SpaceX, OpenAI, and Anthropic Are Woth More Than Every Single IPO in the Last 45 Years

SpaceX, OpenAI, and Anthropic are collectively valued at roughly $5.2 trillion, exceeding the $4.07 trillion combined first-day value of 3,365 U.S. technology IPOs from 1980-2025. The valuations rest on aggressive future economics: OpenAI projects $278 billion of cumulative negative free cash flow from 2026-2030 while spending about $856 billion on computing infrastructure, and Anthropic does not expect positive cash flow until 2028. The article highlights rising AI-bubble risk, with Apollo warning that major technology companies may need to triple cash flow by 2030 or face weaker AI investment, wider credit spreads, reduced capex, and slower U.S. GDP growth.

Analysis

The investable fault line is financing duration, not near-term AI usage. If frontier-model economics remain capital-intensive for longer, the value pool shifts from model owners toward toll collectors: NVDA and networking/power beneficiaries (AVGO, VRT, ETN) retain revenue visibility while AI labs absorb depreciation, lease obligations, and funding-round dilution. That advantage narrows once customers begin optimizing inference costs or delaying incremental clusters, making order-cancellation and backlog-conversion data more important than headline capex commitments over the next 1-3 quarters.

A public listing of a major model developer would create a live valuation anchor for private AI portfolios and could force markdowns across late-stage venture marks before it creates a durable equity opportunity. APO has upside from secondary-market liquidity, private-credit demand, and fundraising activity, but it is also exposed to wider financing spreads and lower realizations if growth-company marks reset. The second-order macro risk is that hyperscaler capex is increasingly functioning as a growth support; a synchronized capex pause would hit power equipment, data-center REITs, semiconductors, and credit spreads simultaneously over 6-18 months.

Consensus is too focused on whether AI is "real" and insufficiently focused on who bears the residual asset risk. Model providers may ultimately win revenue share yet still generate weak equity returns if compute suppliers capture economics and the cost of capital remains elevated. Conversely, an apparent valuation correction could be contained if inference monetization improves quickly enough to lift utilization and reduce the need for externally funded compute; quarterly cloud-margin commentary and disclosed contracted capacity are the key falsifiers.

Key Decisions for Investors

• Maintain a 3-6 month relative-value position: long NVDA / short a basket of AI-capex-sensitive model-owner proxies when available; until direct listings are liquid, use long NVDA versus short ARKX or a high-duration software basket. The thesis fails if NVIDIA data-center guidance decelerates materially while model-provider monetization evidence accelerates.
• Reduce unhedged exposure to VRT and ETN after sharp AI-capex-driven rallies; retain only with downside hedges. A 10-15% cut in hyperscaler capex guidance or evidence of data-center delivery delays would likely compress backlog multiples faster than reported revenue, while sustained order-book growth and pricing power invalidate the caution.
• Watch, rather than initiate, APO into any AI private-market issuance window. Buy only if fee-related earnings guidance and deployment pace offset any realization losses; avoid if private-credit spreads widen by roughly 75-100bp or management signals weaker exits, as balance-sheet and carry-mark sensitivity would overwhelm transaction-fee upside.
• Use 6-12 month put spreads on QQQ or SMH as portfolio insurance against a capex-led de-rating, funded by trimming crowded AI infrastructure winners. The hedge should be closed if hyperscalers demonstrate stable cloud margins alongside continued double-digit capex growth, which would validate that infrastructure spending is being absorbed rather than merely financed.