October 19, 2026
Dear Leland and Everett,
I want to talk about Elon Musk, because as much as Elon is in the news these days, I don’t think people appreciate what is coming with Elon’s companies. That topic happens to sit at an intersection of places of particular interest, and perhaps even expertise, for me. I’m no technology expert, but I follow technology trends and have for several years. Tech company strategy, moreover, is something near and dear to my heart. Almost all MBAs fashion ourselves business strategy experts, and these days tech company strategy is the sexiest, most interesting sector. Of course, residing in Silicon Valley surrounded by tech companies, having worked for one of them, and knowing people who work for several, increases both my level of interest and my familiarity. Finally, Elon’s companies are somewhat unique in the tech world in that his companies not only operate in the world of “bits” (e.g. pure software), but also “atoms” (e.g. hardware). Apple, my former employer, is of course also a hardware company. But outside of Amazon, most other tech behemoths are purely software. Think Google, Facebook, Microsoft, and the AI companies: these are companies that primarily make digital products rather than physical ones. The physical world changes things dramatically, in ways software folks often underestimate. The practical effect is that developing a capability to produce hardware typically takes years, and once a company develops a capability in the world of atoms it typically takes aspiring competitors years to catch up. Given the size of Elon’s companies’ head starts, it’s inspiring and a little daunting to think about the type of impact his companies might have on humanity going forward.
Elon is CEO of four companies. The least interesting of these right now might be the Boring Company. I still find the company fascinating, if only because the Boring Company is, by far, the cheapest and most efficient company in the world at digging underground tunnels. They have built several operational tunnels successfully, most notably a tunnel that facilitates traveling from the Las Vegas airport to various parts of the city rather efficiently. Elon’s vision for the company was to dramatically enhance traffic efficiency by adding underground tunnels to the transportation network; unfortunately he needs municipalities to engage, and uptake has been disappointingly slow. Still real, but slow.
I hesitate to call Neuralink the second most boring of Elon’s companies; perhaps it’s more accurate to call it the company with the least capacity for broad societal impact in the near term. For the niche of people it serves, the impact looks to be very big indeed. What Neuralink does (particularly what it hopes to do) is somewhat hard to describe. For now I’ll stick with what it does: within the last few weeks Neuralink enabled an ALS patient who was losing the capacity to talk (he had become hard for loved ones to understand) the ability to “speak” via a computer interface. Neuralink placed an implant in his brain; that implant, after some training, is able to read his brain signals for speech and relay them to a computer interface that “talks” on his behalf. His first words, spoken to his wife, were “I love you”. Neuralink, like all Elon companies, has audacious ambitions; for now I think what they are pulling off is promising enough on its own.
I’m not sure which of Elon’s companies will be the more transformative, but I’ll talk about Tesla next. After many years of ogling, I finally purchased a Tesla almost exactly two years ago (to your delight: you both love that car). What finally convinced me to pull the trigger (purchasing a vehicle for only the second time in my life) was the self-driving technology. I couldn’t believe how well it worked. Don’t get me wrong, there were still plenty of hiccups. The car particularly struggled with pedestrian traffic (thankfully, erring on the side of caution), and every few miles just made mistakes that required me to override. What I did not anticipate was the pace of improvement: every month or so Tesla rolled out updates; each of these noticeably improved the self driving, and reduced the number of times I needed to intervene. Now I almost never intervene, other than because I prefer to take a different route. At this point I say with some confidence that Tesla self-driving is a better driver than I, which is saying something because I consider myself a pretty good driver. About the only thing I consistently do better than Tesla these days is anticipate uncomfortable bumps from road damage like potholes…and my understanding is that Tesla is working on this specific problem as we speak.
On some level, Tesla’s pace of improvement makes sense. It took years for Tesla to write enough code to get the cars to work most of the time in self-driving mode. Once they got over that hump, however, things began to snowball. Better functionality convinced more people (like me) to use self-driving. More people using self-driving allowed Tesla to gather more data. In particular, Tesla was able to focus on instances where drivers overrode the self-driving, taking each one of these as an opportunity to improve. They were able to contrast what the car would have done with what the driver wanted (as evidenced by the override), and using a neural net (a type of AI that tries to mimic how humans learn, less by being prescriptive and algorithmic and more by compiling contrasting examples of success and failure, and allowing the processor to infer which actions are most appropriate dynamically), consistently improved precisely the situations where drivers like me were most likely to take over. And so, now, like I say, the car is an excellent driver. I recently drove my Tesla to San Francisco and back for a dinner reservation your mom and I had with friends. I drove out of our neighborhood (I enjoy neighborhood driving, for whatever reason) and set the self-driving before I got on the freeway (less than half a mile from our house). The car got on the freeway, drove ~40 miles on the freeway, exited, then navigated residential and then downtown San Francisco until we pulled up to the restaurant where we were eating (at which point I intervened because I didn’t want to pay the exorbitant parking garage fees). Coming home, same phenomenon: I backed out of my parking space, turned on self-driving as I made my first turn, and then let self-driving go, uninterrupted, until we pulled into our neighborhood.
Tesla’s self-driving advancements allowed the company to start producing CyberCabs, a fully self driving car that lacks even a steering wheel. The idea is that customers can use CyberCabs for their own purposes, rent them out like Ubers (or, more accurately, Waymos), or do some hybrid of both. CyberCabs are now hitting Austin, and reviews sound pretty positive.
You both are already familiar with Waymo, but for context: Waymo started in San Francisco, and has become quite popular there. Waymo is a startup incubated at Google; they too have created self-driving cars. Unlike Tesla they’ve never offered anything other than self-driving cars, and my understanding is that they own the fleet of cars (rather than selling to customers, like Tesla intends with CyberCab). By all accounts Waymo is quite effective at self-driving; truly, customers kinda love it. The problem I see is intrinsically related to what people have long perceived as Waymo’s greatest strength, which is its reliance on LiDAR. I forget what the LiDAR acronym stands for, but it basically sends out rapid pulses of laser light, and then uses the reflection time and angle to compile a remarkably accurate view of its surroundings. LiDAR is far more precise and sophisticated than the simple cameras Tesla uses, and so people long assumed Waymo would “win” based on the superior technology.
The problem with LiDAR is the cost. The basic equipment is expensive, and retrofitting a vehicle to incorporate the cameras even more expensive. My understanding is that older Waymos cost something approaching $200k to build. Precise estimates are hard to find, but current production costs might be in the $100-150k range. A newer model is apparently in the works that aims to cut the cost down to ~$75k/vehicle. The problem is that the $75k is the cost to build, whereas Tesla says people will be able to purchase CyberCab for $30k (and that includes Tesla’s profit, which isn’t incorporated into the Waymo $75k cost).
What about Tesla’s inferior camera technology? Well, Tesla decided that they could use software to decipher imperfect camera images in much the same way humans learn how to decipher imperfect images with our eyes. Humans develop all sorts of tacit algorithms for how to deal with darkness, the glare of sunlight, or blind spots. Tesla realized they could just mimic those same learning behaviors, and even improve upon human processing by using software to constantly monitor several cameras at once, without suffering from fatigue or lapses in concentration like a human.
So now, think about it: you will soon be able to purchase a car (Tesla says for ~$30k, but I won’t be surprised if they start more expensive before costs come down with production volume) that you can either use for yourself or defray the costs by renting it out to others when you are not using it. Teslas require less maintenance than cars with internal combustion engines (they have far, far fewer parts, and don’t require oil changes, new spark plugs, or really much beyond tire maintenance), so the cost per mile of owning (or renting) a Tesla becomes far, far cheaper than any vehicle on the road once people start sharing them. Waymos, for all their virtues, generally cost more than renting an Uber or taking a taxi; Cybercabs will cost far, far less.
And then there are the robots. Tesla is currently building humanoid robots called Optimus. The idea is that Tesla will use neural nets, like in the cars, to allow humans to train Optimus in a manner similar to how we train humans. Basically, we will demonstrate what it is we want Optimus to do, and then correct mistakes. The difference between Optimus and a human: once we teach one Optimus how to do something, we will have (with some time lags) effectively trained every Optimus with an updated software package how to perform the same activity. So, much like with Tesla’s cars, I anticipate Optimus will learn and adapt rapidly, becoming a remarkably useful tool surprisingly quickly. Supposedly, Tesla is building robots for use in their own factories first, and then will roll out for public sale later. Elon pitches the Optimus as an in-home helper, capable of cleaning and cooking and doing laundry. Personally, I think Optimus is far more interesting as a factory worker. Right now we primarily use Chinese labor to build the products we use because Chinese labor is cheaper than American labor. Well, Chinese labor isn’t as cheap as robot labor. So, pretty soon, I think it will make more sense to build products closer to home and save the shipping costs than it will be to build those products in China.
Before I get to SpaceX, Tesla and SpaceX are actually working on a collaboration called Terafab. You guys are probably too young to know that chips are a bottleneck in all types of products these days, and in fact contributing to cost increases on a wide variety of product lines. Everything from iPhones to cars (all cars, not just self-driving cars), and even most appliances, are built with chips these days. Complicating matters is an explosion in demand for compute capacity from AI data centers. This explosion in demand is reestablishing a bottleneck in silicon chip manufacturing. Chip manufacturing is insanely capital intensive, requiring multi-billion dollar facilities that take years to build before chips can be manufactured at scale. Understandably, chip manufacturers are careful about how much capacity they install, lest they spend tens of billions of dollars that go wasted due to an economic downturn leading to depressed demand. So these bottlenecks could last for several years. Elon’s solution is Terafab, where the companies aim to dramatically increase global chip capacity – potentially even doubling global chip supply if they pursue and achieve their most ambitious targets. Doing so will allow Tesla and SpaceX to secure the chips they need for the foreseeable future, and may well enable them to sell off excess capacity to other customers. The result is likely a decrease in chip prices, and thus the goods that use said chips (roughly, all electronics we buy). Looking at company valuations already attached to chip design (NVIDIA is worth roughly $5 trillion) and cutting edge manufacturing (TSMC is worth $2 trillion) starts to put Terafab’s potential scale (and, indirectly, impact on humanity) in perspective.
Which brings us to SpaceX. The first thing to understand about SpaceX is that they are approaching 90% of global payload. Meaning, if you add up NASA, China, Russia, and all other competitors combined, they collectively account for 10-20% of global payload capacity to SpaceX’s 80-90%. With Starship coming online, SpaceX capacity is likely to increase dramatically in the near future, which means they will meaningfully increase global payload capacity going to space (and, presumably, increase their market share to or even above 90%).
SpaceX achieved this feat by figuring out how to make (and recover) reusable rockets. Every other entity launching rockets (at least until Jeff Bezos’ Blue Origin solves reusability), uses a rocket one time, and then disposes of it after a single use. Reusability dramatically decreases the cost: imagine how expensive air travel would be if airlines scrapped each airplane after a single flight. As a result, SpaceX is able to relaunch rockets over and over, at dramatically lower cost than anyone else in the industry.
Lower prices and increased payload changes the calculus of what can get launched into space. Historically, entities like NASA focused maniacally on mass and volume, because each gram and cubic centimeter increased the cost of a mission astronomically. As SpaceX lowers the cost of space travel by 90% and more, more entities can afford to launch items into space, and become more experimental with what they launch.
Rather than wait for the industry to decide what they might do with the increase in capacity, SpaceX came up with their own solution. For years SpaceX has put satellites into space that now allow basically global high-speed internet without meaningful dead zones. Now, timed with the much larger Starship, SpaceX is launching its next generation of satellites into orbit. In the first launch, SpaceX deployed 26 satellites which mark the beginning of a deployment whereby SpaceX aims to more than double global bandwidth over the next year or so.
Perhaps more interesting, SpaceX purchased additional spectrum bandwidth that will enable the company to offer cellular service. Pretty soon (and perhaps as soon as next year) customers will be able to choose SpaceX as their cellular provider instead of Verizon, AT&T, or T-Mobile. Unlike those other carriers, SpaceX will offer cellular service without towers and, presumably, without the corresponding dead zones.
Some of SpaceX’s most audacious goals have to do with AI. Despite starting long after industry leaders OpenAI and Anthropic, SpaceX has mostly closed the gap in performance with their Grok AI engines. And while the AI arms race continues, SpaceX is taking some novel approaches. First, they target to leverage their launch capabilities to build data centers in space. Data centers in space should be quite a bit cheaper to maintain than data centers on earth (cooling will not be an issue in space, and solar cells should be more efficient outside earth’s atmosphere). If they work, they could give SpaceX a cost advantage. But the bigger bet seems to be in compute. SpaceX appears to be betting that one of the key bottlenecks to AI will be compute capacity. OpenAI and Anthropic outsource compute to companies like Microsoft, Amazon, and (counterintuitively) Oracle. SpaceX appears to be betting either that those companies are not installing enough capacity or, if they are, that SpaceX can do so more cheaply by way of vertical integration. I’ll be totally transparent, I’m not totally sure I understand the strategy nor the implications, primarily because I don’t understand the direction of the AI industry very well. All I know is that, like Elon’s companies do in all other industries, SpaceX is making a bold and potentially differentiating bet in AI such that, if it pays off, they are likely to have a huge competitive advantage in a hugely valuable industry.
Which starts to get at the complication I see for Elon. His companies are already worth something like $4 trillion combined. If I am right about the direction those companies are headed, they are likely undervalued and headed toward much larger values. Their products are likely to impact us to a degree we’ve rarely experienced and amass more concentration of value than we’ve perhaps seen before. Having any one person become so rich and, by extension, so powerful and influential is enough to make even Elon’s fans a little queasy. And Elon already has lots of enemies, to the point where lots of folks instinctively dislike Elon’s companies and his companies’ products just because of their association with Elon. I anticipate the backlash to Elon to become much, much more visceral and vitriolic, to the point where I hope Elon is traveling with a fair amount of security these days.
Of course, that perhaps leads us to X, the social media platform that now sits under the SpaceX corporate umbrella. I perhaps have the least to say about X, only because it (so far anyway) seems the least revolutionary to me. Elon appears to have a vision for X to become something like America’s answer to the Chinese app WeChat. WeChat serves as a social media platform, but also incorporates financial transactions, product sales, and games, such that it becomes almost something of its own operating system (such that Chinese users can spend more time in the WeChat ecosystem than they do in the rest of their phone app ecosystem). Elon may have grander ambitions for X than I appreciate, but mostly X ownership appears to be something of a defensive position for Elon. So many media and social media outlets are hostile toward Elon that owning X grants Elon a platform to appeal to humanity directly, without media filtration or social media algorithmic manipulation. All I can say, as a user, is that I’m glad Elon purchased it, because X is much more enjoyable to use now than it was before his purchase.
Oh, and of course then there’s DOGE, where it appears Elon spearheaded an effort that survives without him. That effort apparently reduced the size of the federal workforce by ~12% while also reducing contractors and outsourcing costs, without obviously collapsing the federal government’s capacity to function.
Elon is just a remarkable human being. He’s already all over the news, and will continue to be so. If anything, his footprint in our awareness will increase over time rather than decrease. When you read this, you can be the judge of how my predictions played out. Perhaps more valuable, you can see a snapshot of how things looked (at least from my vantage point) in a way that puts what happened next in context.
In the meantime, I expect the coming years and decades will be a wild ride as a result of technological change, in no small part due to Elon’s companies. Buckle up.
Love,
Dad
