N環球 N GLOBE
US Technology

Wired: Nvidia's Rivals Were Already Closing In Before Huang's Exit, as AMD Widens Its MI400 Lineup

· N GLOBE

Wired reports Nvidia was already facing growing competition from AMD, cloud giants' custom chips and inference specialists before Jensen Huang stepped down; AMD expanded its Instinct MI400 lineup the same day with the MI450X and the science-focused MI430X, claiming MI455X inference results up to 30% ahead of Nvidia's Vera Rubin — with caveats. Chip stocks fell in premarket trading ahead of Nvidia's earnings on Tuesday.

Nvidia was already facing intensifying competition before founder Jensen Huang resigned as chief executive, Wired reports: AMD's share of the AI accelerator market rose from 2 percent in 2022 to between 7 and 10 percent in 2026, Google, Amazon and Microsoft are each building their own chips, and inference specialists are gaining ground. On the same day, AMD expanded its Instinct MI400 lineup with the MI450X and the science-focused MI430X, moving to win orders while Nvidia's leadership is in flux.

Wired: rivals were closing in before Huang's exit

Wired cites Silicon Analysis data showing Nvidia still holds about 80 percent of AI accelerator sales, with revenue rising from $26.91 billion in the fiscal year ended January 2022 to $215.94 billion in the fiscal year ended January 2026. But the threats are coming from several directions, the report says: AMD keeps gaining share and earlier this year bought Taalas, which builds model weights directly into silicon; the three cloud giants are developing their own specialised chips to reduce their dependence on Nvidia; and as AI agents spread, inference — running already-trained models — matters more, an area where Nvidia's edge is smaller.

Patrick Moorhead, chief analyst at Moor Insights & Strategy, told Wired that inference gives specialised rivals an opening. Nvidia has responded, acquiring an inference-focused chip startup in December 2025. But Wired says Nvidia's real moat remains its CUDA software ecosystem: AI companies have long stayed on Nvidia hardware largely because switching costs are high — though startups now use AI to move models onto AMD hardware and serve them more cheaply, eroding that barrier.

AMD widens MI400 lineup, with caveats on the benchmark

AMD expanded the Instinct MI400 family the same day: the new MI450X carries 288GB of HBM4 memory, a smaller-memory option for AMD's Helios rack-scale platform, while the MI455X has 432GB; both offer 23.3 TB/s of peak memory bandwidth and the same core compute. AMD says the MI455X achieved 21,333 tokens per second per GPU running DeepSeek-R1 in an MLPerf Inference Offline configuration, with throughput up to 30 percent higher than Nvidia's published Vera Rubin NVL72 result.

The comparison comes with qualifications: AMD set its own October 2026 laboratory measurement against Nvidia's earlier MLPerf Inference v6.1 submission — not an independent, same-conditions head-to-head test — and the Offline scenario measures batch throughput rather than latency for interactive use. The science-focused MI430X also pairs 432GB of HBM4 with 23.3 TB/s of bandwidth and adds native FP64 double-precision compute. AMD says software compatibility, networking efficiency and sustained performance will decide whether the MI400 range is a practical alternative to Nvidia infrastructure.

Chip stocks fall in premarket; market awaits Nvidia earnings

Chip stocks fell in Monday's premarket trading: AMD dropped 8.75 percent on Friday and slipped further; Intel fell 2.52 percent on Friday and was down more than 4 percent in premarket after a shareholder lawsuit alleged it failed to disclose how weak PC sales and excess inventory would hurt revenue; Nvidia fell 2.01 percent on Friday and slipped again, despite a Wall Street Journal report that the Trump administration plans to name Nvidia among more than two dozen US AI 'national champions.'

Attention now turns to Nvidia's earnings on Tuesday (November 17), the Trump-Xi summit in Shenzhen and Anthropic's planned listing later this week for signals on chip demand. Interim chief executive Colette Kress, Nvidia's chief financial officer, has led the company since Huang resigned on November 12; the search for a permanent successor is still under way.

Sources: Wired via News In Las Vegas; PCQuest, November 16, 2026; TipRanks, November 16, 2026.

💬 Comments

No comments yet. Be the first!