Goldman Sachs Bets on China's Domestic AI Chip Rise

Goldman Sachs released two research reports on the same day that paint a stark picture of winners and losers in China's rapidly shifting AI chip landscape. The investment bank upgraded domestic chipmaker Cambricon Technology while downgrading server assembler Inspur Information, both moves driven by a single thesis: homegrown Chinese AI chips are rapidly displacing Nvidia and other Western suppliers.

The divergent ratings — issued on May 4 — reveal how the same macro trend can create vastly different outcomes for companies occupying different positions in the AI supply chain. For chip designers, localization is a windfall. For hardware assemblers, it is a margin squeeze.

Key Takeaways

• Goldman Sachs raised Cambricon's target price to 2,406 yuan (~$330), maintaining a 'buy' rating
• Inspur Information was downgraded from 'buy' to 'neutral,' with target price cut from 86.5 yuan to 76.5 yuan (~$10.50)
• Cambricon's Q1 2026 revenue hit 2.9 billion yuan (~$400 million), beating Goldman's forecast by 61%
• EBITDA margins surged from 26% to 42% quarter-over-quarter
• Contract liabilities — a proxy for order backlog — skyrocketed to 396 million yuan from near zero
• The core thesis: domestic AI chips are gaining share fast, compressing server prices but boosting chipmaker revenues

Cambricon's Breakout Quarter Stuns Analysts

Cambricon Technology's Q1 2026 performance was, in Goldman's own words, 'significantly above expectations.' Revenue reached 2.9 billion yuan (approximately $400 million), representing a 53% increase from the previous quarter and overshooting Goldman's prior estimate by a remarkable 61%.

Profitability metrics were equally impressive. The company's EBITDA margin jumped from 26% in Q4 2025 to 42% in Q1 2026, indicating that Cambricon is not just growing revenue but doing so with rapidly improving unit economics. This kind of margin expansion typically signals that a chipmaker has crossed a critical scale threshold where fixed R&D costs are being amortized over a much larger revenue base.

Perhaps the most telling indicator, however, lies in the order pipeline. Cambricon's contract liabilities — money received from customers for chips not yet delivered — surged to 396 million yuan by the end of Q1, up from a negligible amount in the prior quarter. This metric functions as a real-time demand signal, and the spike suggests that Chinese cloud providers and AI infrastructure operators are placing large advance orders for Cambricon's accelerators.

Why Goldman Turned Bearish on Inspur

Inspur Information occupies a fundamentally different position in the AI value chain. As China's leading AI server assembler, Inspur integrates chips from various suppliers — historically including Nvidia's high-end GPUs — into rack-scale server systems sold to data centers and cloud operators.

The localization trend that benefits Cambricon creates a structural headwind for Inspur. Here is the core problem:

• Cheaper chips mean cheaper servers. As customers swap expensive Nvidia GPUs (often priced at $20,000-$40,000 per unit) for domestic alternatives costing a fraction of that, the total bill of materials for each server drops significantly.
• Revenue per unit declines. Since server assemblers typically price their products as a markup on component costs, lower input costs translate directly into lower selling prices.
• Margin compression accelerates. Inspur's value-add — system integration, thermal management, firmware optimization — remains roughly constant in absolute terms, but represents a shrinking percentage of a declining total price.
• Competitive moats narrow. When the chip inside the box is a commodity rather than a scarce Nvidia allocation, the assembler's bargaining power diminishes.

Goldman cut Inspur's target price by roughly 12%, from 86.5 yuan to 76.5 yuan, and moved its rating to neutral. The message is clear: in a market where domestic chips are 'good enough,' the value migrates upstream to chip designers.

The Bigger Picture: US Export Controls Reshape the Market

Goldman's dual report arrives against a backdrop of escalating US semiconductor export restrictions targeting China. Since October 2022, Washington has progressively tightened controls on advanced AI chip sales to Chinese entities, most recently expanding restrictions in late 2025 to cover a broader range of Nvidia and AMD products.

These controls have inadvertently turbocharged China's domestic chip ecosystem:

• Cambricon, once considered a niche player, has emerged as a primary beneficiary of forced localization
• Huawei's Ascend series processors have gained significant traction in cloud data centers
• Biren Technology and other startups have secured substantial government-backed funding
• Chinese hyperscalers like Alibaba Cloud, Baidu, and ByteDance have accelerated qualification of domestic alternatives
• State procurement policies increasingly mandate domestic chip sourcing

The irony is significant. US export controls, designed to slow China's AI progress, have instead created a protected market for domestic chipmakers — one where companies like Cambricon can achieve the volume necessary to drive down costs and improve performance iteratively.

Compared to Nvidia's H100 and B200 accelerators, Chinese domestic chips still lag in raw performance on standard benchmarks. However, Goldman's analysis suggests the gap is narrowing faster than previously anticipated, particularly for inference workloads where absolute compute power matters less than efficiency and cost per token.

What This Means for Global AI Markets

Goldman's reports carry implications that extend well beyond two Chinese stocks. The broader signal is that the global AI chip market is bifurcating into two distinct ecosystems — one centered on Nvidia and Western suppliers, the other on a rapidly maturing Chinese domestic stack.

For Nvidia, this represents a slow-motion loss of what was once its fastest-growing market. China accounted for roughly 20-25% of Nvidia's data center revenue before export controls took effect. While Nvidia has developed export-compliant chips like the H20, Goldman's analysis implies that even these downgraded products face growing competition from improving domestic alternatives.

For Western investors, the divergence between Cambricon and Inspur offers a framework for evaluating other Chinese AI plays. Companies that own proprietary chip IP stand to benefit from localization. Companies that primarily assemble or integrate third-party components face structural margin pressure as the component ecosystem shifts.

For AI developers operating in or selling to the Chinese market, the practical implication is clear: software and model optimization for domestic chip architectures is becoming a critical capability. Frameworks like Cambricon's MLU software stack and Huawei's CANN are gaining importance relative to Nvidia's CUDA ecosystem within China.

Risks and Uncertainties Remain

Despite the bullish case for Cambricon, Goldman's report is not without caveats. Several risk factors deserve attention:

• Yield and manufacturing challenges. Cambricon relies on TSMC and other foundries for advanced node production, and geopolitical tensions could disrupt supply
• Software ecosystem maturity. Nvidia's CUDA dominance means developers face real switching costs when moving to domestic chips
• Valuation concerns. At Goldman's target price of 2,406 yuan, Cambricon trades at an extremely high multiple relative to current earnings
• Customer concentration risk. A significant portion of Cambricon's revenue likely comes from a small number of large state-backed buyers
• Technology gap persistence. For training large frontier models, Nvidia's chips remain substantially superior in both performance and energy efficiency

The Inspur downgrade also comes with nuance. Server assembly remains a critical function, and Inspur's deep relationships with Chinese enterprise and government customers provide some defensive moat. The company may also benefit if total server volumes increase enough to offset per-unit margin decline.

Looking Ahead: A Two-Track AI World

Goldman's simultaneous upgrade and downgrade represents more than a pair of stock calls. It is a structural thesis about the decoupling of global AI infrastructure into parallel ecosystems.

In the near term — the next 12 to 18 months — several developments will test this thesis. Cambricon's ability to sustain its Q1 growth trajectory through 2026 will be the most direct barometer. If contract liabilities continue to climb and EBITDA margins hold above 35%, the domestic chip thesis will be validated with hard numbers.

Inspur's response will also be instructive. The company could pivot toward higher-value system integration services, develop proprietary accelerator designs, or seek margin recovery through scale. Its strategic choices in the coming quarters will determine whether Goldman's downgrade proves prescient or premature.

For the global AI industry, the key question is whether China's domestic chip ecosystem can achieve self-sustaining momentum — reaching the performance and cost levels needed to support frontier AI development without relying on Western technology. Goldman's latest research suggests that moment may be arriving faster than most Western observers expected.

The age of a single, Nvidia-dominated AI chip market appears to be ending. What replaces it — a bifurcated world of competing standards and ecosystems — will reshape AI development, investment flows, and geopolitical dynamics for years to come.