SpaceX's Starship Flight 13 | The Brainstorm 142

29 min video5 key momentsWatch original

⚡

TL;DR

Starship Flight 13 proved the heat shield can survive reentry intact, de-risking SpaceX's path to full reusability by mid-2027 and potentially doubling 2036 revenue forecasts; meanwhile, frontier AI models face pressure from cheaper alternatives but command a 97-fold annual cost-decline advantage that keeps enterprises loyal.

Key Insights

Starship's soft landing proved the heat shield survives reentry without charring, letting engineers recover intact hardware for faster iteration instead of analyzing burnt wreckage.

Doubles 2036 earnings — If SpaceX achieves full reusability by mid-2027 instead of the modeled 2027 date, it could double earnings expectations for 2036 and increase 2031 earnings by 20-40 percent.

97-fold annualized decline — Frontier AI inference costs are declining 97-fold annualized, making it irrational for enterprises to switch to cheaper Chinese models when trusting an incumbent vendor and waiting one year saves far more.

Geopolitical risk prevents adoption — Deep Seek failed to dent ChatGPT or Claude adoption despite claims it would commoditize AI, suggesting geopolitical risk and switching costs keep American models entrenched despite cost disadvantages.

First orbital catch attempt — Flight 14 will be Starship's first fully orbital flight and first attempt to catch the ship mid-air—the first designed orbital vehicle since the Space Shuttle meant to launch and land repeatedly.

$2 trillion to foundation models — ARK models $7 trillion in AI software spend split roughly: $2 trillion to foundation model companies, $2 trillion to platform-as-a-service, with frontier labs continuing to capture the majority of economic value.

Want this for every new video ARK Invest posts? Brevyd summarizes each upload automatically, the morning it drops.

Deep Dive

Starship 13: Heat Shield Breakthrough

Brett and Tasha open by debating whether Flight 13 was a success—and land on yes, with caveats. SpaceX achieved a soft landing on the ship but suffered a hard booster splashdown, yet the real win was optical proof the heat shield survived reentry intact. Previous flights showed the shield charring severely, forcing engineers to reverse-engineer from burnt debris. This time it landed so softly the ship didn't explode falling over, and the shield looked undamaged. Brett flags this as the hardest engineering challenge SpaceX anticipated, now largely de-risked. They also deployed 20 satellites that stayed active for 20 minutes, gathering data, and reignited an engine in space to test that capability. The booster redesign prevented some engines from relighting for the soft splashdown, so more work remains, but the heat shield validation is a milestone.

Path to Reusability and 2027 Timeline

Brett digs into the financial implications of the heat shield success. ARK's models previously expected a fully reusable stack by mid-2027, and if SpaceX hits that sooner, the capital preserved for satellite manufacturing and production scaling could double 2036 revenue and boost 2031 earnings by 20-40 percent. The current cost to orbit is hundreds of dollars per kilogram; reusability would drop it to around $100 or less. Brett explains their refurbishment model: the first reuse might cost as much as building the ship from scratch, but time declines to roughly three months over a learning curve, eventually approaching airplane-like cadence. Falcon 9 boosters now turnaround in 20 days, but much of that is barge logistics. Starship is designed to land on the launch pad itself, eliminating that bottleneck. Brett also notes the heat shield's ultimate goal is zero refurb—just like an airplane—but that's aspirational; practical refurb will likely mean some tile replacements. Overall, this flight moves the needle meaningfully on SpaceX's unit economics.

Flight 14: Orbital Catch and Starlink Implications

Tasha asks what's next, and Brett lays out Flight 14's two major firsts: the first fully orbital flight and the first attempt to catch the ship mid-air using the launch tower. This would mark the first designed orbital vehicle since the Space Shuttle built to launch and land and launch again. The satellite implications are huge—Starship's full stack will deliver 20 times more bandwidth per flight than Falcon 9. Even if the top stage never becomes reusable, Starlink can commercialize that bandwidth advantage in short order. Brett notes ARK has modeled scenarios where the top stage never works, and Starlink still transforms connectivity delivery; the business is an amazing commercial opportunity independent of full reusability. Catching the ship would cement the entire vision.

Frontier AI: Cost Decline vs. Cheaper Alternatives

Brett pivots to AI and opens with a calculation: frontier model inference costs are declining 97-fold annualized. He uses DeepSeek as a case study—it launched with hype that it would end American lab dominance, but it's now used mostly in China and emerging markets, and has barely dented ChatGPT, Claude, or Gemini adoption. The new claim is that Chinese or open-source models will destroy frontier companies via cheaper agentic experiences. Brett's counterargument: if you can save 5 percent by switching to a Chinese model, or wait one year and get a 97-fold cost decline, why bother engineering the switch? Current frontier models score in the low-to-mid 70s on agentic benchmarks like Deep Seek's software engineering test; error rates are improving toward saturation in about a year. Brett argues that as benchmark performance approaches saturation, the calculus doesn't favor switching—it favors staying with trusted vendors and letting cost drop dramatically.

Market Size and Frontier vs. Distilled Models

Nick pushes back, arguing the frontier itself is getting more crowded and performance jumps between models are shrinking. Brett disagrees, saying benchmark y-axis scaling masks real progress in agentic tasks. Tasha adds that 85-90 percent of knowledge-work tasks can already be handled by older, cheaper models, so the question is: how large is the frontier market versus the much larger market for distilled models? Nick then reframes the debate: most work doesn't need cutting-edge capability, but business competition is zero-sum. If a competitor spends $10 on a frontier model to beat you and you spend $1 on a cheap model, you lose. Brett extends this into a power-law story—restaurants featured by AI chatbots capture disproportionate traffic, so all restaurants must hire AI agents to market themselves to chatbots, concentrating spend upward. As winners pull further ahead, losing businesses must spend more on frontier models to stay competitive or fade into a commoditized long tail. ARK models $7 trillion in total AI software spend as: $2 trillion to foundation model companies, $2 trillion to platform-as-a-service (Palantir, open-weight models), with the rest distributed below. Brett believes frontier labs will capture most of that value.

Takeaways

✓Monitor Flight 14's ship catch attempt in Q1 2025—success de-risks the full reusability thesis and could accelerate SpaceX's satellite revenue timeline by 12-18 months.
✓If your business currently uses a frontier AI model, comparing switching costs to the 97-fold annual cost decline means staying put and waiting one year will likely beat the engineering effort to migrate to a cheaper alternative.
✓Companies competing in winner-take-most markets (e.g., restaurants, e-commerce) must budget for frontier AI to stay visible to AI chatbots and agents, making cost a secondary concern to staying ahead of competitors.
✓Chinese AI models will exist as a meaningful vector but won't dominate enterprise adoption due to geopolitical risk, regulatory uncertainty, and switching costs; expect them to drive frontier labs to optimize pricing and performance rather than lose significant market share.

Key moments

0:06Cost decline for frontier AI

“The cost decline for inference at the frontier is kind of on like my most recent calculation. It's 97fold annualized.”

2:32Heat shield survival proves reusability

“the hardest engineering challenge, at least that we had anticipated, was whether or not you can make that heat shield reusable on the Starship. And from that perspective, this was a wildly successful flight.”

3:362036 earnings upside from reusability

“if they can make this fully reusable earlier, it means they can lower their costs to orbit from hundreds of dollars per kilogram to you know around 100 or 100 or less. And so then that preserves capital for the company to devote to satellite manufacturing and spinning up kind of all their production facilities.”

11:00Flight 14 targets first orbital catch

“14 is the first um fully So these up until now it's been suborbital. So this is the first orbital flight. And again they plan to try to catch the ship for the first time.”

24:42Business competition drives frontier spend

“business is generally positive sum in aggregate but zero sum in competition. So if you can spend 10 to beat your competitor and your competitor is willing to only spend 1 and get beaten, well have fun guy, you're dead.”

You just read one. Brevyd does this for every upload.

Follow ARK Invest and every new video comes back as a summary like this, in your morning briefing. No watching required.