Intel Rips, Thrive Launches Eternal, GPT 5.5 | Diet TBPN
TL;DR
Intel’s stock ripped on a narrative shift, not just the headline numbers — Q1 revenue hit $13.6 billion and the stock jumped 20% after hours, but the real story was investors buying five converging demand drivers: AI agents, CPU-heavy inference, advanced packaging, U.S. industrial policy, and Elon Musk’s rumored Terafab ambitions.
The hosts argue AI agents may give Intel its second chance — Intel’s data center segment did $5.1 billion versus $4.5 billion expected, and Lip-Bu Tan said the CPU-to-GPU ratio has already moved from roughly 1:8 to 1:4, with some bulls like Evercore’s Mark Lapidus floating an even wilder future flip to 8:1.
TBPN framed Intel as a national-security chip bet as much as a business turnaround — the conversation kept returning to the U.S. government’s 10% stake, Taiwan risk, and the idea that Intel is the only remotely plausible domestic leading-edge foundry if Washington wants guaranteed American chip capacity.
The show used Cursor’s ugly margins to tee up a bigger compute argument — after citing The Information’s report that Cursor ran at negative 23% gross margins earlier this year, the hosts argued margins could improve if more inference shifts onto vertically integrated stacks like xAI or proprietary fine-tuned models.
A Patrick O’Shaughnessy clip pushed the ‘token demand will outrun supply’ thesis — the core claim was that Anthropic, OpenAI, and even tier-two labs may stay sold out because model value is rising faster than available compute, while Bill Gurley’s rebuttal was that real shortages should show up as higher prices, not mysterious excess demand.
The back half was classic TBPN rapid-fire weirdness with real signal underneath — Thrive Capital launched permanent-capital vehicle Thrive Eternal with the San Francisco Giants as its first expected partner, Cohere is merging with Aleph Alpha, GPT-5.5 hit the API, and the hosts treated X’s new X Chat app as a mildly amusing ‘sure, more apps’ update.
The Breakdown
Intel’s 20% after-hours pop and the ‘cruise ship turning’ thesis
The show opens on Intel “absolutely ripping,” with shares up 20% after hours after posting $13.6 billion in Q1 revenue. The hosts note the company still lost $3.7 billion, but say that was distorted by one-off Mobileye and derivative charges; stripped out, Intel earned about $1.5 billion versus expectations for roughly break-even. Their framing: the giant, embarrassing laggard of the AI trade may finally be starting to turn.
Why the market suddenly has an Intel story again
They walk through why Intel was the missing piece in a broader computing boom: it missed mobile, fell behind TSMC, and never found real AI GPU traction. But now AI agents are creating a new bull case, because training is still a GPU game while orchestration, inference routing, memory handling, and always-on agent workflows pull more demand back toward CPUs. Taylor and the others also emphasize the geopolitical layer: if Taiwan risk matters and chipmaking becomes national security, Intel is the obvious U.S. vehicle.
The wild CPU ratio debate: from 1:8 to maybe 8:1?
Lip-Bu Tan’s earnings-call line becomes the centerpiece: CPU-to-GPU demand has moved from one CPU per eight GPUs to one per four, and Evercore’s Mark Lapidus thinks inference and agents could push that all the way to eight CPUs per GPU. The hosts think that sounds kind of crazy, but still agree the directional shift is bullish for Intel even if the ratio never gets that extreme. One of the better lines: a smart model sitting in front of a GPU can “spin off a ton of CPU workloads,” so GPUs may actually create more downstream CPU demand.
Musk’s Terafab and the stack of ‘plausible demand stories’
Then comes the Elon wild card: a rumored Terafab effort involving Tesla, SpaceX, and maybe more Musk companies, with Intel helping design, manufacture, and package chips. The scale is deliberately jaw-dropping — 100,000 wafers per month initially, maybe 1 million eventually, which they compare to roughly 70% of TSMC’s total monthly output. They immediately temper it with reality, quoting TSMC CEO C.C. Wei’s basic point that fabs take years, but still argue Intel now has five believable demand stories all pointing in the same direction.
Cursor’s negative margins and the compute economics fight
The show pivots to Cursor, jokingly miscalling a rumored $60 billion buyout a “bailout,” then highlighting reporting that Cursor had negative 23% gross margins earlier this year. The hosts don’t treat that as fatal; they argue users are sticky and that margin structure could look very different if inference is eventually served on cheaper, vertically integrated infrastructure like xAI/Colossus-style capacity or Cursor-specific model layers. It turns into a broader question of who captures value when demand for tokens keeps climbing.
Patrick O’Shaughnessy, Bill Gurley, and whether AI demand is ‘real’
A long clip from Patrick O’Shaughnessy argues that compute is the choke point: Anthropic may have the best model but not enough capacity, OpenAI can keep buying supply, and even second-tier labs may stay sold out because the economic value of top models is outrunning infrastructure buildout. Bill Gurley’s pushback is cleaner and more old-school — if supply is constrained, prices should rise to clear the market. TBPN’s answer is that this isn’t just VC fairy dust; there are real Fortune 500 buyers, hedge funds, banks, and regular subscribers paying because the tools are already creating value.
Thrive Eternal, baseball explained badly, and the Friday sprint to the finish
The final stretch is pure TBPN: Josh Kushner launches Thrive Eternal, a permanent-capital vehicle spanning durable tech and cultural assets, with the San Francisco Giants as the first expected partner. They hilariously stop to explain that baseball is a bat-and-ball game where teams sell tickets, then race through a soldier allegedly profiting from betting on a mission, people “playing Minecraft through ChatGPT” image generation, Cohere merging with Aleph Alpha, a funny GPT-5.5 reasoning example (“Car, bro, it’s a car wash”), X Chat launching, and Greg dropping that GPT-5.5 is now in the API.