In previous analyses, I have speculated on several aspects of both Tesla and SpaceX’s business strategies. In a widely-shared October 2015 piece, I discussed how Tesla would pursue their self-driving vehicle strategy. This piece was endorsed by people such as Marc Andreessen of Andreessen Horowitz; Jessic Verilli of Google Ventures to Jeff Weiner, CEO of LinkedIn.

In July 2017, I analysed in two pieces how Tesla and SpaceX would tie up so that future Tesla vehicle iterations would be connected to a space-based broadband network.

Nine months after the first piece, Tesla CEO Elon Musk confirmed my hypothesis in his Tesla masterplan part deux. Some months after my second and third pieces, Musk was asked on a quarterly call by Morgan Stanley whether a SpaceX / Tesla tie up was in the works. Musk demurred, but in May of the following year SpaceX COO Gwynne Shotwell all but confirmed the plan (“[Starlink] could be used for Tesla,” she said with a smile).

So where are we now?

In 2019, SpaceX launched 120 broadband satellites, and this year they plan to launch an additional ~1,500 satellites, or about 60 satellites per fortnight aboard Falcon 9 rockets.

As I argued in 2017, while no Tesla car models have yet appeared that incorporate a satellite antenna that can hookup to the nascent Starlink broadband network ,  I believe this to be an inevitable addition, possibly being rolled in as either an optional extra or as standard in Tesla models as early as 2021, most likely on the hood/bonnet of the vehicles.

SpaceX expects to make tens of billions in revenues from its Starlink network , essentially a global Internet Service Provider (ISP). This will add fuel to the tank for Elon Musk’s expensive Mars ambitions. There is far more money to be made from telecoms than there is from launch services.

Tesla meanwhile has since massively scaled its car production  (in particular the Model 3)   while also rolling out upgraded hardware for what Musk says will be Tesla’s autonomy platform. It has in parallel ramped its production of batteries both for cars and for static domestic or grid-scale storage ,  and more recently solar (not without its problems).

But could there be yet another angle that Tesla and SpaceX are working on?

Let’s take another look at their strategies.

*****

The first step in any exercise in forecasting is to imagine a future scenario and work your way backwards. So let’s start with assuming that both Tesla and SpaceX are successful in what we know to be their current plans. Let’s look at it under three broad headings. And let’s also imagine that they are , essentially ,  the same company.

Broadband

SpaceX will deploy thousands of satellites in the coming years. The stated goal is to provide very fast, low latency broadband to large parts of the world. The initial focus for market penetration will be areas poorly served by broadband at present. Assuming a rate of $80-$100 per month and millions of customers, this will be an enormous business.

The mistake people usually make when you mention satellite broadband is they base their attitude on prior experiences. Expensive, high latency, unreliable is what they say. Which are all true of existing systems. But this is a new system, using phased array antennas and low earth orbit satellites on a scale never before seen. It is completely different from incumbent satellite broadband systems.

Satellite broadband will be delivered via flat panel antennas fixed to the roofs of customers, be they static premises, ships, planes or vehicles. There are no moving parts – the panel just has to be able to see the sky (you don’t have to point it in any particular direction). One thing that came from the launch of the Cybertruck in November (left) is that there is plenty of scope for a fixed antenna either built-in or as an optional extra on the final design.

With thousands of satellites in orbit over the coming years, the Starlink network will be a very high speed network ,  including extremely fast backhaul. This will also form part of Starlink’s revenues as data transfers over long distances will be faster than via undersea cables , something important for things such as market trading.

Ultimately, this will mean that regardless of where in the world you are, you will be able to access high speed (Gigabit) broadband. All you need is a way to pay (online) and a power supply.

Let’s assume that by 2030 this entire strategy pays off, the grid is deployed and there are millions of happy customers.

Energy

Tesla has a large and growing energy business. This energy business is divided between storage and production – ie battery storage and solar roofs/panels.

All Tesla batteries are actually formed from thousands of a smaller common component, the 2170 format battery which fits in your hand, and there are two main ways they are sold. The first is via Tesla cars, which are essentially lithium-ion batteries on wheels. The second is stationary batteries , via the domestic Powerwall solution, or via the industrial scale Powerpack solution. (The largest example, South Australia’s Hornsdale Tesla storage facility (below) , is now being expanded).

The second part of the business is the nascent solar roof business, plus whatever business / output remains from the SolarCity acquisition. This ties two key parts together — the production of energy from the sun, either to fuel homes, Supercharger stations or businesses; and the storage of energy in the batteries of homes, businesses and vehicles.

Tesla also uses its retail and online market position to sell access to Starlink too, while also selling solar roofs/panels, batteries and cars.

Let’s assume that Tesla continues to sell vehicles, batteries and its solar business begins to grow well. This will mean millions of electric vehicles; static domestic or industrial batteries and of course solar panels. By 2030, there are millions of Teslas on the roads.

Tesla also uses its retail and online market position to sell access to Starlink too, while also selling solar roofs/panels, batteries and cars. Customers could buy a solar roof, an integrated broadband antenna (powered by the sun during the day and battery storage at night), and a battery-powered / Starlink-connected vehicle.

Autonomy

Tesla’s autonomy strategy relates to most current, and all future, Tesla vehicles. Tesla’s view is that the latest iteration of its graphics chip, the Full Self Driving Chip (or FSDC), has hardware powerful enough to support future software upgrades that will lead to Full Self Driving. The chip has been custom designed by Tesla (and made by Samsung) to support its reliance on neural networks as it develops its self-driving software.

As I wrote back in 2015,  Tesla then plans to allow Tesla owners to do things like lease out their vehicles to transport people around autonomously.

The cars will use the FSDC, its onboard cameras, 12 ultrasonic sensors, as well as its connection to either a local 3G/4G network, or in the future on a Starlink connection. To both navigate and learn, the system needs connectivity  — and these connected Tesla cars collect billions of datapoints per day —which it can use to improve its autonomy algorithm. Tesla can then deploy Over The Air (OTA) updates to improve the autonomy skills of its fleet iteratively. And let’s not forget the data needs of passengers in cars that can drive themselves.

Let’s assume that Tesla is successful in its autonomy ambition and in 2030 the world contains millions of self-driving Teslas, many or all connected to the Starlink network.

*****

Now let’s go one step further.

The FSDC is an extremely efficient processor. Because it was designed to suck energy from the batteries inside a moving vehicle, where range is a key requirement, it had to be.

It’s also extremely fast. Here’s Wikichip’s breakdown:

Fabricated on Samsung’s 14 nm process technology, the FSD Chip incorporates 3 quad-core Cortex-A72 clusters for a total of 12 CPUs operating at 2.2 GHz, a GPU operating 1 GHz, 2 neural processing units operating at 2 GHz, and various other hardware accelerators. The FSD supports up to 128-bit LPDDR4–4266 memory.

The chips are limited to drawing 100w of power or less. The peak performance of the GPU and the NN chip is 600 GFLOPS (600 billion floating point operations per second) and 73 TFLOPS (73 trillion floating point operations per second) respectively.

Now let’s do some back of the envelope calculations.

Let’s assume that the vehicle is at rest, parked in a driveway (it may not always be like that but for the sake of argument). Let’s take the larger figure from the NN chips and extrapolate. Let’s assume that as of late 2019, all Model 3s have FSDC chips (they don’t, since many 2018 models have to be retrofitted, but that will happen eventually for those who paid). That’s now over 500,000 vehicles.

Let’s combine those 500,000 vehicles and their theoretical peak performance.

500,000 x 73 TFLOPs = 36,500,000 TeraFLOPS, or 36,500 petaFLOPS

As of today, the fastest supercomputer in the world is the Summit – built by IBM at the Oak Ridge National Laboratory – at 200 petaFLOPS. Tesla’s combined current fleet has a theoretical processing power 200 times that of a Summit Supercomputer.

In addition, each week, at least 500 petaFLOPS of processing power rolls off Tesla assembly lines. That’s two Summit supercomputers per week. At current production levels, including the upcoming Model Y, Tesla will be approaching 70,000 petaFLOPS in theoretical peak processing power across its fleet by the end of 2020.

What could Tesla do with all of this connected processing power? Your guess is as good as mine. But you could certainly dole out tasks to latent processors globally – on demand – and perhaps compensate Tesla owners for the trouble. You could use them to crunch numbers at a rather ridiculous scale.

If you have the Starlink network complete, and you have processing power within the Tesla cars connected to that network, then orbital data centres makes sense.

But for any supercomputer, you need three main things: power, processing and storage.

And, as Steve Jobs used to say, there’s one more thing.

The missing piece in a globally connected supercomputer is storage. The Summit has 250 Petabytes of storage. Where could you put all the data being produced? Teslas don’t have that much of it. They transmit data back via mobile networks. Tesla has no large data centres to speak of. Data centres, as we know well in Ireland, are expensive to build, and extremely hard on electricity. They also need cooling systems.

There are two approaches Tesla/SpaceX could take. One is building data centres close to gateways or internet nodes on land.

The other way is to build data centres in orbit.

If you have the Starlink network complete, and you have processing power within the Tesla cars connected to that network, then orbital data centres make sense. It’s cold, has plenty of access to the sun (and maybe batteries for darkness?), and would be relatively inexpensive to build if you have reusable rockets and economies of scale in mass satellite production.

Placing say Solid State Disks (SSDs) onto satellites and then putting them into geostationary orbit (to relay back to LEO satellites) may sound far fetched, but reusable rockets such as the first stage of the Falcon 9 or the upcoming Starship rocket will mean the actual costs of putting data centres in space will very rapidly fall. They could even use the existing Starlink satellite design but make storage versions using similar parts. And not forgetting that Starship would give the ability to launch extremely large payloads to orbit – of 150 metric tonnes per repeatable launch (or about the weight of a fully loaded Boeing 767-300).

So now we have all of the key features of a global supercomputer:

• Power – via battery storage on vehicles, and solar in both space and on land? Check.

• Processing – via connected GPUs and NNs inside every Tesla vehicle? Check.

• Storage – via orbital satellite data centres (that can be mass produced and launched)? Check.

Who are SpaceX and Tesla competing with in our 2030 scenario? The list grows ever larger. Amazon, Google, Microsoft, IBM, Uber, Lyft, Hertz, Verizon, Comcast, AT&T, Vodafone, Ford, GM, Volkswagen, Toyota, Engie, Chevron, Exxon and many many more.

Musk has often said Tesla will be a trillion dollar company, or 10 times its current size. 

I can already hear people saying: but what about iPhones, or Xbox, or Playstation – those are all fast processors that are connected? But I think the key difference is Tesla/SpaceX owns the entire stack, the processors, the storage, the power supply and the network. It will have full control over all pieces of the puzzle.

And one last thing, depending on how the Starlink satellites are designed, SpaceX could ultimately enter the mobile handset/telephony market, with devices able to access the internet directly via Starlink connections alone, bypassing land-based mobile networks entirely. This requires some advances in power/connectivity designs, but it is something that should be considered as an obvious future option in the coming years. (Recent reports that Apple are investigating space-based connectivity show that they realise the risks from SpaceX too)

Musk has often said Tesla will be a trillion dollar company, or 10 times its current size. If he can pull together aspects of SpaceX and Tesla, he might be right. 

*****

Disclosure: I started buying Tesla stock in 2013 and retain a small shareholding.

Gavin Sheridan is chief executive of legal intelligence firm Vizlegal. He was previously director of innovation at Storyful.