TSLA Market Action: 2018 Investor Roundtable

[EVNow]

A $35k Model 3 doesn't reduce anymore CO2 emissions than a $60k Model 3.

Actually it does - because Tesla is going to sell 5 to 10x more of the $35k Model 3s than $60k Model 3s. A lower priced EV is going to have a much bigger impact than a higher priced EV for this reason.

 

[mongo]

That depends. If the $35k Model 3 is driven ~200 miles a day and the $60k Model 3 is driven ~30 miles a day the $35k Model 3 does reduce CO2 emissions more.

Also if the $35k Model 3 is replacing an old gas guzzler while the $60k Model 3 replaces a modern Prius the $35k Model 3 does reduce CO2 emissions more.

For the same use case, 5 SR 3's replacing 5 ICE cars does more for CO2 than 3 LRs replacing 5 ICE...
(insert your own ratios for total cell usage/ sales)

 

[Pezpunk]

aaand a sudden, irrational, puzzing dip began at the precise moment i posted this, haha.

good news: i was able to end the dip by mocking my original post. you're welcome, guys.

 

[MP3Mike]

For the same use case, 5 SR 3's replacing 5 ICE cars does more for CO2 than 3 LRs replacing 5 ICE...
(insert your own ratios for total cell usage/ sales)

And it is more likely that someone buying a $35k Model 3 has a current car that is near EOL and will get scrapped, where someone buying a $60k Model 3 likely has a newer car that will get re-sold. So each $35k Model 3 could directly eliminate one ICE, instead of just passing down the polluting ICE to someone else.

 

[KarenRei]

Karen, Even if Fremont ships BiW's to China to finish assembly, isn't there a parts tariff in place? I don't think Tesla could achieve much in savings if all of the major parts to complete the cars are shipped in from the U.S. It would make more sense to build another GA line in a tent at Fremont and ship those LR cars to Europe ahead of any China builds.

I don't know how China's tariff works, and would appreciate information from anyone who knows better. But beyond the potential for reducing the costs for any value-added local work, the big benefit is simply that they'd be adding a GA line and paint shop, which are bottlenecking Tesla's production at present.

They need them somewhere. They might as well be in China. Not just because they're going to need them there, but because GA is labour intensive, and labour costs in China are low.

 

[mongo]

Historically, doesn't seem like it. Granted, Tesla might be able to make the argument that shipping can be $$, but historically if one looked at a model s, the premium NET of the VAT was about 20%. It just makes sense, that is where the threshold of retail pricing is in Europe/UK. Almost no company will LOWER retail regardless of where currency fluctuations go. Why do it, pricing of products like this and many other types simply tends to RISE with time not fall and the consumer is fine with that. If the currency works in the companies favor, so be it. and if it goes against you and seems like it's going to stay there, well then RAISE pricing for sure - but currency cross rates can be the boogey man.

If one nets out for VAT, a similarly equipped Model S 100 D is about 95K USD USA (not including federal and state credits/refunds and not playing "how much are you saving on gas, maintenance, time at the pump and other crap) compared to a currency adjusted price of about 116K USD. That can't all be shipping.

It's much higher with VAT of course, on the contingent, but I'm not factoring in any state or national taxation.

Similarly to what? (edit due to your edit)
Europe/UK also adds the overhead of Tilburg. For China, the tariff induced changes were applied practically real time. Up and down.

Edit with numbers:
Current S100D price for UK: 93,150 pounds including 16,000 in VAT. 77,150 without VAT which is 98,172 US dollars.
US S100D price is $96,500 with $1,200 destination fee, $95,300 without
So $2,872 price difference to get to UK.

 

[Chartered321]

New era of profitability starts now! What a surprise !!
Price of the stock is all over the place. Fundamentals are in the right place.
What matters to me is the amount of shares to hold for the long time, no matter what's the price.
Only sad thing is that I had to cancel my Roadster founder series and reinvest the proceeds in shares at 271 USD (experience shared on here). But I'll reconsider ordering again soon.
Last thing: I've seen the P&L of goldman today.. I will only report the expected revenues of Mr Tamberrino model:
2018E Revenues: 20.155 USD bn
2019E Revenues: 25.441 USD bn
2020E Revanues: 28.161 USD bn
....what is this guy smoking....

 

[wipster]

SA will never be the same.

That is the best example of clickbait I may have ever seen...

 

[EVNow]

You mean income inequality like during the industrial revolution? Or during the depression? Or during the civil rights movement? Or countless other times in our history? It's nothing new folks. We have been struggling with these issues since the founding of our nation.

From the context of market thread - we aren't talking 300 years of history - we are talking compared to last couple of decades, where all the growth has gone to the top 10% of population that has increased social unrest. You can downplay it as much as you want - but I think we are at a more socially combustible time than anytime in the past 3 decades. Election of Trump is but a symptom of the problems (its happening all over the world too).

 

[KarenRei]

I don't think that ensures a body ready to paint in China. Rework, scrap, inferior quality are all risk. And why not just finish painting the body if you've already run it over to the dipping line. It's not clear at all what is a time saver.

Temporary protective coatings (an option in addition to dry packaging) are much cheaper than consumer-grade paint jobs. And the reason "why not just finish painting the body" is the simple reason, "Tesla does not have the paint shop capacity to finish painting the body". And if they're going to build a new paint shop somewhere, it ought to be in China where they're going to need one.

 

[Jayjs20]

Currently, the stock price is hanging around the upper BB, which is around 312. I hope we can close decently higher than that to continue this bull run.

 

[ZachF]

I hope the stock ends the day at $314.15

$(100π)

 

[EVNow]

A mini-rally is going on. I wonder whether we'll end up higher than open of 317. 314 as I type.

 

[beachbum77]

Most auto OEMs are also reporting declining profits year over year, by double digit percentages, with 0 prospects for significant growth. Meanwhile, Tesla continues to significantly reduce manufacturing costs every quarter, at the same time as dramatically increasing production/delivery(per the ER, increasing the peak Model 3 output by 40% is just a matter of reducing downtime on the existing lines, with minimal capex).

Internally funded or not, the growth prospects for Tesla dwarf those of those other OEMs.

Who cares about the other guys? I was asking what everyone thinks the P/E ratio should be for Tesla? If Tesla becomes perma-profitable then the market measurement becomes their P/E ratio. My point was I believe this may be what is driving some of the lower analyst valuations. I think they are picking a point between auto companies and tech companies; between single digits and 20-30. If we go out over the next 3 quarters for a total of $8-10 we are in the 30-40 P/E range. That is as high as I see Tesla going. If we use 35 then we need a total of $12 in earnings by the end of 3Q19 to get to a $420 SP.

 

[humbaba]

So this is pretty unpopular, but you deserve to have an EV tax. We all do. Since we aren't buying gas, than the portion of the gas price that would go into road maintenance wouldn't be paid. And since we are using the roads..... you get the idea. Now, the rebuttal to this is of course, "semi trucks should be paying the majority of road repair because they damage it exponentially more than cars do!". And this is very true, but not the way the system is setup right now. We use the roads, we pay to maintain the roads. That is fair.

maybe you missed the point where I would be paying significantly more than I currently do with the gas tax. Just a detail, I know.

 

[jimmy_d]

Yeah.

So the reason I suggested a shared RAM design is that I think there's a chance that the Tesla AI NN chip has a really radical design: DRAM integrated onto the NN CPU die itself. This is a relatively modern technique that Intel (Haswell and later) and IBM (Power chips) are using:

https://www.eetimes.com/author.asp?section_id=36&doc_id=1323410

(Having all the weights in SRAM doesn't seem possible currently: the simplest SRAM cell design would require about ~48 billion transistors for 1 GB of weight data which would result in too large dies - and indications are that they are using at least that much weight data.)

The Tesla NN chip might have gone one step further and basically integrated the NN forward calculation functional units into the DRAM cells themselves. One possible design would be that there's an NN input/output SRAM area in the 10-30 MB size range, and the functional units propagate those values through the neural net almost like real neurons.

Such a design would have numerous advantages:

• Heat dissipation properties would be very good, as all the functional units would be distributed across the die evenly in a very homogeneous layout.
• Execution time would be very deterministic as there's effectively no caching required.
• Lack of caching also frees up a lot of die area to put the eDRAM cells on.
• This design would also allow very small gate count mini-float functional units and very high inherent parallelism.
• Scaling it up to higher frequencies would also be easier, due to the lower inherent complexity and the lower critical path length.
• All of this makes it very power efficient as well, i.e. a very high NN throughput for a given die size, gate count and power envelope.

In such a design external RAM modules have a secondary role: they are basically just for initializing the internal "neurons" (multiplier and saturated-add functional unit) and "axons" (weight value) with the static neural net, and to store the output results.

Other designs are possible too - such as self-contained all-in-one 'neuron' functional units that are programmable to perform a given loop of weight calculations with no external communications other than the input fetches from other functional units, the eDRAM cell fetches and the output stores (i.e. intermediate state would not be stored anywhere external outside the functional unit, it's all within small local registers in the functional unit itself with no bus access to them whatsoever) - but the basic idea is to have the NN weights data on-die.

If that's the NN chip design Tesla invented then I'd expect the NN chips on multi-chip boards to share any external RAM, as it's not a performance bottleneck anymore.

But maybe I'm missing some complication that makes such a design impractical - for example the latency of eDRAM cell fetches would be a critical property.

Let me preface this comment by saying that I like this speculation and I think it's not without merit. Embedding DRAM or otherwise co-packaging with something like HBM has a lot of merit for a general purpose NN processor. That said I think it's unlikely that Tesla is doing this for HW3.

Tesla's stated performance for their NN processor is "500% to 2000%" increase - say about 10x the performance of what they have in HW2. This would be about 100Tops 8bit integer inference performance (GP106 in HW2 is about 10Tops). To get 100Tops 8bit on an Inception V1 style NN (the only one we've see in use for the cameras so far) you don't need to co-package a large DRAM memory. All you really need is to build a chip that is specific to doing large matrix multiply operations, includes a nonlinearity module for the multiply-accumulator outputs, and has sufficient SRAM on chip to store the inter-layer results to save sending them on a round trip to external memory. The weights bandwidth requirement for a CNN is not onerous so pulling that from external DRAM is fine.

Something that often gets lost in the smokescreen that Nvidia puts out is that GPUs are not particularly good at NNs; they're just *better* than CPUs. It doesn't take a particularly advanced IC design to beat the pants off of the best Nvidia GPU if all you are going to use the part for processing CNNs. Google's TPU V1 is an existence proof for this.

Of course, it would be wonderful to see a truly advanced design that incorporated large amounts of embedded DRAM and which had support for compressed weight storage and sparse matrix operations. But the time/cost/risk/reward balance right now seems to me to favor doing a minimalist CNN specific part and save the real design fireworks for HW4. I could see something with 1000Tops showing up in 2021 to enable an even more powerful NN capability.

As for HW4:

HW3 isn't going to be the end. Tesla will continue extending its lead by leveraging even greater amounts of data as the fleet grows to millions of cars and beyond. This will create a use case for 1K Tops, then 10K Tops, then 100K Tops. Metcalfe's Law (Metcalfe's law - Wikipedia) applies to fleet learning networks of self driving cars because the value of the network grows linearly with both the number of cars using the network and the number of cars feeding the network, i.e. the square of the number of vehicles. Once you get to a certain size the lead becomes insurmountable as AT&T taught us so well. I think Elon understands this and is planning for Tesla to be "Ma Bell" in the age of the self driving vehicle fleet. That means Tesla has to keep growing aggressively until the point at which creating a new competitor becomes pointless - a decade or two. Building and keeping that lead is going to require successive generations of HWX's.

 

[LoveTSLA]

A mini-rally is going on. I wonder whether we'll end up higher than open of 317. 314 as I type.

It did. Check the volume of last 5 minutes.

 

[wipster]

Ok, how about a commercial funded by Customers? ~$300K for a slot on Big Bang. Get Hollywood to match, then watch it re-run on every news station as the real story and then watch it again and again. $300 each x 500 people = $150K? Do you realize how many people would sponsor this to help save the planet? I bet Micheal Moore would eat it up! "Paid for by the Customers" in bold at the end.

Hello, anybody out there? Maybe wrong crowd...

Better yet, have Sheldon discover how easy it is to drive a Tesla than a regular ICE car! He could have it drive him to CalTech and home again! Then the customer paid commercial could come on.

Of course this all supposes that Tesla NEEDS the advertisement... they already have a supply challenge.

 

[ZachF]

Who cares about the other guys? I was asking what everyone thinks the P/E ratio should be for Tesla? If Tesla becomes perma-profitable then the market measurement becomes their P/E ratio. My point was I believe this may be what is driving some of the lower analyst valuations. I think they are picking a point between auto companies and tech companies; between single digits and 20-30. If we go out over the next 3 quarters for a total of $8-10 we are in the 30-40 P/E range. That is as high as I see Tesla going. If we use 35 then we need a total of $12 in earnings by the end of 3Q19 to get to a $420 SP.

It has nothing to do with being "auto companies" or "tech companies", The reason auto companies are valued so low is that most of them are bloated with debt (at 70-100% of revenue) and have next to no revenue growth (0-4%/year for the last decade).

Tech companies are valued high because they have rapid revenue growth, not necessarily because they are "tech"... Tesla has rapid revenue growth. 70% per year so far, and probably at least 30%/year going forward.

 

[Tslynk67]

Weird - big $AMZN beat and they're 6% down AH...?

Edit: same now for Google (Alphabet), bit beat and down...

$TSLA being dragged down too...

I don't understand stock markets at all