3 Model Drops. $15M/Day in Burn. One Product Dead. Nobody Connected Them.
TL;DR
Sora’s shutdown was really an inference-cost alarm bell — Nate argues OpenAI killing Sora on March 24 wasn’t just a failed video product story; it exposed a brutal economics gap, with an estimated $15 million/day in inference burn against only $2.1 million in lifetime revenue.
Ads finally entered the AI interface in a way that could hit Google where it hurts — Criteo’s integration into OpenAI’s ad pilot showed early signs that LLM-driven traffic converted 1.5x better than other referral channels, suggesting search intent and ad dollars may migrate from blue links to chat.
Washington may clear AI regulation, but it can’t force data centers into unwilling communities — even as the White House pushed a federal AI framework on March 20, at least 12 states and 54 local governments were moving to pause data-center construction over power, water, and land-use concerns.
The SaaS crisis is less about AI magic and more about broken pricing models — Atlassian’s 1,600 layoffs, after Mike Cannon-Brookes had recently said the company would employ more engineers, became Nate’s case study for how per-seat pricing is getting punished before many software companies are ready to adapt.
Safety posture has become a go-to-market decision, not just an ethics stance — Anthropic’s refusal to meet Pentagon terms reportedly cost it a $200 million contract and triggered a federal backlash, but also created a trust signal that may pay off with consumers and enterprise buyers.
The through line across March was sustainability, not capability — Nate’s big frame is that AI has moved from a phase obsessed with what’s possible to one dominated by what can be delivered with margin, defended politically, and sustained as a business.
The Breakdown
March Was Loud, But Nate Wants the Structural Story
Nate opens by saying everyone already saw the hot takes: frontier model drops, GTC, SaaS panic, product deaths. His point is sharper: the real skill now is “reading differently,” pulling signal from the noise and spotting what actually changes industry power dynamics over the next 12 months.
Sora’s Death Marks the Shift From Training Wall to Inference Wall
The first big example is OpenAI quietly shutting down Sora just six months after launch, including the API, app, and ChatGPT video generation. Nate says the real story isn’t “AI video failed,” it’s that Sora reportedly burned $15 million a day in inference costs against just $2.1 million in lifetime revenue — numbers even Sora head Bill Peebles admitted were unsustainable. His takeaway: the hard constraint in AI is moving from training scale to inference cost per unit of revenue.
Criteo’s ChatGPT Ads Hint at a New $600 Billion Interface Shift
On March 2, Criteo became the first ad-tech company integrated into OpenAI’s advertising pilot, and within days it was pitching 17,000 advertisers. Nate zeroes in on one datapoint: in a sample of 500 retailers, traffic coming from LLM platforms converted at 1.5x the rate of other referral channels. In his framing, this is the moment the ad industry got a credible answer to what happens when the search page dissolves into one trusted recommendation inside a conversation.
The US May Deregulate AI on Paper While Blocking It in the Real World
Nate then pivots to the White House’s March 20 AI framework, which sounds industry-friendly: one federal standard, no new regulator, lighter permitting. But he says the important contradiction is physical, not legal — at least 12 states had data-center moratorium bills in play, 54 local governments had enacted freezes, and communities were pushing back on power and water use. Add in Iranian drone strikes on AWS facilities in the UAE and Bahrain, and suddenly the question of where AI can physically live becomes geopolitical, local, and messy all at once.
Atlassian’s Layoffs Expose the Real SaaS Problem
Atlassian’s March 11 layoff of 1,600 employees lands as Nate’s cleanest SaaS case study, especially because CEO Mike Cannon-Brookes had said just five months earlier that the company would likely employ more engineers, not fewer. Nate doesn’t fully buy the public “AI made us do it” narrative; he says many cuts are really overhiring plus investor-friendly storytelling. But the structural issue is bigger: if AI collapses seat demand, then per-seat SaaS pricing breaks, and Wall Street is punishing companies faster than those companies are reinventing themselves.
Anthropic’s Pentagon Clash Turned Safety Into Market Positioning
The final major March shift is Anthropic’s refusal to let Claude be used for all lawful Pentagon purposes, including its red lines around autonomous weapons and mass surveillance. That reportedly led to a government-wide ban and the loss of a $200 million contract, but Nate says the more important shift is that “safety posture” is now part of commercial sorting. In his view, companies are being forced to choose where they sit on the spectrum between deploy-first and safety-first, and that choice now has direct revenue consequences.
The New AI Question Is No Longer “Can We Build It?”
Nate closes by tying the threads together: Sora, ad monetization, data-center politics, SaaS layoffs, and safety battles are all sustainability stories. The industry’s center of gravity is moving from capability to economics — away from pure frontier excitement and toward the harder question of what can actually be built, sold, governed, and sustained with margin.