Year in Review: Data-Center Chips Buck a Downturn

The year 2022 saw AI accelerators and DPU becoming mainstream in data center servers. Chiplets came to more CPUs while the chip power consumption reached a new high, necessitating a rethink in cooling.

Anand Joshi

The data-center-semiconductor market continued to grow in 2022. AMD furthered its gains in server-market share thanks to its lead in performance and power efficiency. The company began shipping its fourth-generation Epyc processor at the end of the year, and Intel deferred the general availability of its new Xeon to the start of 2023. Data processing units (DPUs) went mainstream through support from VMware and other prominent software companies.

The market for data-center AI chips continued to balloon, with Nvidia’s data-center business unit surpassing $3 billion in quarterly revenue in 2Q22. The company introduced Hopper, its latest AI GPU, offering a 2x performance boost over the previous generation, Ampere. Intel’s Habana subsidiary began shipping second-generation training chips, called Gaudi2, while Google introduced its fourth-generation Tensor Processing Unit (TPU).

AMD acquired Pensando, a DPU-chip maker, to gain a product portfolio that competes with those of Nvidia and Intel. Data-center semiconductors began adopting chiplet technology, and data centers raised their power consumption to a new high.

While AMD, Intel, and Nvidia continue to reap benefits from this market, we expect turmoil among startups in 2023. Shrinking valuation, higher performance standards, and rising development costs will lead to consolidation.