How Arthur Rock Financed Two Silicon Valley Breakaways

In 1968, Robert Noyce and Gordon Moore left Fairchild Semiconductor, the company they had helped start eleven years earlier. Its chip business was thriving, but the two men believed its parent was putting too little of the proceeds back into research. They turned to Arthur Rock to help finance a company of their own. He had helped them leave an employer once before. Intel’s founding archive · Arthur Rock’s oral history

Noyce and Moore each put up $250,000. Rock helped raise another $2.5 million and became the company’s first chairman. The new business, Intel, sold no products in its first year. Intel’s account of the financing · Intel’s first-year results

The two financings trace a change in who could start a technology company and on what terms. In 1957, eight engineers got the money to set up shop by giving a corporate backer control and a right to buy the business. By 1968, two of those engineers could leave again, put up substantial sums of their own and raise the rest from a group of investors. Rock helped arrange both deals. The intervening years gave him something he could take to those investors: a record of what the founders had already built.

A buyer before a product

Rock was working at the New York investment bank Hayden, Stone in 1957 when a group of engineers at Shockley Semiconductor Laboratory sought help leaving. The eight included Noyce, a physicist, and Moore, a chemist. Rock and his colleague Alfred “Bud” Coyle helped find a backer willing to finance the group as a new business. They found one in Fairchild Camera and Instrument, an East Coast manufacturer. Computer History Museum’s account of the deal · Rock’s oral history

Fairchild Camera agreed to lend the new company $1.38 million. The founders bought shares, and some stock was set aside for future employees. The backer controlled the company through a voting trust and held an option to acquire all its stock for $3 million. The engineers could build a business together and share in an early payoff. If it worked, Fairchild Camera had already negotiated the price at which it could take full ownership. Computer History Museum’s corporate history · The financing terms

One of the first signs that the deal might work came from IBM. In early 1958, IBM needed a silicon transistor for a navigation computer and was having trouble finding one with the performance it required. Fairchild had yet to make a device. It secured an order for 100 transistors at $150 apiece. The order gave the engineers a specific problem to solve and a customer willing to pay if they solved it. Computer History Museum’s corporate history

Fairchild Camera exercised its option in 1959. By 1960, the semiconductor company’s sales exceeded $20 million, according to the Computer History Museum. The arrangement had made it possible for the engineers to leave Shockley together. It also left decisions about the chip business’s profits in the hands of a parent company with other businesses to run. How much of that money would go back into semiconductor research became a point of friction for Noyce and Moore. Computer History Museum’s Fairchild history · Intel’s founding archive

The second financing

By the time Noyce and Moore were ready to leave, Rock had also changed jobs. In 1961, he moved to San Francisco and formed Davis & Rock with Thomas Davis. The partnership backed businesses including Teledyne and Scientific Data Systems. Financing young technology companies had become a business Rock could pursue from California. Harvard Business School’s Arthur Rock archive · Science History Institute oral history

For Intel, Rock raised money through debt that could convert into stock. Investors had a way to participate in the company’s gains if it succeeded. Noyce and Moore had committed their own money, and Rock put in money himself. The new company had a board chaired by Rock and could decide for itself how much to spend developing its next products. Intel’s anniversary history · Gordon Moore and Arthur Rock’s joint oral history

The product opportunity was semiconductor memory. Computers stored data in arrays of tiny magnetic cores. Intel was betting that chips could take over that job. Its 1971 annual report described a market in transition and positioned the company to sell into it. That was a concrete commercial proposition: computer makers already paid for memory, and Intel had a different way to supply it. Intel’s 1971 annual report

Another opportunity came from a customer whose business was calculators. While designing chips for Japan’s Busicom, a team that included engineers from Intel and Busicom developed the 4004 microprocessor. Intel introduced it commercially in 1971. The calculator project gave the chip a first use. A processor that could be programmed to perform different tasks had applications beyond that order. Computer History Museum’s Busicom prototype · The museum’s history of the 4004

Rock later described his investing in terms of the people he backed. The Intel deal shows what that meant in practice. He had known Noyce and Moore for more than a decade. He had seen them help turn a group of engineers into a chip supplier with paying customers. Investors still had to take the risk that the team could do it again in a different market. They had evidence of how the founders worked well before Intel had anything to sell. Rock’s Harvard Business School interview

What cheaper computing made possible

Don Valentine saw another side of the semiconductor business. As a sales and marketing executive at Fairchild, he dealt with the people buying chips. At Sequoia Capital, he used that experience to look for markets that cheaper computing might open. Atari used microprocessors to sell games. Apple used them to sell computers to individuals. Once businesses had computers of different kinds, connecting them became a problem customers would pay companies such as Cisco to solve. Valentine traced those connections in an oral history with the Computer History Museum. Valentine’s oral history

By the 1980s, entrepreneurs were also trying to sell artificial intelligence. Companies such as IntelliCorp built software around rules drawn from human specialists. The promise was that a company could capture some of what an expert knew and apply it repeatedly. Making that work meant eliciting the rules, encoding them and maintaining them as the work changed. The people who built those systems described the commercial and technical problems in oral histories recorded by the Computer History Museum. Expert Systems Pioneer Meeting · Nils J. Nilsson’s history of AI

Rock’s account of his early investments offers a view of how he assessed founders and financed chip businesses. It leaves open how he would have evaluated an AI company. A modern deal offers a way to examine the question on its own terms.

What the money buys in AI

In August 2023, CoreWeave, which rents computing capacity, announced a $2.3 billion debt facility led by Magnetar Capital and Blackstone. CoreWeave said it would use the money to buy and pay for hardware for customer contracts it had already signed and to hire staff. For the lenders, those contracts offered a way to assess who would pay for the machines and what the revenue might be once they were operating. CoreWeave’s financing announcement

CoreWeave’s public offering prospectus later showed how quickly the business had grown and how concentrated its customer base was. Revenue rose from $229 million in 2023 to $1.9 billion in 2024. Microsoft accounted for 62% of 2024 revenue. CoreWeave recorded a net loss of $863 million that year. Signed demand helped the company finance more capacity. The figures showed how much of that demand came from one buyer and how much the company was spending as it expanded. CoreWeave’s IPO prospectus

CoreWeave’s facility was a loan to a company with signed contracts. Intel’s founding round paid for a team to develop products it had yet to sell. Both deals required a judgment about how a business would turn expensive equipment and engineering into something customers would buy. What counted as evidence differed. Rock could point to what Noyce and Moore had done at Fairchild. CoreWeave could show lenders the contracts it was buying machines to fulfill.

That distinction matters as AI changes the cost of building software. For a company selling computing capacity, the financing may pay for machines that must earn revenue for years. For a company using those machines to sell software, the costliest work may be getting a customer to adopt the product, connect it to existing systems and trust it with real work. Those are different risks, even when both companies describe themselves as AI businesses.

Before Intel made its first sale, Rock could explain why he trusted the people who would spend the money and what market they meant to enter. A prospective AI investor faces a version of the same question. What does the company need to do with this money, and what evidence is there that someone will pay for the result?