How AI company valuations broke the old secondary discount rules

For most of the last decade, the secondary market ran on a reliable rule of thumb: shares trade at a 10-30% discount to the last primary round, with a small tier of the most in-demand names occasionally clearing at a premium. The current generation of frontier AI labs has stretched that rule in ways the old playbook didn't anticipate — sometimes toward eye-watering premiums, and at least once toward a discount nobody expected from a company growing this fast.

The premium case: pricing a moving target

When a company's valuation triples in under a year — as several AI labs have — the last primary round stops functioning as a reliable anchor almost immediately. Buyers who believe the growth trajectory will continue bid secondary shares above the most recent round, effectively pricing in the next round before it happens. Anthropic's April 2026 tender offer is a clean illustration: priced at a $350 billion valuation, secondary-market chatter at the time was already suggesting $500-800 billion — a gap wide enough that many employees chose to hold rather than sell at the tender price. See the full case study.

The discount case: Discord's reminder that hype isn't universal

Not every hot-sector name commands a premium. Discord's secondary shares traded around 39% below a primary round that had just closed weeks earlier — a gap normally associated with a company losing momentum, not one that had just raised fresh capital at a higher price. Perplexity showed a milder version of the same pattern: a modest discount on one marketplace despite rapid growth and a hot sector narrative. Both cases are a reminder that discount and premium dynamics are set by the actual pool of secondary buyers, not by press coverage — and that pool can disagree with a primary round's lead investor.

Why the split is getting wider, not narrower

Three forces are pulling AI-sector secondary pricing further from the old 10-30% band in both directions. First, primary rounds themselves are closing faster and at bigger jumps, so the "anchor" is fresher but also more volatile — a round from six months ago can already look dated in a market moving this fast. Second, the buyer pool for the largest labs has become genuinely global and institutional, backed by the scale of capital covered in thebuyer & fund directory, which increases competition for scarce allocations in the winners and increases scrutiny of everyone else. Third, employee tender offers at this scale — OpenAI's $6.6 billion, Anthropic's targeted $5-6 billion — now function almost like public earnings releases, giving the whole market a fresh, closely watched data point to react to.

What this means for anyone pricing an AI-sector secondary

The old shortcut — "shares trade about 20% under the last round, adjust from there" — no longer travels reliably into this sector. A buyer or seller needs to check actual, current marketplace pricing and recent tender results for the specific company, not apply a market-wide rule of thumb. See how a deal actually works for the diligence and pricing steps that matter most when the anchor itself is moving this fast.