Hiroshima researchers improve kinetics of ammonia synthesis by chemical looping of lithium hydride; targeting small-scale NH3 production
Redwood Materials launches EV battery recycling program; initial partners Ford Motor Company and Volvo

CoBank report suggests self-driving vehicles could easily overwhelm current data center market

Widespread adoption of self-driving EVs faces several significant barriers including social acceptance, safety and regulatory restrictions. But according to a new report from CoBank, another major challenge that is often overlooked is the monumental impact these vehicles will have on the data center market.

While technology developments such as efficient hyper-scale data centers and on-board processing may provide long-term solutions, there’s an enormous chasm today between the existing infrastructure and what is needed to support widespread adoption of self-driving EVs, the report says.

The projected impact widespread adoption of autonomous driving EVs would have on data centers is stunning. According to the Automotive Edge Computing Consortium (AECC), self-driving EVs may eventually need to offload as much as 5,000 gigabytes of data per hour of operation. To put that into perspective, in 2020 the average person worldwide generated about 150 gigabytes per day.

Holon Investments estimates the global self-driving fleet could reach 400 million by 2035. Extrapolating these numbers implies the annual global datasphere will surge from 64 zettabytes in 2020 to an eye-popping 10,000 to 15,000 zettabytes in 2035—just from self-driving vehicles.

An abrupt switch to driverless cars is unlikely. The more likely scenario is a gradual adoption as auto manufacturers’ move up the autonomous driving scale (1-5), starting with commercial applications before moving on to consumer use.

However, some local jurisdictions are beginning to allow limited use of self-driving technologies for ride share services. Las Vegas announced that starting this year, it will allow Lyft and Motional to deploy a limited fleet of self-driving, driverless taxis with a full-fledged commercial launch slated for 2023. Tesla CEO Elon Musk has indicated he expects Tesla will achieve level 4 autonomy this year.

The amount of capital required to build the data center infrastructure for self-driving EVs is so massive that it begs the question: Without major technological advancements in compute and storage processes, will the industry be able to handle the deluge of data self-driving EVs will generate?

In 2020, globally we stored just under 3% of all data created, with a total storage capacity of 2.1ZBs. If we were to increase our storage capacity 100 fold (to 210ZBs) by 2035, only about .4% of the datasphere would be stored. A 100-fold increase in storage capacity represents a 15-year 36% CAGR, a good bit higher than the market intelligence firm IDC’s annual forecast of 18% (from 2020 to 2024). So, will the industry have enough storage capacity to support self-driving EVs if we can only store .4% of the data created? And considering that it has taken over $2 trillion to build 2.1ZBs of storage capacity, is there a (data center) business case to support mass adoption of self-driving EVs? At this point there appears to be more questions than answers.

—“Self-Driving EVs Could Overwhelm the Data Center Market. Can the Industry Respond?”

CoBank is a $155-billion cooperative bank serving vital industries across rural America. The bank provides loans, leases, export financing and other financial services to agribusinesses and rural power, water and communications providers in all 50 states. The bank also provides wholesale loans and other financial services to affiliated Farm Credit associations serving more than 75,000 farmers, ranchers and other rural borrowers in 23 states around the country.

Comments

mahonj

I do not see why this is so. The AVs will do most of their computing in the edge (as they call it) and not in the cloud. They will not want to send large image sets every 30 msec or so to the cloud to get them processed.
Currently, while people are still developing these systems they may upload a lot of data, but once it goes live, this will reduce greatly.
They may well upload results and location data to the cloud but this is tiny compared to the visual, lidar and radar images that them would be sending during the development phase.
After all, humans can drive without any cloud storage (or lidar, or radar), just stereo colour imaging on a flexible mount (human neck).
One thing that people do not appreciate is how good the human visual system it.
We have a very high resolution section in the Fovea near the centre of the eye (for "predator" mode) and then lower resolution in the periphery for keeping watch in "Prey" mode.
While we have a fixed focal length in the eye, we have variable resolution (fovea vs. the rest) and we can "zoom" by imaging with the fovea.
And we can recognize everything in the scene in real time and usually correctly predict its next action.
Not bad at all.

SJC

could reach 400 million by 2035...no

mahonj

Maybe they were pitching for business from disk drive vendors.

The comments to this entry are closed.