Welcome to Electrified, it's your host Dylan Lumis. First up today, this article from Laura Kaladny didn't teach us a whole lot of new information, but it did a great job of stirring up the conversation, including from Elon. Replaying to this article on X, Elon did say Laura Liar Kaladny. Laura said thanks to some emails from Nvidia staff. The word was that what Elon told shareholders about how many H100's Tesla was buying was not accurate. Second was word that Elon was prioritizing H100 chips for XAI, rather than for Tesla, effectively delaying the delivery for Tesla. An Nvidia memo from December, so this is already interesting timing, a few days ahead of Tesla's annual meeting, said Elon prioritizing H100 cluster deployment at X, versus Tesla, by redirecting 12,000 shipped H100's originally slated for Tesla to X instead. But based on initial commentary, it felt like no one read the following sentence, which said, in exchange, original X orders of 12,000 H100's slated for January and June to be redirected to Tesla. Had Laura been paying attention, maybe she would have included in her article the fact that Gigatex has had seen some construction delays that was impacting the build out of Tesla's data center in Austin. Also conveniently left out of the article was what Elon said on X back in March that improvement will accelerate dramatically now that we are no longer AI training compute constrained. There was also some confusion out there about the 85,000 number and if that was going to be H100's exactly or H100 equivalents.
On X, Ryan said, I think 85,000 is the right number. The new data center will house 50,000 and they have an existing 35,000 H100's. This should all be in addition to the dojo systems. And Elon confirmed it saying correct. Also, I cannot overstate the difficulty of making 50,000 H100's train as a coherent system. No company on Earth has been able to achieve this yet. Once we have that system working, we will order more hardware. That means Tesla is working on getting 85,000 H100's not equivalent, but actual H100's setup, 50,000 of which will be at Gigatexis. And then all of that will be in addition to whatever training they're doing with dojo. AKA even more compute than many of us were expecting because the 85,000 H100 equivalents could have actually included dojo, but Elon clarified that's not the case. Just to be very clear, this is not a situation of X AI receiving H100 chips that now Tesla will not get. It was simply a switch of the timing between delivery for Tesla and X AI with X AI being prioritized first.
That should have been clear even before Elon added any context, but he did confirm Tesla had no place to send the Nvidia chips to turn them on so they would have just sat in a warehouse. The south extension of Gigatexis is almost complete. This will house 50,000 H100's for FSD training. AKA, Tesla was simply not ready for that specific shipment of 12,000 H100's. Then the second point in the article talking about Elon's public comments about Tesla's AI spend not really lining up with Nvidia internal memos. Here's what Elon said. Of the roughly $10 billion in AI related expenditures, I said Tesla would make this year. About half is internal, primarily the Tesla designed AI inference computer, which is hardware 4, and sensors present in all of our cars plus dojo. So for 2024, that's about $5 billion in spend for hardware 4, sensors, and dojo. We did not get a further breakdown of that group.
Elon said for building the AI training superclusters Nvidia hardware is about 2 thirds of the cost. My current best guess for Nvidia purchases by Tesla are 3 to 4 billion dollars this year. Holmars asked Elon how he saw dojo comparing to Nvidia saying is there a realistic pathway to one day producing training compute internally rather than buying from Nvidia. Elon said training compute for Tesla is relatively small compared to inference compute as the latter scales linearly with the size of the fleet. Remember Tesla's inference compute is going to be hardware 3 and 4, the computers actually onboard Tesla vehicles. Perhaps the best way to think about it is in terms of power consumption. When the Tesla fleet reaches 100 million vehicles, peak power consumption of AI hardware and cars will be around 100 gigawatts. Training power consumption is probably less than 5 gigawatts. These are very rough guesses.
Obviously 5 gigawatts of AI training compute is enormous by current standards but is only about 5% of total Tesla AI compute. There is a path for dojo to exceed Nvidia. It's a long shot as I've said before but success is one of the possible outcomes. Recall what Elon said on the Q1 call this year. Tesla's AI inference efficiency is vastly better than any other company. There's no company even close to the inference efficiency of Tesla. Elon's answer didn't really give us what many were looking for to find out how dojo in its current iteration is actually comparing to Nvidia H100s for training FSD. The last update we heard was that dojo could be making up between 10 and 20% of Tesla's training compute needs but that's really just a rumor. Elon's been trying to make the point very clearly when the conversation of Tesla's AI compute comes up rather than thinking about training coming first. We should be thinking about inference coming first. And on X, Holmars reminded everyone from what he heard from his sources at Nvidia, Tesla has ordered over 100,000 H200s from Nvidia that they'll be taking delivery of over the next 12 months. Deliveries have already started. The simple takeaway from all of this is that Tesla is doing just fine when it comes to compute resources and Tesla will continue to build out its training compute as fast as its infrastructure will allow.
On that front Tim Zaman who used to work at Tesla said the practicalities of building a supercomputer are insane once you get down to it. Let's assume you have a data center location and order GPUs. You're probably thinking of these sleek orderly photos of racks of compute. Just think about how those GPUs got there. The picture on the screen is only a few hundred GPUs. Just taking delivery of thousands of GPUs is a crazy military operation even before any racking, stacking, and cabling or bring up. Data centers are usually built out slowly but in the AI craze you want it up as soon as possible and existing data center facilities do not accommodate for this well. Thanks to the limited amount of trucks, warehousing, unloading docks, ramps, dolleys, jacklifts, corridors, etc. So I asked Tim, I said, hey can you share a rough timeline on how long it takes to get 10,000 H100s up and running after taking delivery. To which Tim said, taking delivery sounds like receiving an Amazon package. Although he did use the exact same phrase in his post. I digress. Tim continued, for taking delivery, calculate back for 10,000 GPUs, GPUs in a server, servers in a box, boxes in a truck, amount of trucks, amount of loading docks, and the time it takes to unload a truck. To which I said, fair enough, but I think you know I meant from the time everything arrives on site to the time the equipment is actually set up, hooked up, turned on, and training data. That was my actual question. Which didn't really get answered, but I suppose something is better than nothing.
So no, Elon is not out to sabotage Tesla in favor of XAI, it's just that sometimes things happen and timelines change and this is one of those situations. It would have been nice to get a few tidbits on how the build out of Tesla dojo is going in Buffalo, still mums the word on that one. I'd also like to go on record again saying that I have a very high conviction that in the long run Tesla will benefit greatly from the advancements and success of XAI. The Tesla China weekly insurance number came in at 15,000. Now just to be transparent, CNE the post reported 15,200 and often different sources will have slightly different numbers. For me, I always like to go conservative, so comparing 15,000 to the same week in quarter one, that number was 12,500. Thus, quarter over quarter Tesla China is now 17.9% ahead of the key one pace. Week nine of quarter two last year came in at 14,500, so now year over year Tesla is down 3.5%. Some people are optimistic Tesla can flip quarter two year over year positive over the next four weeks before the quarter ends, but the numbers from quarter two last year were pretty high the last few weeks of the quarter. Specifically, the average was 16,250 per week. Tesla's insurance as they have a new employee, Alan Laban, who's now the head of insurance partnerships at Tesla.
On LinkedIn, he said, I'm thrilled to share that I've started in a newly created role at Tesla. His goal, to make Tesla vehicles easy and economical to ensure. By partnering with insurance companies, teams across Tesla and collision shops in the US and Canada, we will lower the total cost of Tesla ownership. Why would Tesla, which already has its own insurance company, care about the rates other insurance companies charge Tesla owners because of the mission? High insurance premiums on EVs will slow this transition. That's where the insurance partnerships part of the job comes in. Through better communication, processes and partnerships, we will reduce repair costs and lower insurance prices across the industry. As I'm sure you can tell, this new role at Tesla has more to do with the legacy auto industry than it does Tesla's own insurance line. But as a Model Y owner who is now paying twice as much for car insurance with my Model Y as I was on our Honda CRV, I'm very pleased to hear about this new position. And no, Tesla insurance is not live where I live. Also, Tesla is clearly hiring for some revamped new key initiatives. Perhaps one of my favorite Easter eggs that Tesla has placed as of late, Monroe shared a picture of what looks like a part of a circuit board with the words, high Sandy engraved right on it. This was found during the Cybertruck tear down, so hopefully Monroe got video footage of the moment when Sandy saw that.
That grand opening happening today, but we've learned that products have been shipped out of this Durham facility since December of last year. So this is the company's first North America facility show any product shipping around the country coming out of Durham. With Ken Power celebrating this grand opening in North Carolina, the governor said they know we're leading the EV supply chain with more and more battery chips and parts suppliers coming to North Carolina. The governor said the EV goals they already set have been met, so they're now setting new ones. A quick tip of the cap to Ken Power and North Carolina. The CEO of Hyundai North America said we continue seeing great success in our ego friendly lineup with an overall 50% increase year over year. Both EVs and hybrids continue to gain popularity. More importantly, he said there's a lot of negativity in the marketplace right now regarding EVs, which I find really amazing. EVs have been with us for quite a long time and they are not going away. EVs are the future. It's not a question of if, but a question of when. And so from a strategic perspective, we're really focused and dedicated to promoting EVs. A statement I found to be very refreshing given how many CEOs as of late have been talking about pulling back from EVs. Reuters is saying Tesla will stop production at Gigabrelin for five days in June to improve processes in the factory. They did however talk to a Tesla spokesperson who confirmed the dates and it's actually five days that are non-consecutive starting on June 7th and then wrapping up on June 28th. With three days of paused production in between.
There seems to be some confusion around Tesla's 2024.20 software update specifically when it comes to adaptive headlights, matrix lights and working around corners. I believe this post from Tesla newswire gets it right. They said if you're in the US, Canada, Mexico, Puerto Rico or South Korea, your headlights will adapt to the curves of the road. But for now it's lateral only adaptive headlights, not the full on adaptive headlights where they'll actually change pixels for oncoming traffic on straightaways. To tell if your vehicle has the matrix headlights, look toward the outer right portion of the headlamp and look for the circle. No circle light on the right. No matrix headlight. Obviously when looking at the vehicle that's just for the headlamp on the right. Tesla manufacturing posted congrats giga Nevada team on building their 5 millionth drive unit. There is some Tesla employee chatter on the interwebs about how things are going with the testing of 12.4 and it's really not great. Unexpected braking. Yesterday we talked about bugs making the software unusable. The lane indecision or the hesitation is still persisting. The hard stopping and accelerating is still persisting and the hesitation at turns is still persisting. It is interesting that publicly Elon hyped up version 12.4 so much talking about that 5 to 10x reduction in MPI miles per intervention. But reading all of the tea leaves between the delay on the release to actual customers and what we're hearing from those people testing it, it's not a great start.
This is not at all a reason to panic however I do think it's important to be aware of the ability to now write code for specific problems is no longer there. It may be another few days and dare I say weeks until customers start getting 12.4.1. A Reddit user shared a photo of some new V4 dispensers in Milpedus, California. They're saying the site is still using version 3 cabinets so the speeds are still limited to 250 kilowatts. Norway's largest pension fund, KLP has come out and publicly said they'll be voting against Tesla's plans to redemesticate to Texas and against Elon's 2018 comp plan. They are of course voting in favor for the proposal that would urge Tesla to engage in wage and other labor negotiations. Fidelity is saying that KLP only owns about $162 million worth of Tesla stock. With this one I'll pose the same question that Nick asked, what do you guys think about this Cybertruck light bar wrap? Personally I think it's a cool idea and I'm biased because black is my favorite color but I do think the full light bar all the way across is such a signature of the truck.
The Nikkei Asia shared this chart and if you just focus on the red line you can see the automobile industry capacity utilization rate taking a dip in quarter one. Not quite yet as low as what happened during the illness but certainly approaching it. The capacity rate for auto factories in China is down to about 65% and they typically try to keep the benchmark around 75%. They also pointed out the capacity utilization rate for any V's in 2023 was only around 47.5%. That data would be slightly skewed given how many new NEV companies have been cropping up in the Chinese market. It's a trend to watch because if those rates drop even further than the push from China to export to the rest of the world to find some new demand will be even stronger.
Mary Barra gave a new interview and talking about EV she said I wouldn't say we're recommitting but as we make this transformation it's going to happen over decades. And that's why I could not be more proud of our gas powered fleet as well. We've already known that GM's all electric future was going to be delayed from some initial statements they made years ago but in the same interview she also said we will be just committed because we think in the long term EVs are better. I think over the long term when we have a very robust charging infrastructure people are going to choose EVs because they're better. And look it may not seem like much but I think the more that GM loyal customers and fans here their leader make statements like that the sooner more of them will actually consider buying an EV.
In Europe the all electric explorer has officially begun production and is on its way to customers in Europe. Ford has been pretty clear that this vehicle has been designed specifically for Europe so no plans to bring it to the US. In case you see the article out there talking about a Ford patent where they actually talk about a series of electrically connected modules that could be disconnected and swapped. Think Neo battery pack swapping with a twist. However this patent was filed by Ford back in 2018 and given that Neo is still losing money and Tesla is not doing something like this I would be very surprised if Ford ever tried to commercialize this. If you're interested in leasing a Lucid they're now offering a new $3,750 air credit for new leases. They're also offering a $5,000 on site vehicle bonus if you buy one of their inventory vehicles.
Here we have Ford's US sales and we're just going to focus on year to date through the month of May. Pure EV sales are up 87.8% coming in at 37.2,000 through the first five months of the year. Hybrid vehicle sales have grown 50.9% year over year coming in at 74.000. Mach-E sales are up 88.7% coming in at 18.7.000. F-150 Lightning sales are up 78.5% coming in at 13.000 and on the commercial side their eTransit van is up 111% coming in at 5.3.000. I know the growth percentages seem great but when you actually do the math specifically for the Mach-E that's still less than 4,000 delivered per month. And for the Lightning so far it's been 2,600 per month. Honestly though given all of the production challenges and delays with the Lightning in quarter one pretty encouraging numbers for the Ford EV division.
A quick channel update you can now opt into my electrified newsletter and yes for now it's going to be completely free. The link will always be in the description below or you can just go to electrifiednews.com. I need you guys to know I guard my inboxes like Area 51 so I will be very respectful of yours. To start the newsletter will be for the days when I'm unable to upload a video on YouTube. That way when I do resume uploading I don't have to cover that stale news that I know many of you have already seen. Let's say it's a Wednesday and for whatever reason I can't actually sit down, record, edit and upload a video. I still always follow the news so I'm going to gather all of it as I would and then type it up into a 3-5 minute digestible form. This way if I resume uploading on Thursday then everybody's already in the loop from Wednesday and the Thursday episode will be specific for Thursday news. It's not just for that though I also plan to use this as an outlet to do more of that deep dive research and analysis that I have not done as much on electrified as of late and that's been the case because I haven't wanted electrified videos to become prohibitively long. I know a lot of you just want the news and the data and that's what I try to provide. I do have a few other things planned along the way as a thank you for all of your support and for trusting me with your email inbox. And finally if my YouTube channel was ever taken down this will give me a way to communicate with everybody on what's next. And of course you can always opt out at any time so again it's electrifiednews.com and it'll always be in the description below.
Australian Energy Giant Origin has launched a new service on its smartphone app that'll allow Tesla owners to fully charge their car for less than the price of a cup of coffee. The company is saying this marks the first time an energy retailer in Australia has allowed automated EV charging in their app. This program will give Tesla owners access to a charging rate of 8 cents per kilowatt hour. It's really a fancy way to incentivize scheduled charging and avoid the peak charging times. Origin expects this EV power-up service could save eligible Tesla drivers up to $452 every year. The company said we're starting with Tesla vehicles, the most popular EV brand in Australia, and look forward to progressively expanding EV power-up to support a range of EV models in the near future. As part of Tesla's spring update 2024.14, Tesla is now adjusting the auto-wiper sensitivity if the driver presses the wiper button while the wipers are set to auto. Doing that will temporarily increase the sensitivity of the wipers. This means Tesla is now using user input to actually train the neural nets when it comes to the auto wipers. So not only can we train FSD but now we can also train the wipers too.
Tesla stock closed the day at $174.77 down 0.86% while the NASDAQ was up 0.17%. It was yet another lower volume day for Tesla trading about 31 million shares below the average volume the past 30 days. If there was ever a time to root against the inverse Kramer, now is that time as Jim just said, I am not a seller of Tesla. I think he wins the vote and the stock goes higher. Hope you guys have a wonderful day. Please like the video if you did. You can find me on X-linked below and a huge thank you to all of my Patreon supporters.