CNBC Scott Wapner with Brad Gerstner - May 14th, 2026

Key points

• **Cerebras IPO live blow-by-blow.** Opened on the show. IPO price $185 → first trade $350 (~89% pop). 30M-share offering, only 10% to float = ~3M shares. 25x oversubscribed. Crosses **$100B market cap** on day one. **Most important AI-infra IPO since Snowflake.** Brad's reaction: 'this team is intent on getting to $500B and then $1 trillion. Andrew Feldman is an absolute warrior.' But: 'I would not be YOLO-ing into this thing at $400. At $325 it feels better; institutional begins to move in around these levels.'
• **The 'dribble lockup' as a novel IPO structural innovation.** Traditional: 6-month cliff = all insider shares unlock on day-180 = massive overhang quant-arb'd against. Cerebras: **rolling release.** Employees can start selling within days; investors after the first earnings report (~5-7 weeks); then ~15% released, then 6% every few weeks over 6 months. **Two effects:** (a) avoids the post-cliff overhang; (b) gets supply into a clearly-supply-constrained market faster. Brad posting the chart on Twitter. **Watch for this structure replicating in the SpaceX / Anthropic / OpenAI IPOs coming.**
• **Anthropic added $14B incremental ARR in April alone** = '2 Databricks plus 1 Palantir in one month.' **Cross-references Krishna Rao's $30B+ end-of-Q1 disclosure from ILTB this week** — the run-rate has continued accelerating into April. 'The most parabolic revenue numbers in the history of capitalism.' If Anthropic hadn't shown up with these, the broader market might be down. The AI revenue story carrying the whole index.
• **Software 'in the too hard basket' — Brad's strongest framing yet.** 'Software stocks have reverted from a decade-long superior multiple to a market multiple. When you adjust for real SBC and treat them like all the other Mag-7, they're basically at market.' **'I can't see out more than 2-3 years on AI's impact on software, so paying 20x FCF feels fundamentally at odds with AI.'** Exceptions: data-infrastructure layer (Snowflake, Databricks, ClickHouse) — token consumption drives basic-services consumption. **Point-solution SaaS = 'on the front of the conveyor belt heading toward the guillotine.'** Direct continuation of [Issue 03's bucket-1 melting-iceberg framing](/issues/2026-05-03) and Lemkin's 'terminal state of decay' from this week's 20vc.
• **Power + compute as the actual cycle bottleneck.** **30-40% of announced compute is delayed this year** — same number Chamath quoted on [Issue 05 All-In](/issues/2026-05-17). 'Keep your eye on power and compute. If delays persist, it will roll back onto revenue and ROI of capex.' **The 'Tom Brady launches a data centre' sell signal:** when celebrities start NeoClouds is when Brad takes the portfolio flat for the year. Currently being pitched 'multiple Neo Clouds every week' but no celebrities yet (vodka → tequila → JADA NeoCloud is next, joke).
• **Amazon defence.** 'I remember 2010 when everyone said why is Amazon wasting all this money on AWS. That something turned out to be the greatest gold mine in history.' Bill Baruch trimmed Amazon + Alphabet; Brad: 'we own a lot of Amazon and aren't selling any — this is the start of something, not the end.' Why: 'Amazon is now the premier place for Claude workflows + Anthropic.' Headcount-light, capex-heavy, growth-rate-accelerating (Q1: AWS 28%, Azure 39%, Google Cloud 60%). 'Without these numbers, the market would be down a lot.'
• **On capex risk + debt concerns.** 'They're not racking up a lot of debt — they're producing $100B of FCF/year. We're financing $1T/year of capex without any government money. China has to subsidise this with taxpayer money.' But Malcolm Etheridge counter: 'a lot of debt is being racked up off-book through SPVs. At some point that conversation gets louder.' Brad doesn't dispute it. **Sub-text: the off-book SPV financing structure for hyperscaler infra capex is the unstated tail risk Brad knows but isn't bothered by.**
• **Brad's IPO calendar update.** Cerebras = first big AI-era IPO. SpaceX is next ('end of June', re-confirmed from his Milken interview last week). Anthropic and OpenAI are private-IPO already ($120B raised at OpenAI; $50B at Anthropic in 48 hours per [20vc this week](/issues/2026-05-17)) and may push to 2027. Smaller / medium-sized companies: 'today in Silicon Valley, my portfolio companies are saying we have a shot to do this — you don't have to be a trillion dollar company to go out.' **Implication: a longer queue of mid-sized AI IPOs is opening behind Cerebras now that the test case worked.**
• **Walked back from yoloing.** Earlier this week Brad called Nvidia 'the first $10 trillion company.' On this hit, he stays in but adds nuance: Nvidia did 'nothing for six months' before this week's break; H200 sales cleared to 10 firms in China; Jensen in China for trip; 'multiple expansion alone justifies these levels.' Memory: 'still trading at 5-6x earnings — fundamentally restructuring these industries.' Will host Micron CEO on next BG2 pod.
• **On the Cisco-2002 comparison.** Reject. 'Nvidia is growing 70%, was trading below market multiple. People propagating the Cisco myth for 2.5 years have missed all the upside. The question is always when too much is too much — and we're not there.' Memory at 5-6x earnings is **structural mispricing, not cyclical.** Brad's bet is on the **memory-as-the-real-bottleneck thesis from [Issue 04](/issues/2026-05-10)** combined with the **memory durability re-rating from earlier this week.**
• **Final trade: Lilly** (continuation of leveraged-balance-sheet leaning-in trade). Other panel final trades: Dominion Energy (Josh Brown — AI raises all energy boats), Apple-Gemini partnership (JP — 'Apple cheating on OpenAI with Gemini, this was always meant to be'). **The Apple-Gemini speculation is worth tracking** — would be the direct response to the OpenAI-suing-Apple story from this week's All-In.

Notable quotes

Themes

• Cerebras IPO opens at $350 / ~$100B market cap with novel dribble lockup
• Anthropic +$14B ARR in April = '2 Databricks + 1 Palantir' in one month
• Software 'in the too hard basket' — market multiple reversion, point-solutions on guillotine
• Power + compute (30-40% delays) as the cycle's real bottleneck
• Memory at 5-6x earnings as structural mispricing — Brad's highest-conviction trade