Is SpaceX Coming For Your Phone Bill? | The Brainstorm 153

Space Exploration Technologies

Across the shows: 11 speakers in other episodes, 14 bullish and 7 bearish calls, past 30 days

• Brett Winton · ARK InvestBullish

  SpaceX's direct-to-cell plan, backed by broadband, could be highly disruptive to US telcos.

  Brett says the SpaceX plan is provocative and potentially very disruptive to the telcos. Its cellular service depends on the density of its broadband dishes in each city. Coverage therefore needs a successful broadband business locally, so SpaceX has to sell broadband first to make cellular work. It may be able to sublease bandwidth from cellular carriers as well. The shift from emergency backup at about a dollar or two per subscriber to capturing the full monthly package greatly expands the addressable market. US broadband is about $170B a year, while cellular plus broadband is expected to exceed $500B annually by 2030. On his model, SpaceX reaches about $400B by 2030 and $500B by 2032, roughly 25% of a $1.6T addressable market. Much execution is still needed.

  “The SpaceX plan is provocative, and... there's this interlinking between their, call it, broadband business and their perspective cellular business that is what makes it potentially possible and super disruptive to the telcos”at 13:55

• Sam Korus · ARK InvestBullish

  SpaceX's direct-to-cell offering could reach consumers within about four years, faster than telcos want.

  Sam expects SpaceX's direct-to-cell offering to be about four to five years away, and he thinks four years is more realistic. He points to Starship's progress in how quickly SpaceX can get hardware into orbit. This could mean SpaceX knocks on consumers' doors sooner than AT&T, Verizon or T-Mobile would like. AST SpaceMobile is partnering with the telcos, but Sam thinks SpaceX's offering will be fundamentally different.

  “I think this is a four year... Four to five year. ... this could be knocking on people's doors faster than AT and T, Verizon, or T Mobile really wants.”at 15:43

• Brett Winton · ARK InvestBullish

  SpaceX's satellite broadband capacity could rise from ~900 Tbps to ~24,000 Tbps by 2028.

  Brett expects SpaceX's throughput to rise sharply as it adds satellites and bandwidth. He expects capacity to grow from about 900 terabits per second at the end of this year to around 24,000 Tbps by year-end 2028. Customers mainly want reliability rather than peak speed, and most people never use their full bandwidth. SpaceX can therefore sell a service that is plenty for everyone, and the bigger pipes let it offer the same speed guarantee at a lower price. This should move people from using Starlink as a backup to using it as their primary connection, which in turn seeds the cellular business.

  “the thing that people really want is reliability. And if you have more satellites up and you have more bandwidth up, we think their bandwidth is gonna increase from 900 terabits per second by end of this year to on the order of 24,000 terabits per second by 2028 year end.”at 19:26

• Brett Winton · ARK InvestBullish

  SpaceX's short-term compute rentals can capture the premium on new chips, roughly $30-40B per gigawatt.

  Brett says SpaceX builds its compute and rents it on short-term deals, so it collects the premium that new chips command. Neoclouds such as CoreWeave earn only a little over $10B per gigawatt because they must sign five-year deals to finance their data centers. SpaceX could collect $30-40B per gigawatt across its fleet. Its strong balance sheet and access to large-scale financing let it avoid long-term contracts. It can also absorb oversupply by using the capacity internally for Grok and R&D, which gives it a superior monetization position, as it already does with launch and Starlink.

  “SpaceX builds and is renting short term. Right? And so they get to collect the premium that those new chips have.”at 27:00

• Brett Winton · ARK InvestNeutral

  SpaceX should keep most launch capacity for its own missions given the strong ROI on its own launches.

  Brett says that on a strict narrow economic basis, it is unclear SpaceX should give outside customers any launch capacity. The return on SpaceX launching its own payloads, mainly Starlink, is currently very strong. Most launches are for Starlink and government missions, and SpaceX devotes engineering talent to those. Starship is meant to pull the industry in, but for now the economics favor SpaceX using its own capacity.

  “for now, on a strict narrow economic sense, it's unclear that they should give anybody any capacity on that other than their own because they have such great ROI on just launching their own stuff right now.”at 30:21

OpenAI

Across the shows: 31 speakers in other episodes, 14 bullish and 28 bearish calls, past 30 days

• Brett Winton · ARK InvestBullish

  OpenAI's 722 math proofs show it is outpacing mathematicians with its staff and compute.

  Brett describes OpenAI as speedrunning all of mathematics. Working mathematicians usually spend multi-year projects on results like these and then publish them for recognition. OpenAI appears to produce them with the staff it has and a modest amount of compute time on its powerful models. Many of the proofs could lead to further discoveries in formal logic, and some may have commercial applicability. He sees the notable absence of cryptography as one possibility among others, and does not know which explanation is true.

  “OpenAI is speedrunning all of mathematics and seems to be able to do so with the people they have on staff and some number of hours of compute time with their super powerful models.”at 1:33

Cybersecurity

• Brett Winton · ARK Invest

  An AI-driven attack scrambling a systemic bank's ledger could be more disruptive than stealing crypto coins.

  Brett says crypto is not the only financial institution at risk from increasingly powerful frontier models. He describes a European bank waking up to find its accounts scrambled. A hacker could not run off with the dollars, but could cause heavy damage. Recovering the account records of a systemic institution that has underinvested in security would be very hard, and everyone would need to freeze while trying to work out where the money was and what their balances were. By contrast, with stolen crypto you know where the coins are, even if they are in the wrong hands. He concludes there is a larger nest of potential disruption in financial markets beyond crypto. He also argues that real assets with clear title records are harder to wipe out than bank balances.

  “you could imagine some European bank wakes up and its accounts have been totally scrambled. ... they could certainly cause a lot of damage.”at 9:07

• Brett Winton · ARK Invest

  Post-quantum encryption schemes are newer and less tested, so they may be more vulnerable to AI attacks.

  Brett argues that quantum-safe encryption schemes are more complex and less tested than the existing systems. Because of that, they probably contain undiscovered vulnerabilities that AI could find and exploit. The schemes people were hoping to rely on as a life raft against quantum computing may therefore have more holes than expected. He presents this as a more tenuous position than the cryptographic primitives they are meant to replace.

  “The quantum safe encryption schemes are more complex and less tested. And so if anything, they're, like, a little more tenuous here because they probably have more vulnerabilities that haven't been discovered that AI would then crack.”at 11:39

Bitcoin & crypto markets

• Brett Winton · ARK Invest

  AI poses a nearer-term risk to crypto than quantum computing, mainly through surrounding infrastructure, not cryptographic primitives.

  Brett argues there is greater risk from AI than from quantum computing to cryptocurrencies broadly. The risk is not in the cryptographic primitives but in everything securing the blockchain. Self-custody hardware wallets have had code bugs exploited to steal coins, and a supply-chain attack on Ledger broadcast users' seed phrases. Retail self-custody is becoming riskier against sophisticated attackers, so large infrastructure providers that can pay for security may be safer. Exchanges are large honeypots, and a severe exchange hack could be a disruptive event. Brett also notes that frontier capability trails the labs by roughly six to twelve months and will eventually be available to bad actors, so the industry as a whole is likely to lag but will still get there.

  “there's greater risk from AI than there is from quantum computing to kind of, like, certainly penetrating kind of cryptocurrencies broadly. But the risk is not necessarily in the cryptographic primitives, but in everything else that tries to secure your blockchain”at 2:57

Open-source AI models

• Sam Korus · ARK Invest

  Open-weight models are more likely than closed models to be used for hacking, since bad actors will operate there.

  Sam argues that if the world gets hacked, it is more likely to happen through open-weight models than closed ones. His reasoning is that nefarious actors will operate in the open-weight ecosystem, where they can use models without the access controls and monitoring that frontier labs apply. He suggests frontier labs may then sell cybersecurity packages to defend against this.

  “if the world is gonna get hacked, it's actually more likely to get hacked by open weight models than closed because, you know, those were... Will be where the nefarious actors will operate.”at 5:15

Stablecoins & crypto policy

• Brett Winton · ARK Invest

  Traditional financial instruments are moving onto crypto rails for practical collateral and transfer benefits.

  Brett says traditional financial instruments are being converted to crypto. People are doing this for practical reasons rather than philosophical ones: it makes it easier for customers to borrow against these assets and to transfer them. As a result, crypto becomes more practically useful, beyond serving as insurance against systemic collapse.

  “People are converting it over not because of philosophical conceits, but because, hey, this makes it easier for customers to borrow against these assets and to transfer these assets.”at 8:26

Quantum computing

• Brett Winton · ARK Invest

  Quantum computing is at least a decade away, and AI may erode its codebreaking appeal first.

  Brett says he is skeptical of quantum computing timelines and thinks the technology is at least a decade out, if not longer. He also argues that AI's acceleration is taking commercial opportunity from quantum. The main justification for funding quantum computers has been that they would crack cryptography. If AI can break those codes first, there is less reason to fund the quantum machine, since the same Bitcoin could be taken with superintelligent AI.

  “I'm, at least publicly and personally skeptical on the timelines on quantum computing. I think we're at least a decade out, if not more.”at 12:04

AI infrastructure & data centers

• Brett Winton · ARK Invest

  Marginal GPU rental cost rises toward $50B+ per gigawatt by 2030, while neoclouds stay near $20B.

  Brett models marginal compute rental at about $40B per gigawatt today. He projects it rising to $50B+ by 2030, growing at nominal GDP, and says there are plenty of ways it could go higher. Neoclouds such as CoreWeave earn only a little over $10B per gigawatt because they sign five-year contracts in which the first year, when chips are most valuable, carries the premium. They do this to get the guaranteed revenue that financiers need for data center capex. As a result, neoclouds may keep earning around $20B per gigawatt even as the newest builds monetize at about $50B.

  “If you're renting at the margin today, you have to pay $40,000,000,000 per gigawatt. If you're renting at the margin by 2030, you're gonna have to pay 50 plus, and that's just increasing it at the rate of nominal GDP.”at 26:30