Meta to spend billions on AMD chips, challenging Nvidia's monopoly

Six gigawatts. To put that in perspective, it is roughly the amount of electricity consumed by several million households. That is the volume of server capacity Meta Platforms intends to deploy on AMD processors, under a deal with the chipmaker that could become a turning point in the battle for the AI infrastructure market. Besides buying hardware, Meta will also acquire AMD shares, a move that goes well beyond an ordinary commercial relationship and looks more like a strategic alliance.

To grasp why this matters, it helps to recall the last two years. Since generative AI took off, Nvidia has become the absolute monopolist in accelerators for training and inference of neural networks. Its H100 GPUs, and then the latest Blackwell, became the de facto industry standard. Nvidia's market capitalization at times exceeded three trillion dollars, and waiting lists for its chips stretched for months. AMD, historically the second player in GPUs, tried desperately to catch up with its Instinct MI300 line and later models, but its share of the AI accelerator market remained disproportionately small: by various estimates, five to ten percent.

That is why the contract with Meta is not just a big order. It legitimizes AMD as a serious alternative to Nvidia in the eyes of the whole industry. Meta is one of the world's largest consumers of AI compute. It trains its own Llama family of models, runs recommendation algorithms for billions of users of Facebook, Instagram and WhatsApp, and invests heavily in the metaverse and wearables with AI features. When a customer of that size bets on AMD, it sends a powerful signal to the rest of the market.

Technically, Meta's decision makes deep sense. Dependence on a single supplier is one of the main headaches for any large technology business. Nvidia, aware of its dominant position, sets prices accordingly. Diversifying suppliers gives Meta not only a fallback in case of supply disruptions but also strong leverage in price negotiations with both manufacturers. In addition, in a number of scenarios AMD processors deliver competitive performance at a more attractive ratio of price and energy efficiency, a critical parameter when you are talking about six gigawatts of power draw.

Meta's decision to buy AMD shares deserves separate attention. The practice is not new in the tech industry: consider Microsoft's investments in OpenAI or Amazon's in Anthropic. Buying a supplier's shares creates an additional financial tie and a mutual interest in success. For AMD, this means not only an inflow of capital but also long-term confidence that a key customer is not going anywhere. For Meta, it means a potential financial gain if AMD's stock rises thanks to a broader customer base.

The consequences of this deal reach well beyond the two companies. For Nvidia, it is the first truly alarming wake-up call. Yes, Jensen Huang's company still controls the lion's share of the market, and its CUDA software ecosystem remains practically without alternative for many developers. But if other hyperscalers such as Google, Amazon and Microsoft follow Meta's example and start buying more AMD hardware, price pressure on Nvidia will grow. For the AI industry as a whole, this is a positive signal: competition in chips means lower compute costs, which ultimately speeds up the development and democratization of artificial intelligence.

For AMD, the Meta deal is a moment the company has awaited for years. Under Lisa Su, the chipmaker has methodically built out its AI accelerator lineup, invested in ROCm software as an alternative to CUDA, and forged partnerships with cloud providers. Now those efforts are starting to bear fruit on a scale that could change the company's financial profile. Billions of dollars of guaranteed revenue from a single customer is a foundation on which further expansion can be built.

The AI infrastructure market is entering a new phase. The era of unconditional dominance by one supplier appears to be ending, not because Nvidia has gotten worse, but because the stakes have become too high to depend on a single source. Meta's six gigawatts on AMD processors are not just a commercial deal. They are a statement about what AI infrastructure will look like over the next five years: more competitive, more diversified and, possibly, more accessible to everyone.

*Meta has been designated an extremist organization and is banned in Russia.