2017 Investor Roundtable:General Discussion

[larmor]

No disrespect to anyone, but what he showed in this presentation – which might not even be representative of Mobileye's status quo or potential future – is the complete opposite to how Tesla, comma.ai, Uber and most likely Waymo approach FSD.

• Mobileye's approach, according to this presentation:
  Let's divide driving into x sub-tasks / activities / goals and try to come up with a separate mathematical equation or logic for each and everyone of them.
  
• Everyone else (simplified):

1. Let the neural net figure out what driving is and how it works based on provided training data.
2. Sandbox the AI and set hard limits ("Under no circumstances do this or that ever." etc.)
3. Optional: Have an observer AI judge the driving AI's decision.
4. Let it self-improve in simulations and/or shadow driving until it's x-times safer than a human driver in any given situation. Adjust and improve training data, if necessary (go back to step #1, rinse and repeat)

Bottom line:
Mobileye's goal is to build the best-possible robot-like driver agent, everyone else tries to recreate a flawless human-like driver using AI.

Both approaches and goals come with their respective set of pros and cons.

My bet is on deep neural networks.

​

Taking google's deep mind approach to playing GO, version 1 learned from other prior games played. Version 2 played against itself. Applying this to cars, the AI driver could drive against itself in a simulation-- and subsequently cut itself off...

 

[Anstandswauwau]

Taking google's deep mind approach to playing GO, version 1 learned from other prior games played. Version 2 played against itself. Applying this to cars, the AI driver could drive against itself in a simulation-- and subsequently cut itself off...

No worries, Alpha Go Zero's approach is only feasible for systems with rules of low complexity. Yet

No, in all seriousness: Comparing the complexity of our phyiscal reality with a system that is based on a rigid 19x19 2D array with binary tokens would be a bit of a stretch, don't you agree?

 

[larmor]

No worries, Alpha Go Zero's approach is only feasible for systems with rules of low complexity. Yet

No, in all seriousness: Comparing the complexity of our phyiscal reality with a system that is based on a rigid 19x19 2D array with binary tokens would be a bit of a stretch, don't you agree?

Not unless you have enough nodes and computing power. I/O is probably the problem, getting the video data out, processing it for analysis. The real system of driving is not totally discrete, but not impossible.

What is interesting is that Alpha Go Zero used a creative approach, heretofore not thought possible by a non human.

 

[Yggdrasill]

I'm still trying to wrap my head around the economics of the Semi, and I have to admit I'm a bit concerned. Especially when it comes to required capex and margins.

I think a Megacharger capable of doing 100 charges per day at 0-80% will cost around 14.5 million USD; including solar and batteries. If we assume each Semi on the road will do one charge per day, plus some slow charging, that works out to a capex of 145,000 per Semi, just to keep the buyers happy. Now, I've calculated the breakeven of the Megachargers to be around 15 years, so Tesla likely wouldn't lose money on the Megachargers, however, it is a massive undertaking.

1000 Megachargers could support a fleet of around 100,000 Semis, and would cost 14.5 billion USD. That's not pocket change. And assuming each Semi would be replaced at 1 million miles, and would use 30% slow charging, the Semis would be replaced every 1,000,000 miles / ((365 x 400 miles) / 0.7) = 4.8 years. So, that 14.5 billion USD would support annual sales of 100,000 / 4.8 = 20,800 Semis.

If we assume the ASP is 300k USD, the Semis would represent 6.24 billion in revenue. That's relatively tiny compared to most other opportunities for Tesla. And what are the margins? At 300k USD, I'm thinking maybe 20%. Even the Roadster seems like a more attractive opportunity for Tesla. It should have higher margins, similar revenue, and close to zero required capex for charging.

I don't know, from the technical standpoint I do like the Semi, but I'm less sure about the economics. I'm thinking I would have rather seen Tesla promise 12 cents per kWh, not 7 cents. That would have made the Megachargers a meaningful profit center.

 

[Anstandswauwau]

Not unless you have enough nodes and computing power. I/O is probably the problem, getting the video data out, processing it for analysis. The real system of driving is not totally discrete, but not impossible.

Didn't say it was impossible, just "not feasible yet"

What is interesting is that Alpha Go Zero used a creative approach, heretofore not thought possible by a non human.

The way I see it, it's more of a brute force rather than a creative approach. But that's just a subjective judgment, which is pretty irrelevant as long as it does it job well, haha.

 

[Runarbt]

I'm still trying to wrap my head around the economics of the Semi, and I have to admit I'm a bit concerned. Especially when it comes to required capex and margins.

I think a Megacharger capable of doing 100 charges per day at 0-80% will cost around 14.5 million USD; including solar and batteries.

I think you are way off here.. Do not think Tesla will build chargers which cost 14.5million.

No need to start out with solarpanels enough to support 100 semi charges a day. Gird connection at first. Shouldnt cost more than the 50stall supercchargerstations. Really doubt these cost that many millions each.

Build out the initial network like this. Then when semi-production is properly on its way, they can think about adding needed batteries and solarpanels.

Add batteries when they have these to spare. Production ramped properly, and price lower. Then when solarpanel production is ramped.. add these. no hurry, at least not until it makes economical sense to get less dependent on grid connection.

 

[3Victoria]

My best guess:
EAP AP1 = MB chip (for visual object recognition) + discrete-driving-logic
EAP AP2/2.5 = NN (replacement for MB chip) + discrete-driving-logic.
FSD = NN ( " ) + route-planning + deepNN-driving-logic

One was a bandaid and the other is the real solution.

I agree with this interpretation.

 

[hobbes]

If we assume the ASP is 300k USD, the Semis would represent 6.24 billion in revenue.

Don´t know if this was posted yet, but the Italian company Fercam who reserved said a semi is $250,000:
Fercam bestellt Tesla-Elektro-Lkw vor und verrät Kaufpreis

 

[mongo]

I'm still trying to wrap my head around the economics of the Semi, and I have to admit I'm a bit concerned. Especially when it comes to required capex and margins.

I think a Megacharger capable of doing 100 charges per day at 0-80% will cost around 14.5 million USD; including solar and batteries. If we assume each Semi on the road will do one charge per day, plus some slow charging, that works out to a capex of 145,000 per Semi, just to keep the buyers happy. Now, I've calculated the breakeven of the Megachargers to be around 15 years, so Tesla likely wouldn't lose money on the Megachargers, however, it is a massive undertaking.

1000 Megachargers could support a fleet of around 100.000 Semis, and would cost 14.5 billion USD. That's not pocket change. And assuming each Semi would be replaced at 1 million miles, and would use 30% slow charging, the Semis would be replaced every 1,000,000 miles / ((365 x 400 miles) / 0.7) = 4.8 years. So, that 14.5 billion USD would support annual sales of 100,000 / 4.8 = 20,800 Semis.

If we assume the ASP is 300k USD, the Semis would represent 6.24 billion in revenue. That's relatively tiny compared to most other opportunities for Tesla. And what are the margins? At 300k USD, I'm thinking maybe 20%. Even the Roadster seems like a more attractive opportunity for Tesla. It should have higher margins, similar revenue, and close to zero required capex for charging.

I don't know, from the technical standpoint I do like the Semi, but I'm less sure about the economics. I'm thinking I would have rather seen Tesla promise 12 cents per kWh, not 7 cents. That would have made the Megachargers a meaningful profit center.

If they only charge once per day, it can happen slowly at night. No need for the 30 minute charger.

Even if it were 14.5 million for 100 charges a day. That is 145 thousand a charge-day over the lifetime of the system. Say 30 years, 14.5 million / (30*365*100) = $13.25 a charge. 800 kWh per charge is $0.017 per kW.

If it is financed with a 15 year 5% loan. that's 114,670 a month: 30*100 charges = $38.22 a charge @ 800 kWh that's $0.048 per kWh.
$38.22 @ 2.50 a gallon diesel = 15.3 gallons * 10 MPH = 153 Miles equivalent cost. That's a good deal from day 1.

And that is just during the financing period. If it has a service life of 30 years, the next 15 years provide all free fuel.

 

[Yggdrasill]

I think you are way off here.. Do not think Tesla will build chargers which cost 14.5million.

No need to start out with solarpanels enough to support 100 semi charges a day. Gird connection at first. Shouldnt cost more than the 50stall supercchargerstations. Really doubt these cost that many millions each.

Build out the initial network like this. Then when semi-production is properly on its way, they can think about adding needed batteries and solarpanels.

Add batteries when they have these to spare. Production ramped properly, and price lower. Then when solarpanel production is ramped.. add these. no hurry, at least not until it makes economical sense to get less dependent on grid connection.

The economics of the superchargers work quite well, because people charge 80-90% at home, and they don't use that much energy either.

If you don't have free unlimited supercharger with your Tesla today in California, the rate is 20 cents per kWh, and I think this is a realistic picture for what the electricity would cost from the grid. That means that a Megacharger using grid electricity and doing 100 0-80% charges per day would lose 10,400 USD per day. Or 3.8 million USD per year. Plus depreciation of the Megacharger and maintenance. Even assuming Tesla would net 50k USD from a Semi sale, that profit would be gone in 1.3 years. And before the Semi is ready to be scrapped, Tesla would lose another 130k USD on the Megacharging...

 

[Yggdrasill]

If they only charge once per day, it can happen slowly at night. No need for the 30 minute charger.

Slow charging requires investment and effort by the company, and you're not guaranteed 7 cents/kWh. It would be much easier and cheaper to tell the driver to Megacharge the Semi at a stop.

I think the Semi might get a ~50 kW onboard charger, which means it would take 20 hours to charge, and wouldn't be able to fully charge on off-peak electricity. I may be mistaken about that, though. (What is the logical grid connector in the US? The highest power socket that is at all common in Norway is a ~43 kW 400V 63A three phase socket.)

Even if it were 14.5 million for 100 charges a day. That is 145 thousand a charge-day over the lifetime of the system. Say 30 years, 14.5 million / (30*365*100) = $13.25 a charge. 800 kWh per charge is $0.017 per kW.

If it is financed with a 15 year 5% loan. that's 114,670 a month: 30*100 charges = $38.22 a charge @ 800 kWh that's $0.048 per kWh.
$38.22 @ 2.50 a gallon diesel = 15.3 gallons * 10 MPH = 153 Miles equivalent cost. That's a good deal from day 1.

And that is just during the financing period. If it has a service life of 30 years, the next 15 years provide all free fuel.

I assumed some use of more expensive grid power, but yes, they might be profitable in a decade or two...

 

[mongo]

Slow charging requires investment and effort by the company, and you're not guaranteed 7 cents/kWh. It would be much easier and cheaper to tell the driver to charge the Semi at a stop.

I think the Semi might get a ~50 kW onboard charger, which means it would take 20 hours to charge, and wouldn't be able to fully charge on off-peak electricity. I may be mistaken about that, though. (What is the logical grid connector in the US? The highest power socket that is at all common in Norway is a ~43 kW 400V 63A socket.)

For an 8 hour charge, one 125 kW supercharger @ $25-$60k each would suffice per truck. Tesla could still provide the electricity/ storage, and TOU rates would assist additional needs.

 

[oneday]

(What is the logical grid connector in the US? The highest power socket that is at all common in Norway is a ~43 kW 400V 63A three phase socket.)

Possibly this one. 480v 3-phase 100amp pin and sleeve connector. 4100P7W Pin And Sleeve Plug 100 Amp 3 Pole 4 Wire | ElecDirect

 

[Yggdrasill]

For an 8 hour charge, one 125 kW supercharger @ $25-$60k each would suffice per truck. Tesla could still provide the electricity/ storage, and TOU rates would assist additional needs.

That would certainly help, provided off-peak commercial electricity costs under 7 cents per kWh. Or the operator would cover the installation costs for solar and batteries. If Tesla would provide solar and batteries, the capex challenge would only be moved from the megacharger to the supercharger.

 

[mongo]

That would certainly help, provided off-peak commercial electricity costs under 7 cents per kWh. Or the operator would cover the installation costs for solar and batteries. If Tesla would provide solar and batteries, the capex challenge would only be moved from the megacharger to the supercharger.

Superchargers are 1/16 the rate of megachargers, so that should cut the non-revenue generating cost a bit. If Tesla charges an upfront fee to the trucking company, they can offset costs and, if they have extra capacity, sell to the utilities at market rate.

Maybe set it up similar to the 3: get revenue from trucking co. before the payments on the equipment are due.

 

[Anstandswauwau]

Don´t know if this was posted yet, but the Italian company Fercam who reserved said a semi is $250,000:
Fercam bestellt Tesla-Elektro-Lkw vor und verrät Kaufpreis

Some additional tidbits from the above-mentioned article:

• Fercam is apparently the first European company to order the Semi, probably because …
• … it can only be preorder in North America. They had to deposit the preorder fee of $5k via a proxy company in New Jearsy.
• Semi is said to cost $250k but they article didn't specifically stated what version they have preordered.
• Fercam expects the Tesla Semi to have 10% lifetime savings over conventional ICE semis.
• Daimler – their current main supplier for semis – are yet to give them a binding commitment(?) for their upcoming EV semi. Same applies to Einride as well.

 

[larmor]

I'm still trying to wrap my head around the economics of the Semi, and I have to admit I'm a bit concerned. Especially when it comes to required capex and margins.

I think a Megacharger capable of doing 100 charges per day at 0-80% will cost around 14.5 million USD; including solar and batteries. If we assume each Semi on the road will do one charge per day, plus some slow charging, that works out to a capex of 145,000 per Semi, just to keep the buyers happy. Now, I've calculated the breakeven of the Megachargers to be around 15 years, so Tesla likely wouldn't lose money on the Megachargers, however, it is a massive undertaking.

1000 Megachargers could support a fleet of around 100,000 Semis, and would cost 14.5 billion USD. That's not pocket change. And assuming each Semi would be replaced at 1 million miles, and would use 30% slow charging, the Semis would be replaced every 1,000,000 miles / ((365 x 400 miles) / 0.7) = 4.8 years. So, that 14.5 billion USD would support annual sales of 100,000 / 4.8 = 20,800 Semis.

If we assume the ASP is 300k USD, the Semis would represent 6.24 billion in revenue. That's relatively tiny compared to most other opportunities for Tesla. And what are the margins? At 300k USD, I'm thinking maybe 20%. Even the Roadster seems like a more attractive opportunity for Tesla. It should have higher margins, similar revenue, and close to zero required capex for charging.

I don't know, from the technical standpoint I do like the Semi, but I'm less sure about the economics. I'm thinking I would have rather seen Tesla promise 12 cents per kWh, not 7 cents. That would have made the Megachargers a meaningful profit center.

Just one more question, regarding the above, could one presume the megacharger is built into the pricing for the semi?

Clearly an in house charger for a large fleet of trucks at one company making round trips is the target market.

 

[Waiting4M3]

Not unless you have enough nodes and computing power. I/O is probably the problem, getting the video data out, processing it for analysis. The real system of driving is not totally discrete, but not impossible.

What is interesting is that Alpha Go Zero used a creative approach, heretofore not thought possible by a non human.

Alpha Go has to beat a human player, I imagine creative approaches are more helpful in competition. Autonomous driving's goal right now is to be safe, so being creative may not be well suited. Who knows, maybe in the future autonomous driving AIs will compete with each other on driving speed/efficiency also, maybe then they will take more creative approaches.

 

[Snapdragon III]

Tesla has enough supercharger sites that they must be buying quite a bit of electricity on commercial scale. I would think that they have a pretty good handle on what the costs would be to power a Megacharger station. I would hope that they wouldn't dramatically screw this up.

 

[Todd Burch]

I'm confident Tesla has done far more homework and with far more information than any of us could hope to have and the Megachargers will in an absolute worst-case scenario break even and more likely be a marginal revenue generator.