Michael Burry Questions Tenability of SpaceX and Anthropic Valuations, Citing Compute Overcapacity
Investor Michael Burry has challenged the multi-trillion-dollar valuation targets of SpaceX and Anthropic, pointing to unprofitable S-1 fundamentals and unsustainable compute scaling strategies. Burry warns that the frantic capital allocation toward AI infrastructure is a temporary demand spike that will result in massive hardware overprovisioning as compute power commoditizes.
SpaceX S-1 Disclosures and Valuation Divergence
SpaceX's May 20 IPO prospectus filing (S-1) exposes a significant mismatch between the company's financial fundamentals and its reported $2 trillion target valuation. The regulatory filing revealed that SpaceX generated $18.7 billion in revenue last year, paired with a substantial net loss of $4.9 billion.
From an analytical perspective, a valuation exceeding $1 trillion cannot be justified by these operational metrics. Any upward trajectory in share pricing is driven primarily by market hype and technical trading dynamics rather than structural unit economics. The current S-1 disclosures do not support a $1 trillion valuation, let alone the $2 trillion figure targeted for the public listing.
Anthropic and the High Marginal Cost of LLM Training
Anthropic’s recent capital raise at a $965 billion valuation highlights a broader systemic risk within the generative AI market. The current methodology for developing cutting-edge frontier models relies on brute-force scaling architectures that are highly capital-intensive and unsustainable over longer horizons.
The capital expenditure required for training and running these large-scale models is too high relative to their long-term value capture. As optimization techniques and hardware efficiencies advance, compute power will inevitably undergo commoditization, following the historical pricing trajectory of internet bandwidth. Consequently, proprietary model developers are unlikely to sustain valuations near the $1 trillion threshold as their underlying technological moats erode.
Tokenmaxxing and Infrastructure Overprovisioning
The current macroeconomic rush to secure advanced compute infrastructure is a false demand signal. This phenomenon, characterized as a "tokenmaxxing" trend, has triggered an aggressive buildout of data center capacity and hardware procurement that far exceeds projected long-term demand.
Systems operators and hyperscalers are currently overspecifying and overbuilding infrastructure to support workloads that will not scale linearly in the coming years. Once the current cycle of model training stabilizes and execution becomes more efficient, the market will face a significant supply-demand mismatch, leaving operators with underutilized, capital-intensive infrastructure.