Meta, Google, and Microsoft challenge Nvidia with AI chips to dominate the next decade

The rise of artificial intelligence (AI) is redefining the strategies of major technology players, driving them toward vertical integration reminiscent of the model that characterized IBM in the 1960s.

This move not only responds to the need to improve products and reduce costs, but also to avoid strategic dependence on single suppliers in the context of growing demand and a limited supply of specialized chips.

Bloomberg Intelligence estimates that the market for custom AI chips will reach $122.000 millions of dollars by 2033, reflecting the expansion prospects for those who master this technology.

How are Big Tech preparing to compete in the AI market

In this race, Meta reportedly began testing its own in-house AI chips to train models last year and acquired the startup Rivos to advance its custom semiconductor development.

Meanwhile, Google has consolidated its TPUs to the point that companies like Anthropic, OpenAI, and Meta itself have signed large-scale cloud agreements to access this technology. For Microsoft, the introduction of the next-generation Maia 200 chip marked a breakthrough after several months of delay.

These developments are not limited to chip design, According to Jonathan Atkin, an analyst at RBC Capital Markets, both Microsoft and Amazon are investing in dark fiber—underground fiber optic cables that remain unused until they are activated.

Atkin specified that, while Google and Meta own their own cables, they still rely on external providers to expand their connections. This type of investment allows companies to strengthen the infrastructure that connects their data centers and supports the growing demand that comes with expanding their cloud services.

Why Big Companies Are Focusing on Manufacturing Hardware

The current push by “hyperscalers”—Alphabet, Meta, Microsoft, and Amazon—reflects a concern similar to the one that motivated IBM: the high cost and scarcity of Nvidia chips has driven these companies to develop their own semiconductors.

According to analyst Jay Goldberg of Seaport, “Hyperscalers… recognize that there is a serious strategic danger in having a single AI computing provider. Therefore, they now have a strong strategic reason to manufacture their own silicon.”

Alphabet, a pioneer among the four, is already exploring the possibility of selling its physical TPUs to Meta, which could put it head-to-head with Nvidia in direct competition.

The decision to move forward with developing its own chips not only represents cost savings but also optimization according to the specific needs of the software and services each company offers.

How does this model relate to the one implemented by IBM in the mid-20th century

The logic behind in-house manufacturing of hardware and components harks back to the vertical integration that once gave decisive advantages to companies like IBM.

According to Carliss Y. Baldwin's analysis in the book *Design Rules*, by the mid-1980s this strategy allowed IBM to account for more than half of the computer industry's market value by producing its own mainframes and specialized components.

Likewise, IBM's model arose from the pursuit of a superior final product and the uncertainty surrounding the supply of early computers.

However, this dominance collapsed in the 1990s as a result of the decreasing cost of semiconductor production and the emergence of companies like Microsoft, focused on software, and Intel, a leader in microprocessors.

Google parent Alphabet (GOOGL) reportedly is in talks with Meta Platforms (META) and others to let them use its Tensor AI chips, expanding its Nvidia (NVDA) rivalry. Google and AI chip partner Broadcom (AVGO) rose late. Nvidia stock and Advanced Micro Devices (AMD) fell.Google has traditionally used its customer tensor processing units (TPUs) to its own data centers, renting them out to customers.