Big Tech companies are starting to look like IBM in the 1960s

The race to dominate the burgeoning AI market is pushing tech giants to adopt business models reminiscent of IBM ( IBM ) in the 1960s.

Big Tech “hyperscalers” Alphabet ( GOOG , GOOGL ), Meta ( META ), Microsoft ( MSFT ), and Amazon ( AMZN ) are all in various stages of developing their own custom AI chips to put in their data centers and power their cloud and software offerings. Alphabet, the most distant of the four companies, is even in talks to sell its physical chips called TPU to Meta — a move that would put it head-to-head with chipmaker Nvidia (NVDA).

These efforts have led Bloomberg Intelligence analysts to predict the custom AI chip market will grow to $122 billion by 2033.

Big Tech component manufacturing goes beyond chips: Microsoft and Amazon are actively investing in dark fiber, or currently unused fiber optic cables that are already underground, RBC Capital Markets analyst Jonathan Atkin said in a recent note to clients. Google and Meta also own their own cables but still buy from third parties, he wrote. These cables are needed to connect companies’ data centers and the businesses that use them.

The dynamic in which cloud providers produce their own components (hardware) to run their core products (software) shows that Silicon Valley is turning to vertical integration – an operating model pioneered by oil and steel magnates in the late 19th century and adopted by IBM during the digital revolution.

IBM was one of the most successful vertically integrated companies in the 1960s when it made the hardware for its mainframes, or large computer systems. IBM’s strategy arose out of the idea that making its own specialized parts would improve the end product (mainframe) and profit margins—and amid concerns about parts shortages for early computers. It worked: In 1985, the company accounted for more than half the market value of the computer industry, noted Carliss Y. Baldwin in her book “Design Rules.”

Of course, it all came crashing down later. In the 1990s, falling semiconductor manufacturing costs — as well as the rise of Microsoft’s software power and chip leader Intel — dug into IBM’s once-formidable competitive moat, and the company no longer claimed to be vertically integrated until 2000, Baldwin said.

Just as the advent of computing pushed IBM toward vertical integration, the popularization of AI since the launch of ChatGPT in late 2022 has put today’s cloud giants on a similar trajectory. In particular, the steep costs of Nvidia chips and their limited availability have pushed the tech giants to advance their AI chip efforts. Those custom chips are cheaper and better optimized for companies’ software.

Nvidia founder and CEO Jensen Huang holds up a Rubin GPU and a Vera processor during the CES technology show on Jan. 5, 2026, in Las Vegas. (AP Photo/John Locher) · THE ASSOCIATED PRESS

“Hyperscalers … recognize that there is a serious strategic danger in having a single AI computing provider,” said Seaport analyst Jay Goldberg. “And so now they have a very strong strategic reason to make their own silicon.”

Meta began testing an in-house AI chip for training models last year and recently acquired chip startup Rivos to accelerate its custom semiconductor efforts. Google’s TPUs have become so advanced that Anthropic (ANTH.PVT), OpenAI (OPAI.PVT) and even rival Meta have signed major cloud deals with the company to access them. And after a long delay, Microsoft unveiled its next-generation Maia 200 chip in January.

During Yahoo Finance’s recent visit to Amazon’s chip lab and nearby data center in Austin, Texas, the company showed off its latest UltraServers, a cluster of servers that includes Amazon’s next-generation internal GPU called Trainium, Graviton CPUs, and custom network cables and switches that connect them. Amazon still sells more AI computing in its remote data centers powered by Nvidia GPUs than in custom accelerators, but the tech giant is increasingly touting the advantages of its in-house hardware.

Amazon Web Services CTO Paul Roberts told Yahoo Finance that its Trainium3 chip can offer its cloud customers up to a 60 percent price-to-performance benefit compared to GPUs for inference workloads.

“I think what we’re seeing in the market is a lot of validation of this approach [of making custom chips] — versus sort of generic GPUs — now you can have these specialized processors and accelerators that achieve incredible power efficiency savings,” he said.

These types of energy savings will become a bigger deal as the AI ​​data center boom begins to feel the effects of energy constraints.

But Seaport’s Goldberg believes the move toward vertical integration is reaching its “far limit” and not all Big Tech players will succeed.

“If you want to design a high-end chip, that’s a massive expense,” he said, adding that “only so many companies can afford it.”

StockStory aims to help individual investors beat the market.
StockStory aims to help individual investors beat the market.

Click here for the latest technology news that will impact the stock market

Read the latest financial and business news from Yahoo Finance

Leave a Comment