The great AI chip race

TDT | Manama

Email: mail@newsofbahrain.com

The biggest battle in artificial intelligence may not be happening inside chatbots. It is happening inside the chips that make them possible.

As AI models become larger and AI agents perform increasingly complex tasks, demand for specialised processors is exploding. Nvidia remains the most prominent player, with its Vera Rubin platform now entering production. Rather than selling a single accelerator, Nvidia is building complete AI systems combining GPUs, CPUs, networking, storage and software for massive at-data centres.

But competitors are closing in from different directions.

AMD is expanding its Instinct accelerator business into complete rack-scale AI systems. Its new MI400 series and Helios platform are designed to compete at data-centre scale, while AMD’s market value crossed $1 trillion this month as investor interest in its AI business surged.

Google has taken a different route, designing its own Tensor Processing Units. Its Ironwood TPU is built for large-scale AI training and inference and can be deployed in systems containing thousands of chips. Google is also developing its eighth-generation TPU systems specifically for agentic AI workloads.

In smartphones, Qualcomm is pushing AI processing directly onto devices. Its newly announced Snapdragon 8 Elite Extreme Gen 6 and 8 Elite Gen 6 platforms use a 2nm process and are designed for increasingly capable on-device AI agents, cameras and graphics.

China is building its own alternatives. Huawei is accelerating its Ascend AI chip roadmap as demand for domestic AI computing exceeds its current production capacity. The company is also developing systems capable of connecting enormous numbers of processors to compensate for the performance gap with Nvidia.

Alibaba has joined the race with its new Zhenwu V900, which the company says delivers three times the performance of its previous generation. It is also developing a new Qwen model planned at 5-10 trillion parameters, showing how chip development and AI model development are becoming increasingly interconnected.

The competition does not stop there. Amazon is developing its own AI accelerators, Intel continues its Gaudi programme, and a growing number of companies are designing specialised chips for particular AI workloads.

The real contest is now about more than who makes the fastest chip. Memory, networking, cooling, energy efficiency, manufacturing capacity and software ecosystems can determine how useful an AI processor becomes.

And there is a huge prize at stake. As AI moves from answering questions to running agents, robotics, scientific research and business operations, the companies controlling the computing infrastructure could influence how quickly the next phase of AI develops.

The AI race is becoming a chip race -- and the next winner may be decided not by one breakthrough processor, but by the entire system built around it.