China's inference build-out just got a number. bloomberg reports deepseek is building a ~1gw data…
china's inference build-out just got a number. bloomberg reports deepseek is building a ~1gw data center in ulanqab with at least 160,000 huawei ascend 950dt chips, a ~$2.56 billion order.
training stays on nvidia, but inference is going domestic.
Context
Bloomberg reported on September 4, 2026, citing people familiar with the matter, that DeepSeek plans to deploy at least 160,000 of Huawei's Ascend 950DT accelerators at a data center it is building in Inner Mongolia, to operate its models, with the installation timeline depending on Huawei's production capability. An earlier Bloomberg report on July 30, 2026 said DeepSeek is looking to add one gigawatt of compute in Ulanqab, about 350 kilometers northwest of Beijing, and plans to build its own facility while leasing additional capacity.
Startup Fortune, a secondary outlet, puts the chip commitment at about $2.6 billion and says component shortages could keep Huawei from filling the order for more than a year.
The 160,000 chips, the Ulanqab site and the roughly one gigawatt match Bloomberg. All of it comes from unnamed people familiar with the matter and describes plans, so 'is building' is accurate for the data center in the reports, while the chip order is described as planned.
The $2.56 billion figure was not found in a Bloomberg page read. A secondary outlet gives about $2.6 billion, so the exact figure is unsupported here, not refuted.
'Training stays on Nvidia' goes further than the Bloomberg text read, which says DeepSeek does not currently plan to use the 950DT for training. That is a statement about this chip, not about where all training happens, so the rest is unsupported here, not refuted.
'Inference is going domestic' is the author's line. The report says the chips are for operating DeepSeek's models and that delivery depends on Huawei's production.
Related work
- DeepSeek Plans Big Huawei AI Chip Order to Power New Data Center (Bloomberg, September 4, 2026) ↗Source for the 160,000 Ascend 950DT chips, the Inner Mongolia site and the inference-not-training detail.
- DeepSeek Is Developing Massive AI Data Center in Inner Mongolia (Bloomberg, July 30, 2026) ↗Source for the one gigawatt of compute in Ulanqab.
- DeepSeek Bets $2.6 Billion on Huawei Chips It May Wait a Year to Get (Startup Fortune) ↗Secondary source for the roughly $2.6 billion figure and the supply delay.
Watch next
- Find Bloomberg's own price figure for the order. Watch for Huawei's reported 950DT production volume.
Sources
- DeepSeek Plans Big Huawei AI Chip Order to Power New Data Center (Bloomberg, September 4, 2026)news.bloomberglaw.com
- DeepSeek Is Developing Massive AI Data Center in Inner Mongolia (Bloomberg, July 30, 2026)bloomberg.com
- DeepSeek Bets $2.6 Billion on Huawei Chips It May Wait a Year to Get (Startup Fortune)startupfortune.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 9 October 2026 at 09:04 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →