SpaceX, AI Capex, and the Future of Tech (with Jonathan Ferro, Lisa Abramowicz, Amarie Hordern)

SpaceX shed roughly $400 billion in Monday's session, the second largest single session drop on record. Matt Witheiler of Wellington Management writes it's unique in that it can talk about a future that doesn't exist and get credit for it in their price, exactly how Tesla stock behaves.

Jonathan Ferro

Matt Witheiler

It definitely makes it difficult to bet against Elon Musk. Very few companies could go public and trade at a $2 trillion+ market cap with the current financial stats of SpaceX.

How close are we to the reusable technology advancements needed to justify this market cap?

Lisa Abramowicz

Matt Witheiler

SpaceX has delivered reusable rockets at scale, fulfilling the vision. The next step is orbital data centers, and a lot of embedded value in the stock is in that story.

Masayoshi Son of SoftBank said he doesn't see data centers in space as the likely path; it's about who wins the data center buildout on Earth in the next couple of years. As an investor in Anthropic, do you agree?

Amarie Hordern

Matt Witheiler

I do think there is a question about the long-term viability of orbital data centers. SoftBank's point is right: today we are in a compute-constrained environment for inference. There is more demand for inference compute than supply, which has driven performance in chipmakers and AI companies like Nvidia, Microsoft, and Facebook.

Is there pushback on the cost of capacity buildout? Marc Rowan of Apollo talked about AI as a risk, saying what happens if companies limit token budgets due to weak ROI. How much is this a risk factor?

Jonathan Ferro

Matt Witheiler

It's something we think about, but you have to look at what companies purchasing this capital are saying. SpaceX, Google—they are saying there is sufficient ROI to raise additional funds for buildouts. They see internal demand. Perhaps they are overspending on tokens today, but that doesn't mean AI doesn't deliver ROI across a wide variety of use cases. The proof will be in the years ahead.

Anthropic has been tracking 2000% year-over-year growth. Even if that slows to 500-700%, it's still significant.

Microsoft recently tapped DeepSeek Chinese models to offset costs. Why isn't that the path of travel for many companies?

Lisa Abramowicz

Matt Witheiler

Frontier models (OpenAI, Anthropic) will capture a lot of revenue. But not everything needs to run at the frontier model level. There is a role for DeepSeek and open-source models. The market will bifurcate: value in large language models at the bleeding edge, and value in open-source and Chinese models.

Who is SpaceX's real competitor? Anthropic or Blue Origin?

Amarie Hordern

Matt Witheiler

Probably Blue Origin. But a lot of their competition is actually the core we and hyperscalers. Their economic agreements with Anthropic and Google account for almost $24 billion of annualized revenue from leasing data center compute.

You're an investor in Anthropic. How do you characterize the messaging from leadership, which some call AI doomerism?

Jonathan Ferro

Matt Witheiler

It may be perceived as doomerism, but if you look at how this administration is acting and reports from yesterday, perhaps it's not as extreme as people thought. There was reporting that the latest Anthropic model was able to breach some classified systems, resulting in the administration putting a lockdown on the model's release.

When Anthropic goes public, will you sell the stock?

Lisa Abramowicz

Matt Witheiler

As a $1.3 trillion asset manager, we have lots of different pockets of capital. Some may be selling Anthropic, some may be buying more.

How do you think about the prospect of regulation down the road for Anthropic? Is that a story you want to navigate or take your hands away?

Jonathan Ferro

Matt Witheiler

It's a complicated piece of the mosaic. The trade-off is how much regulatory flex the government wants versus free market leadership. There was reporting that the US government may start taking positions in AI companies like they did at Intel. That could be a route.

Do political realities dictate one or the other? Is it easier to say in China we'll be market leaders, while in the US you have to control this because voters are losing jobs?

Amarie Hordern

Matt Witheiler

You can mandate being a leader, but that doesn't mean you deliver. What makes US AI development unique is multiple free market participants trying to build the best models, which has resulted in a lead in the model space.

SignalTube