Technical Deep Dive
Burry's critique zeroes in on the fundamental economics behind SpaceX and Anthropic. To understand why a trillion-dollar valuation is contentious, we must examine the unit economics and scalability of their core technologies.
SpaceX: The Rocket Equation vs. The Profit Equation
SpaceX's Falcon 9 and Falcon Heavy have revolutionized launch costs, reducing the price per kilogram to low Earth orbit (LEO) from roughly $10,000 (Space Shuttle era) to about $2,500. Starship aims to push this below $100/kg. However, the launch market itself is relatively small. In 2023, the global space launch market was valued at approximately $12 billion, with SpaceX capturing an estimated 60-70% of commercial launches. Even if SpaceX captures 80% of a $20 billion market by 2030, that yields only $16 billion in launch revenue—far from the revenue needed to justify a trillion-dollar valuation.
Starlink, the satellite internet constellation, is the real value driver. With over 2.5 million subscribers as of early 2024, generating roughly $4.2 billion in annual revenue, it shows promise. But its capital expenditure is enormous. Each satellite costs about $250,000 to build and launch, and the constellation requires over 12,000 satellites. Total CapEx is projected to exceed $30 billion before breakeven. Furthermore, Starlink faces competition from terrestrial 5G and fiber, which offer lower latency and higher bandwidth at lower cost. The subscriber growth rate is slowing as early adopters are saturated, and regulatory battles over spectrum and orbital slots are intensifying.
Anthropic: The Cost of Intelligence
Anthropic's Claude 3.5 Sonnet and Opus models are among the best in the world, rivaling GPT-4o and Gemini Ultra. However, the AI model market is rapidly commoditizing. The cost of training frontier models has skyrocketed. Training a model like Claude 3.5 Opus is estimated to cost between $100 million and $500 million, with inference costs also high. Anthropic's revenue is estimated at around $1.5 billion in 2024, primarily from API access and enterprise subscriptions. But the company is spending heavily on compute, talent, and research. Operating losses are likely in the hundreds of millions annually.
The core technical challenge is that AI models are not a high-margin software business. They are a capital-intensive, infrastructure-heavy business with thin margins, similar to cloud computing. The open-source ecosystem (e.g., Meta's Llama 3, Mistral, and community models on Hugging Face) is eroding the moat of proprietary models. Companies like Databricks and Together AI offer fine-tuned open-source models at a fraction of the cost. Anthropic's differentiation—constitutional AI and safety focus—is a feature, not a defensible business model.
| Metric | SpaceX (2024 est.) | Anthropic (2024 est.) | Trillion-Dollar Benchmark (e.g., Apple, Microsoft) |
|---|---|---|---|
| Annual Revenue | ~$8.7B (launch + Starlink) | ~$1.5B | $100B+ |
| Net Profit Margin | ~5% (launch), Starlink still negative | Negative (est. -30% to -50%) | 20-30% |
| Revenue Growth Rate | 25% | 200% (from low base) | 10-15% |
| Capital Intensity | Very High (CapEx > $10B/yr) | Very High (compute costs) | Low to Medium (software) |
| Market Cap | ~$180B (private) | ~$60B (private) | $3T+ |
Data Takeaway: The table starkly illustrates the gap. To reach a trillion-dollar valuation, a company needs not just high growth but massive, profitable revenue. SpaceX and Anthropic are still in the investment phase, with negative or thin margins. Their current valuations already price in decades of perfect execution and market dominance.
Key Players & Case Studies
SpaceX vs. The Old Guard: SpaceX's key competitor is not just Blue Origin (Jeff Bezos) or ULA (Boeing-Lockheed), but the entire aerospace industry's cost structure. Blue Origin's New Glenn has yet to launch commercially, and ULA's Vulcan Centaur is more expensive per launch. However, the real threat is from China's CASC, which is developing reusable rockets at a fraction of SpaceX's cost, and from the rise of small launch providers like Rocket Lab (Electron, Neutron) and Relativity Space (Terran R), which are targeting the small-to-medium payload market with lower overhead.
Anthropic vs. The AI Oligopoly: Anthropic's primary competitors are OpenAI (GPT-4o, Sora), Google DeepMind (Gemini), and Meta (Llama 3). OpenAI is estimated to have $3.4 billion in revenue but is spending over $7 billion annually. Google has near-infinite resources but is constrained by its own ad business. Meta gives away its models for free, commoditizing the market. Anthropic's strategy of focusing on enterprise safety and reliability is a niche, but it's a niche that OpenAI and Google can also serve. The real threat is from open-source models. For example, the open-source model Mixtral 8x22B from Mistral AI achieves performance comparable to GPT-3.5 on many benchmarks and can be run on a single server for inference, drastically reducing cost.
| Company | Model | MMLU Score | Cost/1M tokens (input) | Open Source | Enterprise Focus |
|---|---|---|---|---|---|
| OpenAI | GPT-4o | 88.7 | $5.00 | No | Yes (Azure) |
| Anthropic | Claude 3.5 Opus | 88.3 | $15.00 | No | Yes (Safety) |
| Google | Gemini Ultra | 90.0 | $10.00 | No | Yes (Vertex) |
| Meta | Llama 3 70B | 82.0 | ~$0.50 (self-hosted) | Yes | Limited |
| Mistral AI | Mixtral 8x22B | 80.5 | ~$0.20 (self-hosted) | Yes | No |
Data Takeaway: Anthropic's Claude is competitive but not dominant. Its higher cost per token is a disadvantage in a price-sensitive market. Open-source alternatives offer comparable performance at a fraction of the cost, putting pressure on Anthropic's margins.
Industry Impact & Market Dynamics
Burry's skepticism is a canary in the coal mine for the broader tech valuation bubble. The AI and space sectors are currently fueled by low interest rates and a fear of missing out (FOMO) among venture capital and sovereign wealth funds. If Burry is right, we could see a correction that ripples across the entire ecosystem.
Space Industry Dynamics: The space industry is shifting from a government-funded monopoly to a commercial market. But the total addressable market (TAM) for space is still small. Morgan Stanley estimates the global space industry could reach $1 trillion by 2040, but that includes satellite services, manufacturing, and ground equipment—not just launch. SpaceX's valuation of $180B already represents a significant chunk of that future market. For it to reach $1T, it would need to capture an implausible share of a $1T market, or the market must grow much faster than projected.
AI Industry Dynamics: The AI market is projected to reach $1.8 trillion by 2030 (per Grand View Research). But this includes all AI software, hardware, and services. The foundation model layer is expected to be a small fraction—perhaps $100-200 billion. With OpenAI, Anthropic, Google, Meta, and dozens of startups all competing, the market is already overcrowded. The winner-take-most dynamics that justified high valuations for Google and Facebook are less likely in AI, because models are becoming commodities and switching costs are low. Developers can switch from Claude to GPT-4o to Llama 3 with minimal friction.
| Sector | 2023 Market Size | 2030 Projected Size | Key Players | Valuation Concern |
|---|---|---|---|---|
| Space Launch | $12B | $30B | SpaceX, ULA, Blue Origin, CASC | Market too small for $1T valuation |
| Satellite Internet | $4B (Starlink) | $20B (all players) | Starlink, OneWeb, Amazon Kuiper | High CapEx, regulatory risk |
| Foundation AI Models | $10B | $100-200B | OpenAI, Anthropic, Google, Meta | Commoditization, low margins |
Data Takeaway: The projected market sizes for both space and AI are large, but not large enough to support multiple trillion-dollar companies at the current valuation multiples. SpaceX and Anthropic are being valued as if they will capture the majority of their respective markets, which is a high-risk bet.
Risks, Limitations & Open Questions
SpaceX Risks:
- Starlink Saturation: Subscriber growth is slowing. The initial 2.5 million customers were the low-hanging fruit (rural, underserved areas). Next wave requires penetrating urban markets, where terrestrial competition is fierce.
- Regulatory Hurdles: Spectrum disputes with terrestrial 5G operators (e.g., Dish Network, AT&T) could limit Starlink's bandwidth. Orbital debris regulations could force costly deorbiting schedules.
- Starship Delays: Starship is critical for Starlink V3 satellites and for the Mars mission narrative. Delays or failures could stall revenue growth and investor confidence.
Anthropic Risks:
- Commoditization: Open-source models are closing the gap. If Llama 4 or Mistral 2 matches Claude 3.5 Opus, Anthropic's pricing power evaporates.
- Dependence on Compute: Anthropic relies on cloud providers (AWS, Google Cloud) for compute. Any price increase or capacity constraint hurts margins.
- Safety vs. Utility: Anthropic's focus on safety (constitutional AI) may make its models less capable or more censored than competitors, potentially alienating power users.
Open Questions:
- Can SpaceX achieve a profit margin >20% on Starlink? If not, its valuation is unjustified.
- Can Anthropic develop a proprietary data moat (e.g., unique enterprise data) that competitors cannot replicate?
- Will the AI market consolidate into a winner-take-most oligopoly, or will it fragment into many specialized models?
AINews Verdict & Predictions
Verdict: Burry is correct in his skepticism, but for the wrong reasons in the short term. The trillion-dollar valuations are not based on current fundamentals but on future expectations. However, the market is currently pricing in a best-case scenario that ignores significant risks. We believe both companies will eventually become very valuable—perhaps worth $500 billion each—but reaching $1 trillion is a stretch within the next decade.
Predictions:
1. SpaceX will not reach a $1 trillion valuation within 10 years. Starlink's growth will plateau as terrestrial networks improve, and the launch market will remain a low-margin commodity business. SpaceX's best path to $1T is through a breakthrough in space-based manufacturing or asteroid mining, which is highly speculative.
2. Anthropic will be acquired or merge within 5 years. The AI model market is too competitive for a standalone company to survive. A likely acquirer is AWS (which already invests heavily in Anthropic) or a large enterprise software company like Salesforce or Oracle. The acquisition price will be well below $1 trillion.
3. The next correction in AI and space stocks will be triggered by a macro event (e.g., recession, interest rate hike) that forces investors to focus on cash flows rather than growth. When that happens, companies with negative margins and high CapEx will be hit hardest.
What to Watch:
- SpaceX: The profitability of Starlink's next-generation satellites (V3) and the success of Starship's commercial launches.
- Anthropic: The release of Claude 4 and whether it can achieve a 10x performance improvement over GPT-4o. If not, its differentiation disappears.
- Market: The valuation of secondary market trades for both companies. If they start declining, it's a signal that institutional investors are losing faith.