As companies accelerate their adoption of generative AI, Chinese models such as DeepSeek, Alibaba’s Qwen, and Moonshot AI’s Kimi are gaining traction as potential alternatives to leading US providers. While widespread enterprise adoption in Europe is still developing, many organizations are increasingly exploring multi-model strategies involving Chinese vendors, rather than relying on a single AI ecosystem.
Rising Adoption, but in Addition to US Providers
There are currently no representative figures on the use of Chinese AI models by European companies. However, public examples show that they are being used, though often still in the testing and selection phase. Siemens, for example, has said that it has experimented with Chinese AI such as DeepSeek alongside US and European models.
Actual adoption is likely to be higher than public references suggest. Open-weight allows companies to test Chinese models on their own or European infrastructure without a direct contract with the original provider. Many projects therefore remain below the radar. Current use cases focus mainly on cost comparisons, coding, research, document processing, and internal knowledge systems.
The important point to stress is that companies are not simply replacing US providers. Instead, larger companies are increasingly adopting multi‑model strategies. They are making pragmatic choices between individual models when it comes to specific tasks and needs.
Why Choose Chinese AI Models?
China is repeating its proven playbook from other industries in the AI‑token space. While US rivals focus on technological dominance and push for “higher, faster, further” in models and hardware, China’s approach emphasizes scale and price leadership. As seen in the solar or automotive sectors, aggressive pricing and rapid adoption are intended to pave the way for global rollout. At the same time, Chinese providers remain largely on par technologically: the performance gap between leading US and Chinese models has narrowed dramatically.
On the other hand, the cost reality has flipped. Even as token prices fall, total inference costs can still rise substantially as enterprises move from pilots to high-volume usage. This means that at current rates, running frontier AI at scale can end up costing more than the human equivalent for certain tasks. That’s the opposite of what everyone assumed two years ago.
In contrast, Chinese models often cost substantially less than US counterparts. This gap is making the models themselves more attractive, particularly for organizations looking to reduce ballooning AI costs and deliver ROI. On the technical side, many Chinese providers follow an open‑weight approach: DeepSeek, Alibaba, or Moonshot AI allow companies to run models on their own or European infrastructure. This is extremely relevant for organizations that value data sovereignty, customizability, and independence from individual cloud providers.
The Impact on ERP – And ERP Vendors
ERP systems hold the most sensitive data a company has, such as financials, HR records, and supplier terms. Most CIOs we talk to were never comfortable piping that through an external API in the first place. Open-weight lets them host the model in their own environment instead, which quietly removes what’s been the biggest blocker to embedding AI directly into ERP workflows.
Combining open-weight with the cost advantages of Chinese models now makes it economical for companies to put AI inside every ERP transaction, rather than just providing a handful of flagship copilot features. Companies can look to apply AI on their real workloads, such as the thousands of routine transactions that ERP handles every day across departments such as finance, procurement, and supply chain.
What this means is that longer term, the chosen AI model stops being the interesting part. Once capable models are cheap and swappable, the differentiation moves elsewhere. Who actually owns the enterprise data, the workflow logic, and the business context? That’s been the ERP vendors’ moat for thirty years, so this cuts both ways for them. It is a threat to their AI premium, but also their best defense against complete replacement.
At the same time businesses increasingly expect ERP and enterprise-software platforms to remain model-agnostic, rather than being forced to take whatever model is bundled into their ERP suite. This will pile real pressure on SAP, Oracle, and Microsoft to keep their platforms open rather than model-exclusive.
It also means that ERP vendors face a broad strategic question. Can they monetize the model layer itself or will their value increasingly shift toward proprietary enterprise data, workflow integration, and business context?
Factoring in the Risks
Companies need to be mindful of two key risks with Chinese AI models. The first is geopolitics. We’ve all seen how quickly hardware and software exports and imports can be banned as restrictions tighten. In the event of further escalation between East and West, companies that rely heavily on Chinese solutions could quickly find themselves exposed.
Data protection and information security also remain central issues despite open model weights. When official Chinese apps or APIs are used, sensitive customer, employee, or company data may end up on servers in China. This poses significant governance, compliance, and reputational risks, especially for regulated industries running Chinese-hosted APIs directly against ERP data.
Self‑hosting in a European data center does mean that data is not automatically transferred to China, which significantly reduces the risk. However, questions remain about training data origins, biases, security vulnerabilities, documentation, and compliance with the EU AI Act.
Companies that choose Chinese models primarily because of price also risk creating a new dependency. European firms know this problem from other industries. However, relying exclusively on US providers is equally risky: recent access restrictions have shown how dependent companies can be on political decisions or the product strategies of individual vendors.
The Importance of a Multi-Model Approach
Overall, Chinese AI models offer an appealing package, based on their cost and performance. However, the decisive factor is not the model’s country of origin or ranking on a technology leaderboard but testing with the company’s own data and processes. Essentially, the most promising strategy shouldn’t be “Chinese instead of US,” but keeping multiple models available and reassessing performance, cost, and risk for each use case.
This removes the risk that European companies merely replace US dependencies with Chinese ones. The right answer is therefore neither a blanket exclusion of Chinese models nor their uncritical adoption, but a European multi‑model strategy with its own infrastructure, clear security standards, and investment in European alternatives.
