Ben Thompson0:01
Related to cost structure, he says, 'Hey guys, I appreciate the recent discussions about WEO versus Tesla business models in autonomous driving. What is missing seems to be the elephant in the room for all AI companies: compute cost. Sure, Tesla's Robo taxi maybe $330,000, but Tesla is spending billions of dollars on Nvidia GPUs and Dojo, whatever that is, in order to make the $30,000 Robo taxi possible. $330,000 then is only the marginal cost. Tesla's capex over the last 12 months was $9.8 billion. For context, during the Model 3 ramp in 2017, Tesla's capex was $4.1 billion. Tesla probably spent about $5 billion building full self-driving data centers over the past year, and that will go up.' Any thoughts on that?
Yeah, that's so implicit in this argument about scale and cheapness. When I say cheapness, I mean marginal cost cheap. Like, what's the cost of one additional unit? And the whole bit about compute is one additional unit is free, barring power and all that sort of thing. And so, the Starship is a great example. They've spent billions and billions of dollars developing that. They have a long way to go to make their money back as far as that goes, but the marginal cost of a kilogram going into space is going to be drastically lower. What Elon does is he takes this fundamental financial equation of tech. This starts the reason why the whole Silicon Valley ecosystem exists. It literally started with silicon, that's why it's called Silicon Valley. And the fundamental nature of chips is they cost a ton of money to develop, but once you're making them, every additional chip is basically free. It's just sand, that's your input. Now it's overstating it a bit, you have to actually process it and get silicon and all these sorts of things, but TSMC, they're spending $30 billion or whatever it might be, or $20 billion on a fab. That's an unbelievable amount of money. TSMC is spending so much money, and then the chips come out, and basically from TSMC's perspective, a wafer that goes in is a few hundred dollars, and from that wafer they get 100 chips. So their costs per chip are tiny. All their costs are depreciation of those fixed costs. And so that is how tech works: massive, massive, massive upfront costs and then basically zero marginal cost at the back end. This model is how venture capital came about, because you need to get started, you need astronomical amounts of money, but your payoff is infinite because you can just make an infinite number of things. So Silicon Valley starts, venture capital is a part of the Silicon story. Then software comes along. Software is the exact same model but even more extreme, because now it actually literally is zero marginal cost, it's just copying the whole thing. But again, huge amounts of investment upfront to make it work. And so software becomes huge, and then you have venture capital becomes this massive industry. What Elon is doing is he's taking that and applying it to the physical world. What does it mean to take a silicon mindset, to take a software mindset, and apply it to cars? What does it mean to take it and apply it to space? And it's not a perfect analogy. Apple is a useful example here, because the brilliance of Apple is they make these incredibly expensive objects, but they do it at scale. It's the scalability that's amazing. And again, it's not a perfect example. iPhones have actual costs, Teslas have actual costs, rockets have actual costs. But it is taking to the absolute extreme. To what extent? People talk about Teslas being computers on wheels, and that is correct. To the extent they're eliminating all the controls and trying to put everything, and even once they're self-driving, it's really the case. But it's taking a computer mindset and applying it to physical goods. That's the consistency in Musk's businesses, and that is Apple as well. And so to Travis's point, yes, absolutely they are spending a lot and they're going to spend even more. And that's why you need big dreams, because the risks are astronomical. If you build and spend all this money and it doesn't work, you are out of business. And that's why regular businesses don't do it. The risks are too great.
