Zitron Calls Big Tech's $765B AI Bet a Cargo Cult Ritual

Ed Zitron's latest essay at Where's Your Ed At is a long argument that the modern tech industry has stopped trying to build things and started doing impressions of past successes. The frame he reaches for is the cargo cult: rituals that look like the practices that produced iPhones and AWS, performed by executives who do not actually understand what made the originals work. AI, in his telling, is the latest and most expensive version of that ritual.

The numbers he stacks up are the load-bearing part. He puts OpenAI's 2025 spend at $21 billion, with $8.67 billion of that going to inference in the first nine months of the year and $5.73 billion to sales and marketing. He sets that against a projection of $765 billion in big-tech AI capex in 2026, which he calls more than ten times what it cost Amazon to build AWS to profitability between 2003 and 2017. The original iPhone, by his count, took roughly $150 million over two and a half years. The implied question is what the industry is buying now for so much more money.

The personnel critique is sharper. Zitron walks through researchers like Andrej Karpathy, Noam Shazeer and Barret Zoph cycling between OpenAI, Google and back again, and executives like Snap's Evan Spiegel pushing $2,195 AR glasses without an obvious use case, as evidence that the industry is hiring famous names and shipping symbolic products in lieu of having a plan. "Faith can't fill your belly, or cover $1.1 trillion in compute commitments" is the kicker he leaves you with.

The honest caveat is that this is one writer's polemic, and Zitron is openly an AI skeptic. The dollar figures are his aggregation of company filings and trade reports rather than a single audited source, so individual specifics are best taken as reported, not settled. The essay also does not give you the bull-case rebuttal, which is the part a treasurer or a board member would want before deciding whether the 2026 budget is brave or reckless.

What is worth carrying out of it, even if you discount the framing, is the demand to see present-tense proof rather than historical analogies. The next interesting moment in this cycle is probably the first hyperscaler willing to publish inference margins on a real paying enterprise workload. Until then, the cargo cult line is going to keep landing.