China AI Infrastructure: AI-compute demand supports Range and Sangfor earnings, while valuation resets lower their target prices

Overview

The report examines how China’s AI-compute boom affects compute-rental and infrastructure-software suppliers. HSBC favors Range for GPU-as-a-service and data-centre capacity exposure, and Sangfor for cloud, storage and cybersecurity demand, while lowering both valuation multiples because near-term AI-driven growth may not sustain a broad re-rating.

Core views

HSBC argues that AI training and inference workloads, together with constrained supplies of advanced accelerators, are creating a simultaneous volume and price upcycle in compute rental. US export restrictions and extended delivery lead times have widened the supply gap, while domestic large-language-model workloads have generated incremental demand. A-share companies have disclosed RMB72-79bn of compute-rental orders since May 2026, which HSBC estimates is equivalent to roughly 10,000 H100 eight-GPU server rental orders. GPU-server rental prices have also increased markedly in China and the US: Silicon Data’s B200 on-demand rate rose 30% year to date, while Nebius raised B300 on-demand pricing by approximately 56% cumulatively in 2026. HSBC sees providers with sufficient data-centre capacity and delivery capability as the principal beneficiaries. For Range, HSBC raises 2026-28 revenue estimates by 9%, 6%, and 12%, respectively, reflecting higher GPU-rental prices and higher-than-expected 1H26 capital expenditure. It cuts projected gross margin by 2ppt, 3ppt, and 3ppt because more GPU servers increase depreciation, leaving net-profit estimates broadly unchanged but lifting EBITDA estimates by 11%, 8%, and 13%. The revised 2026-28 revenue forecasts are RMB8.545bn, RMB11.169bn, and RMB15.864bn, while EBITDA forecasts are RMB5.566bn, RMB7.746bn, and RMB11.409bn. These estimates are above consensus, particularly by 22% for 2028 revenue and 27% for 2028 EBITDA, because HSBC is more positive on compute rental, IT-capacity expansion, technology advantages and Range’s energy quota. The report nevertheless lowers Range’s target valuation multiple. It uses an EV/EBITDA approach based on an adjusted historical average of 17x, excluding observations more than one standard deviation above the original historical average. HSBC forecasts 2026-28 EBITDA CAGR of 43%, versus 36% in 2021-26, and applies 21x 2027e EV/EBITDA, down from 28x previously. Although 2026e EBITDA growth is expected to be about 60% year on year, HSBC views the surge in GPU rental demand and pricing as cyclical and potentially unsustainable once domestic GPUs improve and supply eases. On 2027e EBITDA of RMB7.7bn, up from RMB7.1bn, HSBC sets a RMB84.00 target price, down from RMB111.00, implying 24% upside, and maintains Buy. For Sangfor, rising memory and GPU costs are expected to encourage enterprises to improve hardware utilization through hyperconverged infrastructure and distributed storage. Sangfor’s 1H26 cloud-computing revenue rose 58.6% year on year, supported by HCI and distributed-storage demand. Its cybersecurity business also returned to positive growth, helped by demand recovery for mature products, AI features, replacement cycles, and AI-generated requirements for its AI security platform, managed security services and SASE offerings. HSBC raises Sangfor’s 2026-28 revenue forecasts by 10%, 13%, and 14%, citing AI-driven HCI and enterprise distributed-storage demand, faster VMware replacement amid Sino-US technology decoupling, and a better cybersecurity outlook. It lowers projected combined selling, administrative and R&D expense ratios by 5ppt, 4ppt, and 3ppt on expected cloud-business scale benefits and continued expense control. The resulting net-profit estimate increases are 82%, 57%, and 31%, to RMB847m, RMB991m, and RMB1.115bn for 2026-28. HSBC’s forecasts exceed Wind consensus: 2026-28 revenue is 5.0%, 4.8%, and 3.0% higher, while net profit is 35.3%, 20.3%, and 9.3% higher. It also calculates that a 1ppt gross-margin move changes net profit by 13-20% in the same direction. HSBC values Sangfor with a price-to-sales framework. Its historical average 12-month forward PS multiple since 2022 is 5.1x; based on projected 2025-28 revenue CAGR of 18%, versus 14% over 2018-25, it applies a 6.6x 2027e PS multiple, down from 8x previously. With 2027e revenue of RMB11.5bn, up from RMB10.2bn, HSBC sets a RMB176.00 target price, slightly below RMB180.00 previously, for roughly 24% upside, and maintains Buy. The report expects stronger HCI and software-defined-storage demand, alongside expense control, to support earnings recovery and potentially further valuation re-rating.

Analysis framework

HSBC first links AI workload growth and accelerator supply constraints to GPU-rental volumes and prices, using disclosed rental orders and market pricing evidence. It then revises company operating forecasts by segment, compares its estimates with consensus, applies sensitivity analysis, and values Range on EV/EBITDA and Sangfor on price-to-sales multiples referenced to historical averages and forecast growth.

Methodology notes

• Industry AnalysisSupply-demand framework

  GPU compute-rental supply-demand analysis

  The report connects constrained advanced-accelerator supply and expanding AI training and inference demand to rising GPU rental prices and capacity utilization.

• Valuation methodsEV/EBITDA valuation

  Range EV/EBITDA valuation

  HSBC applies a 21x 2027e EV/EBITDA multiple, derived from Range’s adjusted historical average and relative EBITDA-growth outlook, to determine its target price.

• Valuation methodsPS valuation

  Sangfor price-to-sales valuation

  HSBC applies a 6.6x 2027e PS multiple using historical forward PS and revenue-growth comparisons to set Sangfor’s target price.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Range (300442 CH)

  Direct beneficiary of GPU rental demand, data-centre expansion and rising GPUaaS prices.

  Strengths

  Technology advantages, sufficient energy quota, data-centre capacity and delivery capability.

  Weaknesses

  Higher GPU-server depreciation reduces projected gross margin.

  Comparison

  HSBC forecasts 2026-28 revenue and EBITDA above consensus, with 2028 estimates 22% and 27% higher, respectively.

  Risks

  Competition, client concentration, order cancellations or delays, and tighter data-centre energy-quota policies.

• Sangfor (300454 CH)

  Beneficiary of AI-driven demand for HCI, distributed storage, cloud computing and AI-enabled cybersecurity.

  Strengths

  1H26 cloud-computing revenue grew 58.6% year on year; cybersecurity demand has resumed growth; expense control may create scale benefits.

  Weaknesses

  Earnings are sensitive to gross-margin changes, with a 1ppt move affecting net profit by 13-20%.

  Comparison

  HSBC’s 2026-28 net-profit estimates exceed Wind consensus by 35.3%, 20.3%, and 9.3%.

  Risks

  Intense competition, weak macroeconomic conditions affecting SME demand, and chip shortages affecting hardware supply or pricing.

Key data

• Announced compute-rental ordersRMB72-79bn since May 2026HSBC estimates this implies about 10,000 H100 eight-GPU server rental orders.
• GPU rental-price changesB200 +30% YTD; Nebius B300 +c56% cumulatively in 2026Evidence supporting a compute-rental price upcycle.
• Range 2026-28 EBITDA estimate changes+11%, +8%, +13%Driven by higher GPU-rental pricing and capacity expansion.
• Range target priceRMB84.00Lowered from RMB111.00; implies 24% upside.
• Sangfor 1H26 cloud-computing revenue growth+58.6% y-o-ySupported by HCI and distributed-storage demand.
• Sangfor 2026-28 net-profit estimate changes+82%, +57%, +31%Reflects stronger revenue assumptions and lower expense-ratio forecasts.
• Sangfor target priceRMB176.00Lowered from RMB180.00; implies c24% upside.

Impact & implications

The report identifies Range as a direct beneficiary of higher GPU-rental utilization and pricing, and Sangfor as a beneficiary of enterprise demand for more efficient software-defined infrastructure. However, HSBC’s lower target multiples indicate that stronger earnings forecasts do not automatically translate into sustained valuation expansion, particularly if GPU supply conditions normalize.

Risks

• For Range, stronger-than-expected competition could pressure pricing, margins and long-term contract wins.
• Range’s reliance on a small number of large cloud-service clients creates contract-renegotiation and usage-reduction risk.
• Range faces order-cancellation, project-delay and energy-quota-policy risks that could constrain utilization and capacity expansion.
• For Sangfor, competition from IT giants could limit growth in cybersecurity and cloud-computing subsegments.
• A difficult macroeconomy could weaken SME cloud and cybersecurity demand and reduce Sangfor’s revenue growth.
• Chip shortages could disrupt Sangfor’s storage and server supply or raise product costs.

What to watch

• Hyperscaler capital expenditure and resulting data-centre infrastructure demand for Range.
• GPU leasing-rate movements and their effect on Range’s GPUaaS revenue and profitability.
• Range’s utilization, operational efficiency, contract terms and EBITDA-margin progression.
• Sangfor’s cloud-computing and cybersecurity demand recovery, including HCI, storage and AI-security products.
• Policy support for cybersecurity, data security, IT innovation and digitisation.
• Sangfor’s overseas revenue growth and expense-control-driven earnings improvement.