Xi's BRICS AI Coalition Hands Nvidia, Cisco and the US Hyperscalers a $50B Sovereign-Compute Ceiling—And the China Revenue Is Already Zero

NVIDIA's Q2 FY2027 results, filed August 26, 2026, contain the most consequential number for the BRICS thesis: Data Center revenue of $89.0 billion, up 117% year-on-year, with the company's own outlook assuming zero Data Center compute revenue from China. CFO Colette Kress's August 26 commentary was explicit — the $108 billion Q3 guidance range explicitly omits any Chinese Data Center compute contribution. The China gap is not hypothetical; it is in the model.

The structural read: NVIDIA no longer needs China to grow — the91%-of-revenue Data Center segment is locked into US hyperscaler and EMEA/Sovereign-AI build-outs. But China going to zero is now permanent, and the Sep 12 BRICS declaration pushes11 member governments into a procurement framework where 'sovereign AI' is a politically acceptable reason to buy Huawei, Cambricon or Alibaba chips rather than NVIDIA. The next wave of sovereign-AI capex in India, the UAE and Indonesia is the marginal buyer.

The Chinese alternative is no longer a slide deck. Huawei told the FT and Reuters it expects2026 AI chip revenue of approximately $12 billion, up 60% from $7.5 billion in 2025, driven by the Ascend 950PR (mass production March 2026,1.56 PFLOPs FP4) and the upcoming 950DT in Q4. The total company reported revenue of CNY 880.9 billion in 2025 (~$127.5B), with cloud computing including AI chips at CNY 72.08 billion. Cambricon, the pure-play listed AI accelerator, posted H1 2026 revenue of RMB 5.996 billion, up 108% year-on-year, on net profit of RMB 2.31 billion (+122%). Alibaba's Q1 FY2027 cloud segment grew external revenue 45% year-on-year — a 22-quarter high — with AI products now 35% of cloud revenue and an annualized run-rate of RMB 49.5 billion.

The performance gap is real but narrowed. The Carnegie Endowment's June 8, 2026 'Compute Coalition' report estimated China reached only 1–4% of US AI chip production in 2025 and will hit only 1–2% in 2026 — a structural constraint that the Sep 12 proposal explicitly tries to work around via shared LLM training and open-source stacks. Bruegel's July 22, 2026 analysis confirms Huawei's Ascend lags on memory bandwidth and developer tools, requiring 2–4× more power for equivalent output. The strategic response: open-source CANN (Aug 2025), a PyTorch-compatible torch_npu plugin, and DeepSeek-aligned models that run on both NVIDIA and Huawei hardware.

The marginal AI buyer over the next 36 months is not a hyperscaler — it is a sovereign government. JLL's mid-2026 India data-center outlook reports capacity will triple from 1.6 GW in mid-2026 to 6 GW by 2029, requiring USD 110 billion in cumulative investment. PwC's September 2026 forecast sees global data-center capex rise from roughly $800 billion per year in 2026 to $1.8 trillion per year. The global sovereign-AI market — data sovereignty, national AI infrastructure, multilingual models — is projected by Global Market Research to expand from $50.2 billion in 2026 to $249.5 billion by 2035.

The BRICS open-source AI proposal directly attacks the procurement framing. India's Jitin Prasada noted at the Sep 12 summit that BRICS countries 'will need massive data-centre infrastructure as AI becomes more important' — a request that the open-source community proposal gives India a politically defensible reason to route to non-US suppliers. The Sep 12 declaration's commitment to implement the BRICS Leaders' Statement on Global Governance of AI — first issued July 6, 2025 — was followed by China's July 17 launch of the World AI Cooperation Organization (WAICO) with 29 member countries, which Xi Jinping invited BRICS members to join.

The BRICS proposal reshapes the supply chain on both sides of the chip. Upstream, TSMC's CoWoS advanced-packaging capacity is fully booked through 2026, with NVIDIA holding roughly 60% of allocation and the top three customers controlling more than 85%. That lock is a defensive moat for NVIDIA — Chinese accelerators cannot buy their way into CoWoS at scale. But downstream, the data-center networking layer is exposed: Cisco's Q4 FY2026 revenue was $17.3 billion (+18% YoY) with $4 billion in AI infrastructure orders, and Arista Networks's Q2 2026 revenue of $3.04 billion (+37.7% YoY) drove management to raise its 2026 AI networking target to $3.25 billion. Both depend on hyperscaler and enterprise build-outs in the same regions the BRICS proposal targets.

The non-obvious read: the supply-chain losers are not the chip vendors, but the data-center networking and cloud-software layers whose revenue attaches to hyperscaler build-outs in the same geographies the Sep 12 BRICS declaration now targets. Cisco reports its Americas region at 59.7% of FY2026 revenue ($33.7B of $56.5B), EMEA at 26.2% ($14.8B), and APJC — which includes India, Indonesia, Saudi Arabia and the UAE — at the balance. A coordinated BRICS+ sovereign-AI procurement shift redirects the marginal networking dollar away from Cisco and toward Chinese or domestic alternatives, including Huawei's own networking stack.

The thesis the brief flags — that Xi Jinping's Sep 12 BRICS bid forces a sovereign-compute race the US cannot win by default — is supported by the data, but the conclusion is more nuanced than the framing. NVIDIA's Data Center revenue grew117% year-on-year in Q2 FY2027 to $89.0 billion, with China already at zero and the company still guiding Q3 to $108 billion. The CoWoS packaging bottleneck means the BRICS+ bloc cannot physically displace NVIDIA on advanced training silicon in the 12-month window. What the Sep 12 BRICS proposals actually do is reframe the next decision point: when India, the UAE, Saudi Arabia and Indonesia buy sovereign-AI capacity over 2027–2029, the procurement framing is no longer 'best available chip' but 'aligned stack.'

The investors' question is sequencing. In the next two quarters, the numbers favor NVIDIA and the US networking/cloud incumbents — Arista Networks's raised AI target and Cisco's $4B in AI infrastructure orders confirm the demand. Over1–3 years, the question is whether the BRICS+ bloc can build a coherent alternative stack fast enough to capture meaningful sovereign-AI share. The Carnegie Endowment estimate that China will reach only 1–2% of US AI chip production in 2026 is the binding constraint; the Sep 12 BRICS proposal is the political workaround. Watch Cambricon's order book, the rollout of WAICO centers, and India's first major sovereign-AI procurement decision in 2027.