We’ve got a lot to cover in today’s AMA…
First things first, I want to thank everyone who joined me on Wednesday for my Crypto-AI Fusion event. It’s always thrilling for me to get the chance to talk about these incredible technologies, and even more thrilling that my team and I get to share our strategy to use artificial intelligence (AI) to profit from crypto volatility.
We had a great time. If you missed it and are interested in hearing more, we do have a replay available for a short while longer. You can go here to view it.
I’d also like to take a moment to thank you, Bleeding Edge readers. Your feedback and engagement are invaluable to us and we love hearing from you. And we’re always looking for new ways to bring you all into the fold of what we’re doing behind the scenes at Brownstone Research.
As some of you may have caught at the end of yesterday’s issue, we’re currently taking a survey of what you all would like to receive in addition to your daily Bleeding Edge issues.
We’ve got a few ideas in circulation – a free daily crypto newsletter to help us keep a finger on the pulse of crypto at all times… and a morning update with quick insights into what’s happening in the markets that day – but we’d love to hear your thoughts.
You can click here to access it. There’s no need to enter your name, email, or personal information to take the survey, though we do have an optional space for you to leave us comments if you prefer, or if you have something else in mind.
We look forward to hearing from you.
Now, on to the AMA…
Dear Jeff Brown,
Matt Stoller who writes BIG on Substack has this to say about AI’s energy consumption…
An important point from John Mark Newman on how big tech incumbents have an incentive to drive AI to be extremely resource intensive, even if there are lighter better models available. Basically, if you sell gas, you want to make sure people drive gas guzzlers.
Are there ways to develop and run AI in a more energy-efficient way, as proposed by China’s DeepSeek?
– Richard S.
Hi Richard,
This is a very important dynamic to understand, and I’m glad you wrote in with this.
However, the quote that you provided makes no sense whatsoever. And the analogy is nonsensical.
Technology companies are incentivized to optimize and reduce the energy consumption of any AI application, particularly when it comes to inference – the running of an AI application.
Whether that company is running the AI application in-house (i.e. an operational expense), or it is running the application and selling it as a service, it wants to lower the energy requirements for either savings or improving gross margins.
There is a nuance, however. When a tech company is building a frontier AI model, like X and Grok 3, it is less concerned about energy consumption.
Of course, it will optimize its data center to be as energy efficient as possible, but we should remember that this is a race to artificial general intelligence (AGI). More GPUs equals more energy requirements. And the first to achieve AGI has the potential to dominate the market and earn the trillion-dollar prize.
So when it comes to training a model, this is research and development. There is a strong emphasis on the speed of training right now rather than on energy efficiency.
But when it comes to operating the AI models, inference, there is an effort for energy efficiency.
It’s important to state that there are some tradeoffs. AI applications with higher energy consumption are more accurate and can take on more complex tasks.
Complex reasoning, chain-of-thought analysis, agentic technology, and high-resolution image generation are examples.
After a frontier model has been trained, it can be made more efficient though. AI models can be compressed or “pruned” to make them more efficient. Quantization can be used to make the models less precise. And the knowledge can be transferred into a smaller model, a process known as “distillation.”
This is precisely what DeepSeek did, illegally, with OpenAI’s technology.
All of these approaches can be used to make the running of AI models more energy efficient. And for most applications of AI, these things will be sufficient for their purposes.
And every year, new semiconductor technology will become more energy efficient in terms of energy used per unit of compute.
But these things do not mean that AI-related electricity consumption will decrease. More energy-efficient AI technologies will only increase the use of AI and overall energy consumption.
So yes, AI models and semiconductor technology are in a constant state of becoming more energy efficient, but this will ultimately end up increasing overall demand for AI products and services, resulting in higher overall energy consumption.
I very much enjoy reading the editorials on the various questions that come into your service. I have a question related to Elon Musk’s planned future flight to Mars. He indicates that he will send some of his robots on this trip.
What environmental challenges will these machines face as they are deployed on the Mars surface? I will be interested in seeing how they may survive in the Mars climate, that is if Starship lands and is still functional
– John B.
Hi John,
What a great, practical question.
Obviously, humans will still need to wear some kind of space suit to provide oxygen for breathing and temperature control, and also to pressurize the suit. But what about humanoid robots, which will be of incredible utility in constructing and maintaining a presence on Mars, or the Moon for that matter?
One of the biggest environmental challenges is the large swings in temperature on Mars. Temperatures do depend, of course, on where on Mars you’re located. For example, on the equator of Mars, daytime temperatures are a balmy 70° F (20° C), and at night will typically drop below freezing. But at the poles of Mars in the winter, temperatures will drop to -220° F (-140° C).
With that said, the stresses of gravity are less. Mars’ gravity is about 38% of Earth’s gravity, which means on the surface, we would feel like we weighed 38% of what we weigh on Earth. Imagine how much higher we could jump on Mars?
The surface of Mars is also very sandy/dusty, which is always an issue with machinery of any kind.
There are a number of things I expect will be done to modify Optimus, or any other robots, to endure Mars’ environmental conditions. Optimus will need…
In addition to the above, there would have to be some decisions made about power sources as well, such as considering lithium-ion batteries that need regular recharging versus something like a radioisotope thermoelectric generator (RTG) that generates electricity using the heat from the radioactive decay.
This might sound like a lot, and it is, but all this technology exists today. There’s no reason Tesla, or anyone else, couldn’t have humanoid robots ready to travel to Mars by the late 2028/early 2029 launch window.
And who knows? Musk might just surprise us all by sending up a few Optimus robots to Mars on a Starship in late 2026 (Nov–Dec) as a test mission that might carry a payload of critical materials needed for a future manned mission to Mars in late 2028.
And I’m predicting that’s exactly what Musk and his team at SpaceX and Tesla will do.
Dear Jeff,
Have you heard about this class action lawsuit in re: Tesla overstating its electric cars’ battery life and self-driving capabilities in a class-action lawsuit?
Is there a similar action occurring in America? What impact will this have on Tesla’s claims of fully autonomous self-driving capability?
Regards,
– Roberto C.
Hi Robert,
While I’m not specifically familiar with the class action lawsuits in Australia concerning Tesla, I can pretty much guarantee they are a subset of the class action lawsuits against Tesla in the U.S. market.
I did a quick check, and I believe that there are six open class action lawsuits in the U.S. against Tesla:
As we can see, four out of the six are based in California where there have been strong political motivations and attacks from the state against Tesla. The illogical, anti-growth, anti-innovation policies levied by the California government led to Tesla moving its headquarters from California to Texas at great expense to the state (in terms of tax revenues).
It’s also worth noting that there are law firms that specialize in class action lawsuits. It is literally their business model. And they almost always target large companies with a lot of cash.
The lawsuits drag on for years, and the “game” is to get a settlement that is profitable for the law firms. Typically, the settlements result in small payments for consumers, and the companies being attacked often settle just so they can stop spending on legal fees to defend against the lawsuits.
Simply put, the business of class action lawsuits is a very big business.
Apple (AAPL) is dealing with them, Alphabet (GOOGL) is dealing with them, Microsoft (MSFT), etc. There are currently seven open class action lawsuits against General Motors (GM) right now. If we pick a big company with a lot of cash that has products/services for consumers, we’re likely to find class action lawsuits.
From my perspective, Tesla (TSLA) has been very clever and smart about managing the rollout of its Autopilot and self-driving technologies. It has always been clear that its full self-driving (FSD) software is “supervised” and requires the full attention of the driver to be ready to take control of the steering wheel at all times.
There used to be a steering wheel “nag” every couple of minutes that required the driver to touch/tug on the wheel to confirm that the driver was still paying close attention. But in the last six months, the camera in the cabin has been used to monitor the driver’s attention, and touching the wheel is no longer required.
As I’ve shared before, when I’m in my Tesla, I no longer drive, the car drives me. It is relaxing, convenient, and far less stressful than if I was driving the car myself. The refinements in Tesla’s FSD in the last six months have been stunning. There’s no way to fully understand how good and safe the technology is without directly experiencing it.
To answer your question specifically, I don’t believe any of the class action lawsuits will have any negative impact on Tesla’s continued development of its FSD software or forthcoming regulatory approvals to launch both a robotaxi network and “unsupervised” self-driving software.
Tesla already has the data to show that its Autopilot and FSD software is safer than human drivers, and that data will only improve over the next couple of months where I’ve predicted that Tesla will eclipse 6 billion driven miles on FSD.
We’re in for an exciting future. Musk announced that Tesla expects to launch its robotaxi network in Austin, Texas, this June. And when Tesla does that, you know where I’ll be… on my way to Austin to check out the service and enjoy some fantastic Austin barbeque.
Hi Brownstone Research,
I’m writing in about an experience I had traveling to the airport this year after a short CES visit with my sister for digital health.
We had an Uber driver and he had a lot to say. It was an interesting conversation. He told us that up to October of 2024, UBER paid out 60% of the Uber ride price to the driver. Then in October, Uber drivers got a terms of service update that changed that to just 30%. Clearly he was upset as it destroyed his ability to earn a living. He said 30% would barely cover his vehicle, insurance, maintenance, fuel, or general operating costs… He went from making a decent living to now looking for a regular job.
I told him this was likely their transition to autonomous driving…
Would you have any insight into why Uber made this change that will clearly destroy their drivers’ ability to drive for them? Was it just a money grab to drive their stock price? Was it to transition to autonomous driving?
– Kevin M.
Hi Kevin,
Thanks for writing in. I really appreciate the boots-on-the-ground research. When I’m traveling, I always talk with the drivers to get insights on the local economy and ride-hailing services. They’re great sources of real-time information.
What you were discussing with the Uber driver is known as the “take rate.” It’s the percentage of the ride-hailing fare that goes to Uber. On the surface, we’d probably think that there’s only one answer – a specific percentage – but there isn’t, which is why this issue is quite controversial.
Uber has complex formulas for the take rate depending on the type of ride, for example, an Uber X versus an Uber Black ride. Surge pricing, promotions, and regional differences can also impact the take rate.
That’s right. From city to city, the rate can be different as the costs of operating in certain regions or cities are different for Uber.
Some have made claims that Uber has increased its take rate to 40%, but Uber has firmly disagreed with that number stating that “Uber’s take rate is nowhere near 40%.” Uber’s global average take rate in 2023 was 28.7%, the key word is “average.” As I mentioned before, there is a lot of variability depending on the factors that I mentioned earlier.
I did a rough calculation for 2024 by dividing Uber’s mobility revenue by its gross bookings and the number comes out to just about 27%.
Again, these are just averages, but from the data, we can understand that what the driver told you is not accurate. He may be in a region/city and tend to a certain kind of ride that results in higher average take rates, but that’s different than having a flat policy for every ride – which Uber does not have.
Aside from this, these policies are not connected to autonomous driving.
Uber (UBER) has radically improved its operating model since Dana Khosrowshahi became CEO in August 2017. He made some radical, and obviously necessary, changes in Uber to bring the company into profitability and generate free cash flow.
Khosrowshahi was quick to spin off capital-intensive divisions of Uber that contributed very little to revenues and were a drag on free cash flow.
One of those divisions was Uber’s Autonomous Technology Group (ATG), which it “sold” to Aurora Innovation (AUR). Actually, it gave Aurora the division and also gave Aurora $400 million in exchange for 26% of the company. Uber still owns 23.4% of Aurora, which focuses on autonomous driving technology for trucking and logistics.
Uber has been active in partnering with major players in the autonomous driving space as a way to prepare for the upcoming transition. It is quite cautious in its messaging, acknowledging the need for safety that is demonstrably better than human drivers, geofencing limitations, and regulatory hurdles for autonomous driving.
With that said, this is a big month for Uber and autonomous driving. Uber announced that it partnered with Google’s Waymo in Austin, TX. Uber riders based in Austin can hail a Waymo self-driving car directly from the Uber app. This is something I’ve been predicting for years.
Google’s end game is not to own and operate ride-hailing services. It will license the technology and partner with other players like Uber to eventually take over these businesses. Google is developing an in-car operating system for data surveillance to ultimately increase advertising revenues.
And this isn’t Uber’s first foray into an operating partnership.
In December, Uber partnered with China-based autonomous driving company WeRide and launched commercial services in Dubai. It’s the largest robotaxi service operating outside of the U.S. and China.
WeRide’s Autonomous Vehicle Available Through Uber in Dubai | Source: Uber
Unrelated to Uber’s operating models for ride-hailing, the transition towards autonomous driving is already starting. Uber believes that this will be a gradual transition for the reasons that I provided earlier and that it is well-positioned for autonomous driving because of the maturity of its own business.
Uber is correct to state that autonomous fleets will require facilities for parking, maintenance, cleaning, and, in the case of electric vehicles, charging. These are things that Uber is well-positioned to provide due to its operational infrastructure and financial resources. And these requirements will also create new employment opportunities to support these autonomous fleets.
One of the more interesting dynamics of the ride-hailing trend is that passenger prices have increased by about 50% between 2017 and 2023, reflecting the spike in inflation.
The combination of both autonomous driving technology and the coming reduction in inflation will drop ride-hailing costs below 2017, dramatically widening access to individualized point-to-point public transportation.
Thanks again for the questions, Kevin. It allowed us to explore some interesting dynamics related to the autonomous driving industry.
Jeff
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
