Nvidia's next-generation AI rack system has been delayed to 2028 due to manufacturing snags, according to a SemiAnalysis report cited by CNBC. The slip pushes back the successor to Nvidia's current rack-scale architecture, which underpins much of the world's AI training and inference.
Key facts
- Next-gen Nvidia AI rack delayed to 2028.
- SemiAnalysis report cited by CNBC flags manufacturing snags.
- Nvidia did not comment on the report.
- Delay follows Nvidia renting back GPU capacity from neoclouds.
- Rivals AMD and Cerebras compete for data center business.
Nvidia's next-generation AI rack system has been delayed to 2028 due to manufacturing snags, according to a SemiAnalysis report cited by CNBC. The delay, first reported by SemiAnalysis and covered by CNBC, pushes back the successor to Nvidia's current rack-scale architecture, which underpins much of the world's AI training and inference. Nvidia did not immediately respond to a request for comment on the report, the outlet added. The company did not disclose the specific manufacturing issues or the exact nature of the delayed product.
The delay comes as Nvidia faces broader demand softness, with the company recently renting back GPU capacity from neoclouds. This contrasts with the narrative of perpetual GPU shortages that dominated 2024 and 2025. Rivals including AMD and Cerebras Systems are competing for Nvidia's data center business, while Google continues to develop its own TPU line, with the next TPU codenamed Humufish using Intel EMIB-T packaging.
Key Takeaways
- Nvidia's next-gen AI rack delayed to 2028 on manufacturing snags per SemiAnalysis.
- Delay benefits AMD and custom silicon rivals.
Why This Delay Matters
For hyperscalers planning 2027 data center builds, a 2028 rack system arrival means they must lock in current-generation configurations or bet on alternative architectures. The delay benefits AMD's Instinct line and custom silicon efforts from Google, Amazon, and others. SemiAnalysis has previously projected 40GW+ behind-the-meter data center capacity by 2028, and any hiccup in Nvidia's roadmap could reshape that outlook.
The manufacturing snag also highlights the challenges of scaling beyond Blackwell. Nvidia's previous generation, Blackwell, saw strong demand but also faced initial production constraints. The delay suggests that the next leap in rack-scale integration—likely involving tighter GPU-to-GPU interconnects and advanced cooling—is proving harder than anticipated.
What to Watch
Watch for Nvidia's next earnings call, expected in August 2026, where CEO Jensen Huang may address the roadmap. Also track hyperscaler capital expenditure announcements for 2027–2028, as they will reveal whether customers are shifting orders to AMD, Intel, or custom silicon. A confirmed delay to 2028 would give AMD and Cerebras a wider window to capture share in the high-end training market.
Source: news.google.com
