such sensors further allows AI to distinguish small details, thereby being able to see more of its environment. 

However, trying to provide this extra data comes at the increased cost of sensors and challenges in getting that data to an AI fast enough. Additionally, the hardware needed to execute AI can require power-hungry computers, which are themselves costly. 

When gathering data to train an AI, it is essential that data is not only vast in quantity and resolution but also representative of all possible situations. Simply put, providing an AI with training data from one straight road is not enough to prepare the AI for all roads that it may encounter. Instead, numerous roads with curves, junctions, and lanes need to be used, as well as variations of those roads in different conditions, including rain, wind, snow, and traffic.

Getting this training data can be done by streaming sensory data from vehicles in real-time to remote data centres, but considering that such a datastream would consume immense amounts of bandwidth, only networks such as 5G are capable of meeting such demands. Even then, 5G is not available everywhere, meaning that automotive manufacturers have to rely on in-house testing and simulations.  

Overall, trying to pull data from vehicles in real-time is extremely difficult to do with modern infrastructure, and running an AI local to a vehicle introduces numerous technical challenges. The combination of all of this has resulted in self-driving vehicles struggling to take off. 

Tesla claims to be reinventing the data centre, except it won’t

In a bizarre claim, Tesla has recently stated how it will be developing a new type of data centre that the world has never seen before. It is well established that Tesla is looking to create the world’s first truly autonomous cars, and while it has certainly developed interesting cruise control systems (Autopilot), it is far from actually producing a commercial self-driving feature (something which Elon Musk claims every year that it is only a year away, for the past ten years).

Of course, Tesla understands the dire need for vast amounts of data,  and it is likely that Tesla has struggled to get enough data to provide additional training to its vehicles (this is true for any automaker looking to develop self-driving). Furthermore, the computational power needed to train self-driving AI systems is arguably greater than anything currently available to mankind, hence the need for new datacentres with improved computation and AI capabilities.

As such, it makes sense that Tesla has begun to purchase data centres and hire experts in the field of large-scale computing systems. But, what is unusual is that Tesla claims to be developing a new data centre that, when completed, would be a first of its kind. Currently, very little data has been shared on what this data centre would do and what features would make it unique. 

This move into the data centre domain is not without precedent. As highlighted by Electrek, Tesla has been strategically acquiring data centres previously owned by Twitter, indicating a broader plan to bolster its data processing capabilities.

In addition to this lack of detail, one of the data centres acquired by Tesla is in Sacramento. To most, this acquisition would appear to be normal, but considering that Elon Musk himself said that Sacramento is the worse place to have a data centre, it’s very unusual that this is the location chosen.

This acquisition strategy is intriguing, especially given Electrek's report on Tesla's recent takeovers of data centres from Twitter. Despite Elon Musk's previous reservations about Sacramento as a location, Tesla's actions suggest a calculated move to harness the existing infrastructure for its ambitious data processing goals.

Why is this claim utter nonsense?

There is no doubt that Tesla will build a data centre that heavily relies on specialised hardware used to train AI models (such as ASICs, accelerators, and GPUs), but it is highly unlikely to be a first of its kind. To understand why this is the case, one only needs to look at the world’s biggest tech developers; Google, Amazon, and Microsoft. 

These companies have spent decades in the data centre industry, developing all kinds of hardware and software solutions that are able to meet the growing needs of the world. With billions upon billions of dollars in investment, some of the brightest minds in the world, and an extremely large profit incentive, it is highly unlikely that these companies would be suddenly overtaken by an EV maker that consistently has issues with quality, performance, and manufacturing. 

Just like how Elon Musk’s robot demonstrated nothing that didn’t already exist, a data centre developed by Tesla would unlikely present a technological revolution that would make Google and Amazon quake in their boots. Of course, that isn’t to say that Tesla’s data centre won’t deploy advanced tech (such as those used in the Optimus robot), but to say that Tesla will change the industry with its development is ludicrous.