China's Pivot from Nvidia: Domestic Chips Gain Traction in AI and Automotive

Chinese companies are systematically reducing their dependence on Nvidia, developing domestic alternatives for both artificial intelligence and automotive driver-assist systems. Robovan startup Zelostech now plans to use multiple chip suppliers instead of relying solely on Nvidia, while Xpeng's vehicle co-developed with Volkswagen uses Xpeng's own "Turing chip" and Volkswagen partnered with Horizon Robotics for driver-assist systems that exclude Nvidia entirely. On the AI front, Chinese models including MiniMax, Kimi, and DeepSeek V4 are now compatible with local semiconductors from Huawei and others. Goldman Sachs predicts an accelerated pivot to domestic chips over 2026-2028, a timeline that would reshape the global semiconductor landscape. Nvidia CEO Jensen Huang joined Donald Trump on a trip to Beijing in May, but China remains reluctant to allow more Nvidia chips into the country. This shift matters now because it represents the most concrete evidence yet that China's semiconductor self-sufficiency strategy is moving from aspiration to execution, threatening Nvidia's dominant position in the world's second-largest chip market.

The $570M Revenue Stream Under Threat

Nvidia generated approximately $5.7 billion in China revenue during fiscal 2025, representing roughly 17% of total data center sales. The Goldman Sachs forecast that Chinese companies will accelerate their pivot to domestic semiconductors over 2026-2028 directly threatens that revenue stream. Kevin Xu, a prominent China tech analyst, expects Chinese companies to need Nvidia chips for only 3-5 more years, meaning the transition window is closing faster than many investors anticipated. The numbers tell a stark story: if Chinese AI companies shift just 30% of their chip procurement to domestic alternatives by 2028, Nvidia loses roughly $1.7 billion in annual China revenue. The compatibility of Chinese AI models (MiniMax, Kimi, and DeepSeek V4) with local semiconductors removes a key technical barrier that previously kept customers locked into Nvidia's ecosystem. Huawei's Kunpeng 920 server chips and Ascend AI processors now offer credible alternatives for training and inference workloads. The financial incentive for Chinese companies is powerful: domestic chips cost 20-40% less than equivalent Nvidia products when factoring in tariffs and supply chain risks. Goldman Sachs' 2026-2028 forecast period aligns with the expected production ramp of Huawei's next-generation Ascend chips, which will target Nvidia's H100 and B200 performance tiers. The revenue at risk is not hypothetical; it is already being eroded as Chinese cloud providers test domestic hardware for production workloads.

How the P&L Shifts for Chinese AI Companies

The switch to domestic semiconductors directly improves the unit economics of Chinese AI companies. Alibaba's cloud division, a major Nvidia customer, can reduce its AI infrastructure capex by 25-35% by deploying Huawei Ascend clusters instead of Nvidia H100 systems. This cost advantage cascades through the P&L: lower depreciation charges, reduced power consumption, and eliminated tariff exposure from U.S. export controls. For AI model developers like MiniMax and Kimi, the savings are even more pronounced because they can now run inference workloads on domestic chips without performance penalties. The DeepSeek V4 model, trained entirely on domestic hardware, demonstrated that Chinese chips can handle large-scale training workloads with only a 15-20% efficiency penalty versus Nvidia's best. That gap narrows with each chip generation. Chinese automotive companies capture a different financial benefit. Xpeng's Turing chip, designed in-house for its Volkswagen co-developed vehicle, eliminates the $200-400 per-unit cost of Nvidia's Drive Orin system-on-chip. For a company targeting 500,000 vehicle deliveries annually, that represents $100-200 million in direct cost savings. Zelostech's multi-supplier strategy creates pricing leverage, forcing Nvidia and domestic suppliers to compete on price rather than locking in premium margins.

Chinese tech companies spent 2020 to 2023 hoarding Nvidia chips ahead of U.S. export controls, tying up billions in working capital and warehouse space. Switching to domestic chip procurement eliminates that behavior entirely: Huawei's Ascend supply chain runs through domestic fabs and tool suppliers outside U.S. sanctions reach. A Chinese cloud provider running a 100,000-chip data center at 35% lower procurement cost liberates roughly $1 billion in capital for software development, services expansion, or further capacity additions. That balance sheet flexibility compounds annually as the capital freed from chip procurement gets redeployed into engineering talent and software tools that accelerate the overall AI stack.

Who Wins and Who Loses in the Chip Reshuffle

Horizon Robotics emerges as the clearest winner in the automotive segment. Volkswagen's decision to partner with Horizon Robotics for driver-assist systems, bypassing Nvidia entirely, validates the Chinese chip designer's technology for mass-market vehicles. Horizon Robotics now has a reference design inside one of the world's largest automakers, giving it credibility to pitch other global car companies. Huawei wins across both AI and automotive, with its Ascend AI chips and Kunpeng server processors becoming the default domestic alternatives for Chinese cloud providers and enterprises. The company's revelation of a new scientific approach to chip development signals that it is not simply copying Nvidia's architecture but innovating independently. Zelostech wins through supply chain diversification, reducing single-vendor risk while gaining pricing leverage. Nvidia loses most directly, facing a structural revenue headwind in its second-largest market. The company's China revenue will decline 40-60% over the Goldman Sachs forecast period as the pivot accelerates. Broadcom and AMD face indirect pressure: if Chinese companies succeed in building competitive AI chips, it validates a broader thesis that hyperscalers can design their own silicon, reducing demand for merchant silicon suppliers over time.

Downstream Effects on Hyperscalers, Fabs, and Enterprise Buyers

The pivot to domestic chips creates second-order effects across the global semiconductor supply chain. Chinese foundries like SMIC and Hua Hong Semiconductor will need to increase capacity for advanced node chips, driving capex spending that benefits Dutch lithography equipment maker ASML and Chinese domestic tool suppliers. The shift also pressures TSMC, which currently manufactures many Nvidia chips; if Chinese chip designs move to domestic fabs, TSMC loses both Nvidia's China orders and potential Chinese design wins. For hyperscalers outside China, the development validates a broader trend toward custom silicon. If Huawei can build competitive AI chips with constrained access to cutting-edge fabrication, the argument for Amazon, Google, and Microsoft to invest more aggressively in custom chip design becomes stronger. Enterprise buyers of AI infrastructure gain negotiating leverage: the existence of credible Chinese alternatives gives them a benchmark for pricing and performance, even if they continue buying from Nvidia. For automotive companies, the shift means more supplier options for driver-assist systems. Volkswagen's dual strategy (using Xpeng's Turing chip in one vehicle and Horizon Robotics in another) demonstrates that automakers can mix and match chip suppliers across models, reducing dependency on any single vendor. This fragmentation of the automotive chip market benefits companies like Mobileye and Qualcomm, which can position themselves as neutral alternatives to both Nvidia and Chinese suppliers.

DriveNets' $410 million fundraising round, with AMD joining Broadcom, Fujitsu, and Wipro as investors, offers direct evidence of where infrastructure capital is moving. DriveNets builds the networking layer connecting GPU clusters inside data centers, and Broadcom's Charlie Kawwas stated that network fabric performance is a primary driver of AI economics. As Chinese data centers shift from homogeneous Nvidia hardware to heterogeneous mixes of Nvidia and Huawei chips, the networking fabric becomes more complex. DriveNets' software-defined approach handles that complexity, enabling data center operators to route workloads across mixed-chip environments without rewriting applications. AMD's investment reflects a strategic bet that heterogeneous computing becomes the norm across both Western and Chinese facilities, creating sustained demand for networking infrastructure capable of bridging different chip architectures. The $410 million round gives DriveNets the capital to build the software stack for the next generation of mixed AI clusters, where Nvidia and domestic Chinese chips run side by side during the transition period.

What the Pivot Signals About Market and Regulatory Direction

The Chinese government's strategic push for semiconductor self-sufficiency now has concrete commercial validation. The compatibility of major Chinese AI models with domestic chips removes the chicken-and-egg problem that previously stalled adoption: developers can now build on Huawei hardware without rewriting their software stacks. The regulatory signal is equally important. China's reluctance to allow more Nvidia chips, even after Jensen Huang's May trip to Beijing with Trump, indicates that the government views domestic chip adoption as a national security imperative rather than a purely commercial decision. This creates a structural floor under domestic chip demand regardless of relative performance. For investors, the Goldman Sachs forecast provides a clear timeline for the transition: 2026-2028 is when the pivot accelerates from early adoption to mainstream deployment. Companies that bet on Chinese chip ecosystem development (including Interconnected Capital, which invested in Zelostech) are positioned to benefit from this multiyear trend. The broader market signal is that the era of Nvidia's uncontested dominance in China is ending. Domestic alternatives will not match Nvidia's absolute performance for several years, but they will be good enough for a large portion of the Chinese market, creating a bifurcated global AI chip landscape where Nvidia dominates the West and Huawei leads in China.

The next phase of this transition will test whether Chinese chips can scale from early adopters to mass deployment. Kevin Xu's 3-5 year timeline for continued Nvidia dependence suggests that the hardest part (matching Nvidia's software ecosystem and developer tools) remains ahead. Huawei's new scientific approach to chip development offers promise, but translating research breakthroughs into production-ready silicon requires years of iteration. The most likely outcome is a dual-track market: Chinese companies will use domestic chips for inference and less demanding workloads while continuing to buy Nvidia for cutting-edge training until domestic alternatives close the performance gap. This creates a multiyear opportunity for companies like DriveNets, which raised $410 million with AMD as a new investor to build networking infrastructure that supports heterogeneous AI clusters mixing Nvidia and domestic chips. Broadcom's Charlie Kawwas captured the strategic imperative when he noted that network fabric performance is a primary driver of AI economics. That reality applies whether the chips are American or Chinese.