The Uncomfortable Question Behind Every Silicon Valley Origin Story

NeXT, the company Steve Jobs founded after Apple pushed him out, was a commercial catastrophe. Over five-plus years of selling hardware, it moved roughly 50,000 units total. Not per year. Total. So how did it survive more than a decade? That question is circulating again, and one proposed answer is that some of the money came from intelligence agency budgets.

Let me be upfront: this one is thin on verifiable sourcing. I couldn’t find recent community discussion digging into it, and I haven’t confirmed the underlying reporting firsthand. So this isn’t a piece arguing the claim is true. It’s a piece about why the claim lands so easily — because the relationship between Silicon Valley and government money is one of the best-documented things in the industry’s history, with or without this particular story.

NeXT never had a business model that closed

The NeXT Computer launched in 1988 at $6,500. That’s roughly $17,000 in today’s money. It was pitched at university labs, which is a polite way of saying it was priced for buyers who didn’t exist in volume.

The automated factory Jobs built in Fremont could produce 10,000 units a month. Actual output ran around 400. A 4% utilization rate. The factory was more famous as a magazine photo shoot than as a manufacturing operation.

And yet the company ran from 1985 until Apple acquired it in 1996. The publicly known funding: $7 million of Jobs’ own money, $20 million from Ross Perot, $100 million from Canon. Even with all that, NeXT killed its hardware business in 1993, cut more than half its staff, and shrank into a software company.

There’s always been a gap in the story: where did those workstations actually go? Universities and financial firms were the stated market. But in that era, the biggest institutional buyers of high-end Unix workstations were government agencies. That wasn’t a secret. It was just the market.

The Valley was built on defense contracts

This part isn’t contested by anyone. Before the region was called Silicon Valley, its industrial spine was military procurement.

Hewlett-Packard’s early growth came from World War II demand for test instruments. Fairchild Semiconductor’s first major customer was the Minuteman missile guidance program. Through the mid-1960s, the Defense Department and NASA absorbed a large share of every integrated circuit produced in the United States. When there was no consumer market, the government was the only buyer willing to pay whatever the newest chip cost.

ARPANET — the internet’s direct ancestor — was a Defense Advanced Research Projects Agency project. GPS was military. The touchscreen, voice recognition, and lithium-ion battery research that ended up in the iPhone all had public funding absorbing the early risk. Economist Mariana Mazzucato built an entire book around this, The Entrepreneurial State, and the core argument has held up: the state took the losses on the risky decades, and private companies collected on the profitable ones.

The CIA literally started a venture capital firm

In 1999, the CIA founded In-Q-Tel. This is not a conspiracy theory. It’s an organization with a website and a public portfolio page.

One of its early bets was Keyhole, a satellite mapping startup. Google acquired it in 2004 and turned it into Google Earth. In-Q-Tel also appears on Palantir’s early investor list. More than 200 companies have taken its money, concentrated in databases, natural language processing, biometrics, and satellite imagery.

The mechanism is genuinely clever. Instead of developing technology in-house, the agency funds private startups and buys the output later. On the startup’s books it shows up as venture investment, not a government contract. Cleaner accounting, less press attention, and a founder who can tell an origin story that never mentions Langley. If something structurally similar happened in the 1980s, before In-Q-Tel existed to formalize it, it would be far harder to trace today.

The same structure is reassembling inside AI labs

If this stayed a 40-year-old story about Jobs, it would be trivia. What makes it worth attention is how precisely the pattern is repeating.

Since 2024, the major AI companies have been signing defense and intelligence contracts in sequence. OpenAI quietly revised the usage policy language that had prohibited military applications. Anthropic announced a partnership with Palantir and AWS to deploy Claude in classified environments. Meta authorized Llama for national security use. In 2025, the Defense Department signed agreements worth up to $200 million each with multiple AI firms.

Run the two claims side by side. One: training frontier models costs billions and nobody has a clear path to profitability. Two: a customer has appeared who pays consistently and doesn’t negotiate hard on price if the capability is real. That is the Fairchild-and-Minuteman arrangement with better branding. Early-stage costs the consumer market can’t absorb get carried by a buyer who cares about capability more than cost.

The useful version of this story

Reading all this as “Silicon Valley was secretly a spy operation” is the lazy interpretation. The sharper point is narrower and more damning: the genius founder in a garage narrative has functioned, for decades, as a device that erases where the money came from. Not through conspiracy. Through storytelling preference. Origin myths about capital allocation don’t sell books.

Whether the NeXT claim holds up or not, the fact that it’s this hard to check tells you something. It’s 40 years old, and the structure was designed not to leave a trail. The AI defense contracts, by contrast, are public. You can read the dollar figures. You can pull up the old usage policies and see exactly when the language changed.

So ask the question now, while the paper trail is still sitting in the open — not in 2065, when someone publishes the retrospective about whose money was really behind the model boom. The next version of the AI tool you use every day is being designed to satisfy someone. It’s worth knowing who.