AI Data-Center IPOs Face a Reality Check as Hyperscaler Capex Reaches an Inflection Point

The gulf between artificial intelligence infrastructure spending and the market's willingness to fund new entrants is widening. The four largest hyperscalers have guided to $720-$745 billion of combined 2026 capital spending, roughly double the 2025 pace, with Amazon alone near $220 billion. Goldman Sachs puts the group closer to $800 billion and pegs 2027 consensus at $1.1 trillion. At those levels, the sector needs approximately $300 billion of annual revenue to break even and $417 billion to generate a 10% return. That math is setting a high bar for any company trying to go public on the back of AI infrastructure demand.
Elice, a South Korean AI data-center firm, illustrated the gap on October 20 when its shares fell 32% on their first trading day, triggering immediate questions about whether public markets can absorb new AI infrastructure paper at current valuations. The Nasdaq recovered intraday, lifted by falling oil prices and slipping Treasury yields, but the Elice print landed as a data point the broader market could not ignore.
The macro backdrop is not making new listings easier. The 10-year Treasury yield pushed above 5% in October 2026, reaching levels last seen in 2002, before settling near 5.07% in the week ending October 16. Higher long rates compress the multiples applied to growth-stage companies and raise the cost of the very debt that funds the data-center builds underpinning AI revenue projections.
The contrast with established operators is sharp. $SPCX (Space Exploration Technologies Corp.) reported October 21 revenue of $14.58 billion against a $12.72 billion estimate, up 161% year over year, with AI revenue of $7.56 billion beating the $6.3 billion consensus. Shares rose 9% to $160.70. The result underscores that scale and existing customer relationships are what the market is rewarding, not the infrastructure build-out story alone.For the broader sector, the Elice debut and the 10-year yield at 5% set the conditions for the next wave of AI-adjacent IPO pricings. Sponsors and issuers will need to demonstrate a clear path to the revenue thresholds Goldman has outlined before public investors assign growth-stage multiples to new data-center names.