Topics Pioneers Insight Method Author
Back

SpaceX, Palantir, Anduril & Onodrim | The Return of Conglomerates

·Sourcery

Rendering the long image…

Topics Cross-Sector Frontiers: Tech and Healthcare Next-Gen Disruptive Computing Architectures AI Infrastructure Investment and Due Diligence Early-Stage VC Investing Dynamics Specialized Edge AI Reshaping Defense Tech Venture Capital Bets Big on Hard Tech
Pioneers Christian Garrett

Summary

  • Garrett’s SpaceX thesis now rests on a vertically integrated hyperscaler with several distinct growth engines. He cites roughly 160% Q3 revenue growth, $4.4 billion in net income, $8.8 billion in EBITDA and compute paybacks below 12 months; at a $2 trillion valuation, he puts the business around 20–30× EBITDA. Owning applications, models and infrastructure offers flexibility across changing AI economics: “You probably want to own the vertically integrated players.”
  • Starship’s investment significance extends from cheaper launches to enabling orbital compute and new businesses. Garrett expects orbital compute to succeed, citing initial satellite launches in the “back half of next year, I believe,” while cautioning that timelines cannot be forecast accurately. He expects Starship to reduce launch costs by an order of magnitude, with full reusability still the next major economic milestone.
  • AI infrastructure increasingly rewards financing advantages alongside engineering execution. Deployment speed, payback and cost of capital determine whether companies can fulfill enormous spending commitments; Garrett acknowledges genuine uncertainty about revenue, pricing and future capacity. His firm has invested in Cognition and just co-led Hex’s round; he says, “I don’t think I have a better-informed view than someone else.”
  • Duration and concentrated ownership matter more to Garrett than one exceptional growth year or a modest entry premium. He contrasts SpaceX’s roughly 40% compounding over 20 years with investors’ fixation on annual growth, and would accept a 25% premium for a quality asset offering a 5× outcome. Excess capital can still produce graveyards: “The power law will probably pay for that amount of capital,” but individual investors must own the winners.
  • The conglomerate thesis depends on shared technology and commercial demand supporting defense capacity. Anduril’s Lattice connects products operationally, while Garrett sees Palantir and SpaceX as broader models spanning commercial and government customers. His historical shorthand is that General Mills “made cereal and weapon systems”; commercial scale can sustain capabilities that industrial policy alone cannot economically maintain.
  • Monodrom targets Europe’s defense suppliers because funding and contracts have outpaced production capacity. Garrett contrasts roughly 4,000 European drone companies with insufficient support for components, engines and manufacturing. The acquisition platform plans long ownership, management partnerships, technology deployment and capital investment; its first acquisitions and major partnerships remain forthcoming. “Now we need to execute.”
  • Industrial technology can improve difficult physical processes without automatically producing software economics. Garrett points to Applied Intuition simplifying vehicle electronics and General Matter introducing modern tools into uranium enrichment. His underwriting caveat survives the enthusiasm: “Just because you deploy software into a business doesn’t necessarily make it a technology business.”

Deep dive

1. SpaceX’s AI surge sits inside a longer compounding machine

  • After 16 years invested, SpaceX remains Garrett’s largest position. He cites Q3 revenue growth of roughly 160%, $4.4 billion in net income and $8.8 billion in EBITDA. At a $2 trillion valuation, he places it around 20–30× consensus EBITDA, mentioning approximately 27×. His excitement rests on dominance across several critical industries, rather than one revenue stream.

  • O’Shea presses the identity question: more than 50% of revenue reportedly comes from AI “in some capacity,” despite SpaceX’s origins in launch and exploration. Garrett sees businesses moving through different S-curves: AI is expanding rapidly from a smaller base, while launch and Starlink have their own runways. Revenue mix is a snapshot inside “a multi-decade or century view.”

  • The hyperscaler argument combines infrastructure execution with monetization across applications, models and compute. Garrett points to Cursor and Grok chatbots, alongside software and services being built around the compute stack. Reported paybacks below 12 months signal immediate demand exceeding supply. Owning multiple layers, he argues, provides exposure whether open-source deployment expands or closed models retain their lead.

  • His downside protection comes from flexibility: if external demand slows, capacity can shift toward internal products; fast construction also permits shorter investment cycles and faster retrenchment. He calls the risk “nominal” when infrastructure can serve the company itself. That is his conviction, alongside an explicit limitation: forecasting demand four or five years out remains difficult, even with strong current utilization.

2. Starship’s cost curve connects launch, connectivity and orbital compute

  • Garrett says Starship launched payloads that delivered Starlink V3 satellites, which are delivering compute. He claims a launch carrying only a handful added roughly 1% of global internet capacity, against plans to deploy tens of thousands. That scale underpins his argument that connectivity remains early in its development, even while AI temporarily dominates the growth narrative and revenue mix.

  • O’Shea brings the orbital-compute question back to economics after visiting Starcloud and seeing successive satellite generations under development, with Starcloud 3 intended for Starship. Garrett cites initial orbital-compute launches in the “back half of next year, I believe.” He expects success, but separates that conviction from precision on timing or capacity: complex engineering schedules cannot be reliably forecast.

  • The concrete precedent is Varda: lower launch costs enabled a business serving pharmaceutical and defense customers. Garrett expects Starship to lower costs by an order of magnitude, potentially enabling a corresponding expansion in companies built around space infrastructure. His forecasting constraint matters: several business models now important to SpaceX were not even contemplated one to five years earlier.

  • Full reusability is the next economic milestone, rather than his threshold for a viable launch vehicle. Today’s Starship can launch and deliver payloads, making it a “fully functioning launch platform” against competitors that also lack complete reusability. He expects continuing changes in size and performance over five to ten years; current vehicles are early iterations, not the final platform.

3. AI underwriting turns on financing and independence from individual labs

  • O’Shea asks whether recent financing stress exposes vulnerabilities in the hyperscaler and neocloud buildout. Garrett distinguishes macro volatility and market positioning from genuine questions: can revenues support spending commitments, and can companies obtain the capital needed to fulfill them? Debt is an efficient funding source, making “speed and payback period” and financing costs central competitive variables. Lower-cost borrowers can absorb demand on better terms.

  • Asked which parts are most vulnerable, Garrett declines a definitive ranking. Model pricing, open versus closed deployment, compute pricing, power access and capital costs can all change the eventual economics. Valuation depends on earnings beyond the immediate surge, yet even two-year forecasts are difficult. “You don’t really know what market dynamics will be like in the next three months, let alone the next three years.”

  • Garrett says his firm has invested quite a bit in Cognition and just co-led Hex’s round. They sit at the harness and application layers, where customers can route among models instead of depending on one lab for coding, security or analytics. He expects substantial open-source deployment. Avoiding other categories reflects difficulty underwriting durability, rather than dismissal; he calls OpenAI and Anthropic two of a generation’s most impressive businesses.

4. Long duration can justify premiums without eliminating valuation risk

  • Garrett’s growth filter is duration and quality. SpaceX, he says, compounded around 40% for 20 years; 40% in isolation is good, but sustaining it is extraordinary. His simpler illustration is 9% annual growth over 30 years producing roughly 13×. Minority investors cannot control how long they remain invested, so he wants businesses he would still welcome owning decades later.

  • O’Shea’s pushback concerns the capital train: Cognition at $48 billion, with another round rumored around $60–70 billion; Ramp at $60 billion; Anduril at $90 billion. Are valuations identifying clear winners or merely financing momentum? She also relays Index’s willingness to pay a 25% premium while reconsidering a 2× price difference. Garrett’s answer is that operating growth and capital-driven scarcity both matter.

  • Invoking his godfather Teddy Forstmann, Garrett would rather overpay modestly for quality than secure a discount on weakness: a “quality asset that’s a 5X” at a 25% premium beats a poor asset at a 25% discount. Ramp supplies his operating case: consolidation of financial back-office workflows, continuing product expansion and services that remain underpriced or insufficiently monetized leave further runway.

  • The COVID fintech boom is his warning against confusing category success with universal success. Many companies died, while Stripe and Revolut became $100-billion-plus opportunities and Ramp continued toward that scale; he also cites Nubank. “The power law will probably pay for that amount of capital.” Growth-stage returns remain concentrated in a handful of companies, which shapes his firm’s portfolio construction.

5. Shared technology changes the conglomerate’s economics

  • Garrett distinguishes today’s conglomerates from the disparate subsidiaries that helped discredit the model in the 1980s. Older groups offered diversification, scale and financing advantages, but often lacked operational synergies; weaker divisions dragged on stronger ones and complicated valuation. The newer model rests on a common technology platform, vertical integration, or both. SpaceX combines separate industry exposures with infrastructure shared across the business.

  • Anduril is his “lowercase-c conglomerate”: multiple products, systems and capabilities within a sector. The differentiator is Lattice and shared software and AI infrastructure, alongside extensive technology use in operations. Each additional product can increase the value of the others. That relationship, rather than simply assembling more defense assets, was central to his original investment thesis and expected economies of scale.

  • His larger ambition is commercial-defense integration. General Mills “made cereal and weapon systems”; Chinese defense primes, he says, get 60–80% of their revenue from commercial activity. Palantir’s latest earnings showed majority-commercial revenue, while SpaceX also derives most revenue commercially. He sees those two as the clearest existing examples of businesses reaching the scale needed to sustain critical defense capabilities.

  • Shipping is the decisive analogy: China’s commercial shipbuilding dominance supports its defense production. Garrett argues that industrial policy alone cannot maintain all necessary capacity; commercial demand supplies scale, supports otherwise difficult economics and permits defense output to expand when needed. He expects markets to value diversification and sees opportunity in commercially dominant businesses crossing into defense, including beyond software and cybersecurity.

6. Monodrom targets the suppliers beneath Europe’s crowded defense startups

  • Garrett describes Monodrom as an “aerospace and defense scale-up platform” acquiring and developing European businesses. The US precedents include TransDigm, Teledyne, HEICO and Honeywell. Europe has mature aerospace capabilities around Airbus but, in his account, lacked comparable supplier consolidation because defense was not a growth market. Fragmentation across national markets makes independent suppliers’ expansion and technology adoption harder.

  • The neglected opportunity sits below heavily funded primes and neoprimes. Garrett thinks the emerging winners already have substantial backing and, often, sovereign advantages in countries with meaningful budgets. Against roughly 4,000 European drone companies, he calls the category “overfunded” and “too crowded.” His question is who supplies their components, engines and aerostructures—and whether those suppliers can increase production.

  • Monodrom intends to acquire businesses, hold them long term and partner with existing management. Support includes capital expenditure, engineering for new products and technology to modernize manufacturing and operations. Dual-use demand matters: commercial business can help sustain capacity between defense orders. The ambition is to turn smaller companies into pan-European, potentially global champions, rather than rely on government spending alone.

  • Selection starts with durable revenue, low churn, critical capabilities and identifiable production bottlenecks. CapEx risk, outdated operations and capacity moved out of Europe can prevent contracts becoming deliveries. Garrett cites a team spanning private equity, government, aerospace and technology, with people from nine nationalities. First acquisitions and partnerships with leading commercial and defense companies remain forthcoming: “We’re very early” in proving the thesis.

7. Becoming a co-founder makes investor usefulness more concrete

  • Garrett did not set out to become an operator. European relationships and exposure to both US and European defense companies drew him toward the production problem, then toward helping assemble Monodrom. He describes himself as a minority co-founder supporting the people building it, rather than its operating leader. “It’s not what I think is my superpower skill set.”

  • Company building supplies depth that venture’s breadth and intermittent immersion do not. Garrett says it has made him a better investor and more aware of investors’ anxious, sometimes unhelpful behavior, while acknowledging that his experience differs from being CEO. His partner Justin’s formulation captures the boundary: “A bad investor can destroy your company. A good investor won’t make your company.”

  • The most useful assistance is often straightforward: “It turns out you need revenue and talent.” Investors frequently promise more help than they deliver, without necessarily acting maliciously. Experiencing fundraising from the company’s side has improved Garrett’s understanding of what founders actually need, which questions reveal the real difficulty, and when investor intervention adds pressure rather than solving a problem.

8. Industrial technology improves physical systems without guaranteeing software margins

  • Applied Intuition illustrates a deeper intervention than selling engineering software. Garrett describes work simplifying vehicle processing units, controllers and wiring, improving cost structures while enabling better maintenance, cabin experiences and autonomy. The company has expanded beyond automotive into construction, mining and trucking. The attraction is technology changing the underlying vehicle architecture as well as the tools used to design it.

  • General Matter supplies another example: rebuilding uranium-enrichment capacity while applying modern technology to processes often operated by state-owned organizations using Cold War-era systems. Even adding sensors can create opportunities. Garrett pairs that enthusiasm with admiration for earlier engineers: “They were definitely smarter than us.” New tools can improve efficiency and performance without diminishing the sophistication of what already worked.

  • Onodrim has increased his conviction in industrial software, from paperwork automation to faster integration with PLCs as AI improves. He credits Palantir’s earlier deployment into legacy manufacturing and cites Hadrian among companies extending the approach. But operational improvement and investment economics remain separate questions: “Just because you deploy software into a business doesn’t necessarily make it a technology business.” Margin and outcome implications vary.

9. Political weaponization threatens the cooperation Garrett values in technology

  • Garrett sees technology’s non-zero-sum culture as an advantage worth protecting, particularly as the industry expands into manufacturing, defense and energy. He acknowledges the pessimistic explanation: high growth may create that behavior more than culture does. Nevertheless, petty internal fighting worries him. “Maybe I’m naive,” he allows; he does not think the industry has already lost the characteristic he values.

  • His political concern applies to both sides: judging companies through founders’ affiliations can overlook politically mixed teams, economic contributions and national-security capabilities. He cites burning Tesla vehicles and hostility toward Anthropic as examples of the direction he dislikes. His preferred feedback mechanism is product choice: people can refuse to buy a car or use a model without attacking the company or its leadership.

  • The position comes with an explicit reservation: greater overlap between technology and politics may require reconsidering the framework. “Maybe I’m wrong.” He nevertheless wants consumer and policy decisions grounded in products and business merits, rather than participants’ politics. His warning remains categorical despite that uncertainty: attacking the industry from outside for political reasons “will backfire,” while internal hostility erodes a valuable cooperative culture.