Deutsche Bank: China’s RMB 2 Trillion AI Data Center Gamble Faces Oversupply Cliff
Deutsche Bank warns that Beijing's planned RMB 2 trillion (US$277.8 billion) state-directed AI data center build-out could more than double China's computing capacity to 76GW by 2031 — but risks engineering a structural oversupply that crushes pricing and margins before the decade is out, with Alibaba shares already absorbing an 8.5% blow on fears of a telco-led cloud price war.
The report, published June 11, 2026 by Deutsche Bank analyst Peter Milliken, frames the investment plan — first reported by Bloomberg earlier this week — as a classic infrastructure boom-bust setup dressed in AI clothing. The government intends to channel funds through special-purpose vehicles, policy funds, and state-directed loans, with China's three major telecommunications carriers acting as project orchestrators and procuring more than 80% of core equipment — including semiconductors — from domestic suppliers. The 15th Five-Year Plan (2026–2030), ratified in March 2026 by the CPC Central Committee, explicitly mandates deepening the "East Data, West Computing" initiative and accelerating mega-scale cluster construction, providing the policy scaffolding for the spending surge.
The immediate market verdict was unambiguous: Alibaba, trading on both the Hong Kong Stock Exchange (9988.HK) and NYSE (BABA), shed 8.5% from Tuesday's close through June 10, closing at HKD 113.5 — well below Deutsche Bank's Buy-rated target price of HKD 190.0 (USD 195.0 for the US-listed ADR). The stock now sits near the lower half of its 52-week range of HKD 102.90–185.10.
Doubling Capacity in Five Years Triggers a Familiar Warning Signal
Deutsche Bank's arithmetic is straightforward but consequential. At an estimated development cost of RMB 50 million per megawatt — with hardware and semiconductors comprising 60% of that figure — a RMB 2 trillion (US$277.8 billion) program translates into approximately 40GW of incremental data center power capacity. Layered on top of China's existing 36GW base as of early 2026, the total supply footprint would reach roughly 76GW by 2031.
The problem, Milliken argues, is not the ambition but the arithmetic of demand. Private independent data center (IDC) operators currently add approximately 5GW per annum on their own trajectory. If state-directed and private-sector build-out proceed in parallel rather than in coordination, the market absorbs supply from two distinct pipelines simultaneously. Historical precedent in the Chinese data center market offers a cautionary note: every prior episode of capacity doubling within a five-year window has resulted in oversupply and earnings disappointment. The compounding factor in 2026 is Moore's Law-adjacent chip performance improvement — as each successive generation of AI accelerators delivers more tokens per server, the effective compute capacity of a fixed physical footprint expands annually, further pressuring utilization rates.
Deutsche Bank's base case is therefore not that the build-out fails, but that it succeeds too well. The bank expects IDC lease contracts rolling to lower pricing from the late 2020s onward, compressing returns for operators who locked in capital expenditure at today's elevated costs. GDS Holdings, the largest independent data center operator in China with a current share price of USD 33.48 against a Deutsche Bank target of USD 50.00, faces a capex-to-sales ratio projected to reach 70.3% in 2026E and 111.1% in 2027E — a capital intensity profile that leaves little room for error if pricing softens ahead of schedule.
Telcos Absorbing the "DeepSeek Dividend" Threatens Cloud Margin Stack
The more immediate equity risk, however, sits in the cloud and Model-as-a-Service (MaaS) segment. China's state-owned telcos have moved rapidly to integrate DeepSeek's open-source large language models into their cloud service offerings, creating what Deutsche Bank characterizes as a "telco + DeepSeek" competitive vector. The strategic logic mirrors the playbook from the traditional public cloud cycle: leverage infrastructure ownership and government procurement relationships to undercut private-sector providers on price, using subsidized capital costs as a structural weapon.
For Alibaba Cloud — the company's highest-margin growth engine and the pillar of Deutsche Bank's recovery thesis for fiscal years 2027–2029 — the threat is direct. Telcos historically dominate government and state-owned enterprise (SOE) accounts, precisely the customer segments where AI cloud and MaaS adoption is accelerating fastest under the Five-Year Plan mandate. Deutsche Bank's financial model already reflects a severe fiscal 2026 earnings trough: DB-adjusted EPS collapsed 61.5% year-on-year to CNY 3.15, EBITDA margin compressed to 9.0% from 17.7% in fiscal 2025, and free cash flow turned deeply negative at CNY -49.85 billion as net capex surged to CNY 126.1 billion.
The recovery narrative — DB EPS rebounding 76.5% to CNY 5.57 in fiscal 2027E and EBITDA margin recovering to 12.5% — is explicitly contingent on Alibaba Cloud defending its positioning against the telco encroachment.
Four Structural Buffers Prevent a Binary Collapse Scenario
Deutsche Bank is not calling a structural defeat for private cloud operators. Milliken identifies four countervailing forces that complicate the bearish narrative:
Open-source sustainability. The telco threat is partly contingent on DeepSeek or a comparable state-of-the-art model maintaining open-source licensing. Any shift toward proprietary models — whether driven by commercial incentives or regulatory pressure — would erode the telcos' cost advantage in model deployment.
Policy architecture remains undefined. The computing interconnection framework has explicitly incorporated private-sector participation in its design. How government procurement mandates interact with market-driven leasing — a distinction drawn directly in the 15th Five-Year Plan text — will determine whether telcos crowd out or coexist with private operators.
Customer segmentation provides a natural moat. Telcos' strength in government and SOE accounts is real but bounded. Private enterprises, foreign-invested companies, and digitally native businesses have historically preferred private cloud operators' service quality and customization capabilities. Differentiated positioning limits, though does not eliminate, competitive overlap.
Hardware self-sufficiency as a margin lever. Alibaba is not a pure software company. Its in-house AI chip development program — a capability that becomes more valuable as domestic semiconductor supply chains mature under the 80%+ localization mandate — provides a cost-control mechanism that pure-software MaaS providers cannot replicate. As chip performance improves, Alibaba's ability to capture those efficiency gains internally rather than passing them to customers represents a structural margin expansion lever through fiscal 2028–2029.
GDS Positioned as Near-Term Beneficiary, Long-Term Casualty
For GDS Holdings, the calculus is more nuanced. As a pure-play IDC operator with deep expertise in high-performance data center construction — expertise that telcos managing large-scale government projects will need to source externally — GDS is a logical partner or subcontractor in the state-directed build-out. The company's recently announced expansion plan is directionally consistent with this role, and Deutsche Bank maintains its Buy rating with a USD 50.00 target.
However, the long-term risk profile is equally clear. GDS's net debt stood at CNY 24.2 billion at end-2025, rising to an estimated CNY 23.4 billion in 2026E and CNY 34.0 billion in 2027E as capex accelerates. Net interest cover remains thin at approximately 1.2x through the forecast period. If IDC pricing softens materially from the late 2020s onward — as Deutsche Bank's oversupply thesis implies — GDS's heavily leveraged balance sheet leaves limited capacity to absorb a revenue shortfall.
The bank's summary framing is blunt: "Things get better before they get worse." The near-term window — government-backed project flow, favorable financing terms, and AI demand still in its exponential growth phase — provides a genuine earnings tailwind for both IDC operators and cloud providers willing to invest aggressively. The risk crystallizes later, when supply growth structurally outpaces demand and the pricing cycle turns.
Policy Timeline Anchors the Investment Thesis
The regulatory architecture underpinning the RMB 2 trillion program has been building for years. The February 2022 "East Data, West Computing" initiative established eight national computing hubs and ten clusters to redistribute computing resources from China's resource-constrained eastern regions to its energy-rich west. The May 2025 "Action Plan for Computing Power Interconnection", issued by the Ministry of Industry and Information Technology (MIIT), set binding targets: by 2026, establish a three-tiered national/regional/industry computing power interconnection platform; by 2028, achieve standardized interconnection of public computing power nationwide. The 15th Five-Year Plan, effective March 2026, elevated these objectives to the highest level of policy priority, explicitly authorizing government procurement as a demand mechanism to absorb the new supply.
The domestic chip supply chain assumption embedded in the 80%+ localization target is perhaps the most consequential — and least-tested — variable in the entire program. Deutsche Bank's report treats this as an expression of government confidence rather than a validated capability, noting that the plan's feasibility "suggests confidence that the domestic chip supply chain can handle such rapid expansion." Whether that confidence is warranted will be the central determinant of whether the RMB 2 trillion program delivers its intended 40GW by 2031 — or stalls at a fraction of that target, paradoxically protecting the market from its own oversupply risk.
Related Coverage:
Global ESS Battery Shipments Double as AI Data Centers and Geopolitics Rewire Supply Chains