A national compute-coordination layer is becoming visible.
New standards evidence moves the thesis beyond provider-level control planes toward monitoring, scheduling, billing/trading and compute–electricity coordination across the national compute network.
Still unproven: nationally fungible compute, routine cross-region workload optimisation and commercial returns.
National coordination and compute–electricity standards sharpen the architecture
New National Integrated Computing Power Network projects specify interfaces for monitoring, scheduling, billing/trading and compute–electricity coordination. Separately, GDS has participated in the launch of Hebei's operational provincial monitoring/scheduling platform, while Huawei's new 3D AIDC and compute-operations stack provides an increasingly end-to-end reference architecture.
Thesis change: the site now distinguishes provider-level control planes from an emerging national compute-coordination layer, and treats data-centre load flexibility as an explicit research branch of compute–energy convergence. This is stronger evidence of system integration, but not proof of nationally fungible compute or commercial returns.
LATEST THESIS RUN · 21 SEPTEMBER 2026
Evidence reviewed through 21 September 2026. Individual items retain their actual event dates and evidence qualifications. This is a reviewed research record, not an automatic news feed.
New evidence enters once — then propagates through the thesis.
This is the site's living research layer. Consequential developments are classified by evidence state, mapped to architecture layers and tested against the system-level thesis. When evidence changes an existing conclusion, the affected architecture, company and economics sections are revised rather than leaving the finding isolated in a news archive.
A coordination layer above provider control planes is becoming more explicit.
What changed: national-network projects now specify monitoring, scheduling, billing/trading and compute–electricity coordination; GDS separately provides an evidenced private-AIDC ↔ provincial-scheduling edge in Hebei.
Propagation: Control Plane ↑ · Compute × Energy ↑ · National Architecture ↑ · GDS relationship added. No VNET/KC valuation change.
Architecture change → · Control plane → · Power & energy → · GDS →
KC’s Huawei deployment evidence was missing from the site.
The 10 September hospital disclosure names Huawei and Kingsoft Cloud in a local AI platform. This corrects an outdated site label; it is not a new announcement today. The 17 September KC thesis already incorporated it.
Thesis effect: strengthens the evidence for software operating above third-party compute. Revenue materiality, repeatability and cash returns remain unproven.
Full KC thesis and source links →New DRAM mass production strengthens execution evidence; HBM relief remains unproven.
What changed: CXMT reports fifth-generation DRAM mass production and two 24Gb LPDDR5X products. These are mobile-memory products. Its claim of at least 50% more gross dies per wafer excludes defective dies.
Thesis effect: stronger domestic manufacturing execution; no demonstrated reduction in AI-server memory costs. Yield, customer qualification and cost per usable die remain the next tests.
Company effect: no direct KC/Alibaba procurement or VNET/GDS deployment link established.
Reuters • 20 September · Chinese-language report · Dashboard impact
Capacity growth needs a paid-demand test.
PBOC adviser Huang Yiping warned AI could deepen China’s strong-supply, weak-demand imbalance. This strengthens the case for monitoring utilisation and cash collection; it does not establish a decline in cloud demand or a new funding programme.
SuperPoD is becoming a programming abstraction, not merely a larger cluster.
What changed: Huawei exposed memory/message communication semantics, PTO ISA and developer access to very large Ascend resources, extending system design upward into the runtime.
Connection: UnifiedBus → SuperPoD → PTO/runtime → AICS/control plane → MaaS/agents.
Still unproven: CUDA-like developer depth, realised portability and independent cost-per-useful-task economics.
NPO moves closer to the compute boundary.
What changed: Huawei describes Hi-ONE NPO as mass-production ready and is pushing an NPO implementation agreement into the standards ecosystem.
Connection: reinforces the site's interconnect thesis: optics migrates inward as scale-up systems become larger and denser.
Still unproven: sustained hyperscale reliability and manufacturing economics.
Energy geography and compute geography are being designed together.
What changed: the Xinjiang-to-Chongqing intelligent-compute project provides a physical example linking western renewable resources, remote compute, telecom connectivity and eastern cloud demand.
Connection: East Data–West Compute becomes a demand-to-energy architecture rather than a map of data-centre locations.
Still unproven: routine real-time workload migration according to electricity price or availability.