Friday, June 26, 2026
EN·DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

Broadcom and Marvell are advancing custom ASIC chips to compete with Nvidia GPUs, with Amazon developing proprietary processors.

Custom ASIC proliferation signals structural competition eroding Nvidia's GPU monopoly and forcing hyperscalers to build custom silicon for AI infrastructure.
Trade pressSlicast · August 28, 2025 · Global · Source: benzinga.com
importance 70

Speaking on Nvidia's fiscal second-quarter earnings call, CEO Jensen Huang dismissed concerns that hyperscalers' investments in application-specific integrated circuits could erode Nvidia's dominance in AI compute. Huang emphasized that "accelerated computing is unlike general-purpose computing. It's a full-stack co-design problem," highlighting the complexity of AI data centers and Nvidia's advantage in providing an end-to-end platform spanning data processing to inference. He noted that Nvidia's GPUs are supported in every cloud and computer company and are built to handle evolving AI model architectures, including multimodal systems. "So when you build a data center with Nvidia platform in it, the utility of it is best. The lifetime usefulness is much, much longer," Huang added.

Despite this confidence, Broadcom and Marvell are leading a surge in demand for custom AI accelerators. The ASIC market is forecasted to hit $30 billion in 2025, growing 30% annually. Marvell, holding 15% market share, has won designs for Amazon.com, Inc.'s Tranium 2 and Microsoft Corporation's next-gen chips, though its stock is down 41% year-to-date. Huang countered by highlighting Nvidia's performance-per-watt efficiency, which he said is unmatched and makes it a top choice for data centers constrained by power costs. "Perf per watt drives directly to revenues," he said.

Huang also highlighted Nvidia's broader ecosystem, which spans CPUs, GPUs, SuperNICs, NVLink interconnects, Spectrum-X Ethernet switches, and the newly unveiled Spectrum-XGS, designed for "AI super factories" with multi-gigawatt compute capacity. Nvidia reported second-quarter revenue of $46.74 billion, up 56% year-over-year, beating Wall Street's $46.02 billion estimate. Adjusted EPS came in at $1.05, and gross margins hit 72.7%. For the next quarter, Nvidia guided revenue between $52.92 billion and $55.08 billion, slightly above the Street's forecast, excluding any H20 chip sales to China.

Despite strong results, Nvidia shares fell 3.14% in after-hours trading following the earnings announcement.

Read the original
Broadcom and Marvell are advancing custom ASIC… · Slicast