Long time Tesla hater Lora Kolodny is out with a story saying
@elonmusk redirected Nvidia chips from Tesla to xAI. I can tell everyone is going to be talking about this one, so let's dig in.
The first thing to know about Lora is that she has been working closely with Tesla short sellers like my stalker Aaron Greenspan for years. She even helped spread accusations that I had child pornography when Tesla shorts were trying to smear me back in the day. Everything she writes should be taken with a huge grain of salt and she has often been wrong in the past. This time is no different.
Lora's evidence for this story is an internal Nvidia memo from December 2023 which said "Elon prioritizing X H100 GPU cluster deployment at X versus Tesla by redirecting 12k of shipping H100 GPUs originally slated for Tesla to X instead. In exchange, original X orders of 12k H100 slated for Jan and June to be redirected to Tesla"
The conclusion she draws from this is that Tesla is being ripped off at the expense of xAI. But it is already June now, so both companies have received the 12k GPUs. While she assumes that this is because he hates Tesla and wants to prioritize xAI, this is an assumption that excludes several more benign possibilities. For example, it's possible that Tesla was still setting up some of the existing H100s they received and were not ready to take delivery of more. It's possible that Tesla was under margin pressure and needed to spread out delivery over a longer time period to minimize the impact on cash flow. While many will rush to say Elon made a decision to harm Tesla we don't know all the factors that went into this decision and I'm not going to assume the worst just yet.
She then points to a memo that says their $10 billion in AI spending "conflicts with bookings and FY 2025 forecasts". But the Nvidia staffer who wrote the memo didn't realize that that $10 billion includes spending on Tesla's car inference computers as well as their own in house Dojo supercomputer.
Another point worth considering is that Tesla will use xAI language models in products like the Optimus robot and their in car voice assistant. Grok 1.5 vision looks like it could have powerful self-driving applications as well. So it is not just xAI vs Tesla — the models being produced will actually be useful to both companies.
From what I've heard, Tesla is building out training compute massively and is no longer as compute constrained as they were in the past. They have close to 40,000 H100s and are planning to expand to 85,000 by the end of the year. In addition their in house Dojo supercomputer is now in mass production and expanding rapidly.From what I have heard from my sources at Nvidia, Tesla has ordered over 100,000 H200s from Nvidia that they will be taking delivery of over the next 12 months. Deliveries have already started.
Ahead of the vote on Musk's compensation, Tesla short sellers and their allies like Kolodny will be doing their best to make sure you hate Musk and think that he is ruining the company. The reality is that nobody except Musk could have pushed the company to deliver FSD 12.
Sadly many people will believe this story and repeat it as fact. But I have read Lora's story and see no solid evidence that Tesla isn't getting enough delivery chips because of xAI.