The post went up late on a spring evening, a long argument about startup economics from a name the valley still stops to read. Within hours it was the most-discussed item on Hacker News, collecting 1,561 points and holding the day’s top spot. By morning, the AI industry was arguing with itself in public.
The author was Mitchell Hashimoto, the developer who created Vagrant and Packer and later co-founded HashiCorp. His argument, in brief: much of the AI startup world is built on delusion. Companies across the sector are burning through hundreds of millions of dollars, he wrote, with no working product and no users to show for it.
Hashimoto did not name individual companies, according to people who read the post. He described a pattern instead. Funding rounds sized for category winners. Burn rates that would require venture-scale revenue inside two years. Teams that measure progress by headcount, press releases, and model benchmarks rather than by customers.
A pattern the industry recognized
The post landed at a particular moment. Enterprise AI budgets are under renewed scrutiny, and several high-profile startups have quietly trimmed burn rates after recent rounds came in below internal targets. Public-market investors have spent the past year pressing the same question in earnings calls: where is the revenue behind the headline numbers?
Hashimoto’s framing matched that question almost exactly. He reportedly described startups that raise on the strength of a demo, hire a large research team, and then discover that nobody will pay for the underlying model at a price high enough to cover the compute bill. The result, in his telling, is a sector in which the winners’ cost structure is subsidized by the capital allocated to the losers.
That specificity gave the post its force. This was not a sweeping complaint about hype. It was a description of a failure mode – raise, demo, hire, stall – that a meaningful share of people in the thread said they had watched happen.
Two camps in one comment section
The response split quickly. Supporters argued that Hashimoto had said aloud what investors and engineers whisper in private. Several commenters shared their own examples of AI companies whose pitch decks outran their invoices. Some pointed to a pattern of “paper products” – demos that never become deliverables – as the sector’s defining problem. A few predicted a consolidation wave that would fold dozens of funded startups into larger players at fire-sale prices.
The counter-argument was just as loud. A large bloc of commenters said Hashimoto does not understand how products are built. They noted that infrastructure tools like Vagrant and Packer had years-long adoption curves of their own before reaching mainstream use. Generative AI, they argued, is a platform shift with a longer gestation than a single fundraising cycle, and judging the whole sector by early revenue is how incumbents talk themselves into missing transitions.
Others pointed to the difference between his market and theirs. Hashimoto’s career sits in developer tools, where buyers are engineers rather than procurement departments. Enterprise AI sales run on a different clock: longer deal cycles, slower referenceability, pilots that convert to production contracts only after a year or more.
Why his word carries weight
This is not a random developer venting on a forum. Hashimoto co-founded HashiCorp in 2012 with Armon Dadgar and built it into one of the most successful infrastructure software companies of its generation. It went public in December 2021 at a valuation of roughly $15.5 billion. Vagrant, Packer, Terraform, Vault, and Consul became standard tools in the modern data center, and HashiCorp’s revenue grew from a few million dollars a year to hundreds of millions under his watch.
He stepped away from day-to-day operations in late 2023. In October 2024 he joined OpenAI, the most valuable private AI company in the world, spending several months inside the industry’s center of gravity before resigning in February 2025.
That biography matters for the debate. He is one of the few people who has both built a public infrastructure company and worked inside a frontier lab, and his critique carries unusual credibility on both sides of the argument. He has also been candid since leaving OpenAI, using his public platform to write about what he saw rather than to cash in on it.
A consensus starting to crack
The thread is the clearest sign yet that the AI industry’s internal consensus is cracking. For two years, the prevailing assumption has been that capital efficiency can be deferred because the market is large enough for everyone. Hashimoto’s argument – that a generation of startups is spending like winners while shipping like also-rans – is the bluntest version of a question investors increasingly ask in private.
People familiar with the thinking of several venture firms said the post matched conversations that have been happening behind closed doors for months, particularly around startups whose valuations were set when capital was cheaper. Analysts said the timing compounds the message: with public markets questioning AI revenue quality, the fundraising climate for early-stage AI companies is already tightening.
The most interesting part of the debate is not who is right. It is that the debate is happening at all on the front page of Hacker News, where the industry’s builders actually read the comments. Whatever the merits of the two sides, the argument Hashimoto started is one the AI industry can no longer avoid having.
