Baidu unveils Tianchi 1024-card supernode, launching next year
11/12/2026 — 11/12, 12:20·1 sources (all)·1 reports (all)
Key takeaways
- At Baidu World 2026, Baidu unveiled its Tianchi 1024-card supernode, set to reach the market next year. A single system can support training…
Background & analysis
On November 12, 2026, during the opening ceremony of Baidu World 2026, Shen Dou, Executive Vice President of Baidu Group, unveiled the Baidu Tianchi 1024-card supernode and announced that this compute cluster will reach the market next year.
According to Baidu, the Tianchi 1024-card supernode is built on industry-leading optical engine technology, which triples the inter-card interconnect bandwidth over the previous generation. Compared with the prior product, training performance improves by 2.5x and inference performance by 3x. A single cluster can support training of large models with 5 trillion parameters, and the architecture is designed for horizontal scaling, allowing expansion to clusters of up to 10,000 cards. The companion G3.5 KV Cache memory acceleration cabinet strengthens cache reuse, boosting long-context inference performance by 2x. In multi-turn dialogue scenarios where the cache hits, TTFT drops by 90% and overall throughput rises by 7x.
The report also notes that Baidu had previously announced at last year's Baidu World conference that the Tianchi 1000-card-class supernode would launch in 2028, so the current timeline pulls the schedule forward by one year. The release of the Kunlunxin N-series has also been moved up to 2028. At present, the Tianchi 1024-card supernode is slated for market availability next year, and further details on its rollout are still pending from the company.
AI-generated from 1 reports · updated 1 day ago
- Heat index
- 39
- All sources
- 1
- All reports
- 1
- First seen
- 1 day ago
Reports on this story headlines open the original
At the Baidu World 2026 opening keynote, Baidu executive vice president Shen Dou unveiled the Tianchi 1024-card super node, which the company says will reach the market next year. Baidu claims a single unit can train models with up to 5 trillion parameters, with 2.5x training and 3x inference performance over the previous generation and 3x the inter-card interconnect bandwidth. An accompanying G3.5 KV Cache appliance is said to double long-context inference performance and cut time-to-first-token by 90% in cache-hit multi-turn scenarios.
IT之家AI score 53
Other stories people are talking about
- Heat index 2236Nvidia stock plunges after Jensen Huang resigns, with retail investors buying the dip35 sources
- Heat index 1463RisingGoogle Opens Gemini 4 Hydrogen to All Users as High-Speed Model21 sources
- Heat index 1429SurgeJensen Huang exits as NVIDIA CEO, leadership vacuum lingers16 sources
- Heat index 785SurgeAnthropic hires gas supply lead to power data center expansion after IPO10 sources
- Heat index 673SurgeMicrosoft races to fix global outage hitting Outlook, Teams and Copilot8 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

