TMT Breakout

TMT Breakout

TMTB: TomorrowX Highlights - SpaceX on Compute, AMD’s Lisa Su, Fireworks on Inference, Druckenmiller on Positioning & More

TMT Breakout's avatar
TMT Breakout
Nov 18, 2026
∙ Paid

Great inaugural showing from TomorrowX. The vibes in Austin tomorrow should be even better if semis hold their AH gains. Some great panels today…

Table of Contents

  1. SpaceX: Hosting economics, customer prepayments, orbital compute and direct-to-cell
    Bret Johnsen (CFO, SpaceX).

  2. AMD: Lower software switching costs, the $220B CPU opportunity and Helios deployments
    Lisa Su (Chair & CEO, AMD).

  3. Fireworks: $1.5B revenue run rate, specialized inference and the case against commodity compute resale
    Lin Qiao (Co-founder & CEO, Fireworks AI).

  4. Druckenmiller: Capital preservation today, AI applications and task-based monetization next
    Stanley Druckenmiller (CEO, Duquesne Family Office).

  5. Altimeter: Anthropic’s inference economics, IPO positioning and concentrated model-layer value
    Brad Gerstner (Founder & CEO, Altimeter).

  6. Apollo: Financing AI without underwriting the same risks as the equity investor
    Marc Rowan (Co-founder & CEO, Apollo Global Management).

  7. Dell: 6,500 AI-factory customers and why enterprise returns extend beyond lab revenue
    Michael Dell (Chairman & CEO, Dell Technologies).

  8. Meta: Securing permission to build, industrial smart-glasses pilots and AI safety governance
    Dina Powell McCormick (President & Vice Chairman, Meta).

  9. TomorrowX Founders: Why a slower buildout could extend the cycle, and how retail changes information distribution
    Gavin Baker (Managing Partner & CIO, Atreides Management); Antonio Gracias (Founder, CEO & CIO, Valor Equity Partners); Ron Biscardi (Co-founder & CEO, iConnections).

  10. Goldman Sachs: Application-layer value, fewer routine coding roles and reinvesting AI efficiencies
    David Solomon (Chairman & CEO, Goldman Sachs).

  11. Sound Ventures: Anthropic conviction, distribution as differentiation and private-market access
    Guy Oseary (Co-founder, Sound Ventures).


1. SpaceX: Hosting economics, orbital compute, financing and direct-to-cell

Speaker: Bret Johnsen, CFO, SpaceX.

Summary: Johnsen said SpaceX is deploying terrestrial compute at costs in the low-to-mid $30s per watt versus his estimate of $50–60 elsewhere, with hosting paybacks below one year. He detailed plans to fund expansion through operating cash flow, customer prepayments, vendor payment terms and debt, including a $40B GPU financing facility. On orbital compute, SpaceX plans initial satellites in 2027, a constellation ramp in 2028 and meaningful gigawatt-scale capacity in 2029.

1. Terrestrial compute costs and vertical integration

Johnsen said SpaceX’s first couple of gigawatts of terrestrial compute average costs in the low-to-mid $30s per watt, compared with approximately $50–60 per watt for the broader industry. He expects SpaceX’s cost curve to decline further with volume, manufacturing improvements and technology advances, while power, labor and construction constraints increase costs elsewhere.

He attributed SpaceX’s advantage to vertical integration across engineering, general contracting, manufacturing and power generation. The company has spent more than a year developing its power-generation capabilities, allowing it to bring up infrastructure in months and energize installed racks within days.

As an example, an external turbine supplier indicated that blades and vanes would arrive 18–24 months after the turbines themselves. SpaceX is establishing its own investment-casting facility to manufacture those components internally, drawing on expertise developed for Raptor engines and other manufacturing operations.

Johnsen said that if SpaceX meets its previously communicated plans, it could become one of the largest AI infrastructure providers in 2027.

2. Sub-one-year paybacks and infrastructure financing

Johnsen said SpaceX’s current terrestrial hosting deals generate paybacks of less than one year, sometimes substantially less.

The company plans to fund growth through operating cash flow and improved working-capital management. That includes collecting customer deposits earlier and extending payment terms with suppliers. Because SpaceX has only been providing third-party compute hosting for roughly two quarters, Johnsen said these opportunities have not yet been fully utilized.

He described financing from three sources: operating cash flow, working-capital optimization and debt. On debt, he referenced investment-grade bonds issued in June and a $40B GPU financing facility arranged with Apollo and Goldman Sachs, with public syndication beginning November 17.

Gavin Baker suggested that customer prepayments could cause operating cash flow to exceed EBITDA materially, and Johnsen agreed. With hosting paybacks inside a year, Johnsen characterized the financing requirement as bridging months rather than years.

3. Short-term contracts, compute demand and Anthropic insourcing

Johnsen said demand for compute continues increasing across the industry. Because SpaceX has a profitable satellite and space business, it did not need to secure large long-term hosting contracts in advance to finance its buildout.

Instead, SpaceX has retained flexibility to offer shorter-term and spot-market capacity, which Johnsen said generates higher returns while preserving the ability to update hardware and redirect capacity.

Asked about Anthropic’s intention to bring more compute infrastructure in-house, Johnsen said its projected growth would require substantially more capacity than simply replacing existing rented compute. He also emphasized demand from other customers seeking available power and compute.

The short-duration hosting strategy gives SpaceX flexibility to allocate capacity toward its own software products as they grow.

4. Orbital compute: 2027 satellites, 2028 ramp and gigawatts in 2029

Johnsen described the successful recovery of Starship’s second stage after deploying 26 satellites as a major milestone toward orbital compute.

He expects rapid second-stage reusability to produce another 10x reduction in launch costs per kilogram. Establishing a regular reflight cadence over the next one to two years is central to SpaceX’s plan to make orbital compute more economical than terrestrial alternatives.

The timetable is:

  • 2027: First orbital-compute satellites launch.

  • 2028: Orbital-compute constellation begins ramping.

  • 2029: Meaningful gigawatt-scale orbital-compute capacity reaches the market.

SpaceX intends to use an architecture based on the Starlink V3 satellite platform, replacing portions of the communications equipment with GPUs, additional solar generation and deployable radiators. Satellites would operate in sun-synchronous orbits to maintain favorable solar exposure and reduce battery requirements.

Johnsen highlighted SpaceX’s existing experience with radiation shielding, radiative cooling, laser inter-satellite links and RF communications across approximately 11,000 Starlink satellites flown.

He described orbital operations as eliminating terrestrial electricity, cooling and land costs, making launch and satellite deployment the central economic considerations.

5. Grokbot, Cursor and first-party software monetization

Johnsen said SpaceX’s software portfolio provides another way to monetize its compute investments, including Grok, Grokbot and Cursor.

He described Grokbot’s growth as strong and said the company is working on cost optimization as usage increases. Cursor adds product development, sales and distribution expertise that SpaceX did not previously have to the same extent.

Baker estimated that SpaceX could be approaching $10B in first-party software revenue, explicitly identifying the figure as his own estimate. Johnsen did not provide a revenue figure but expressed enthusiasm about the software business.

On Grok 4.8 and 4.9, he indicated that releases were coming soon without specifying dates.

6. Starlink Direct-to-Cell: 5G service targeted for 1H28

Johnsen said Starlink broadband now serves more than 13 million people, while SpaceX’s longer-term ambition is to connect billions more through ordinary mobile handsets.

The initial direct-to-cell constellation of approximately 600 satellites was designed as a proof of concept and supports text messaging, including emergency communications.

SpaceX plans to begin launching its next-generation direct-to-cell satellites in 2027, targeting true 5G-quality service directly to unmodified mobile phones in the first half of 2028.

The company has acquired mid-band spectrum from EchoStar and additional low-band spectrum. Johnsen said SpaceX is working with international regulators and telecommunications providers toward global service.

Unlike conventional cellular networks, the proposed system can backhaul traffic through satellites, allowing smaller phased-array terrestrial installations rather than relying primarily on large macro towers.


2. AMD: Easier software migration, the $220B CPU opportunity and Helios deployments

Keep reading with a 7-day free trial

Subscribe to TMT Breakout to keep reading this post and get 7 days of free access to the full post archives.

Already a paid subscriber? Sign in
© 2026 TMT Breakout · Publisher Terms
Substack · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for writing, podcasting, and media founders.