Where risks are underestimated and opportunities overlooked
By Mike Glöckner, analyst in the research team at DJE Kapital AG
The AI market (AI stands for artificial intelligence) is in a pronounced correction phase, which has led to oversold situations in places. There are new relevant developments almost daily, but the weakness in AI stocks - especially in the infrastructure segment - continues. DeepSeek's assessment that more efficient models could reduce the need for AI hardware in the future remains a key negative factor. Furthermore, although NVIDIA's latest quarterly figures were solid, they did not reach the level of overachievement seen in previous quarters. Even NVIDIA was unable to stabilize the ailing AI market. Another important negative factor is the expected tightening of US restrictions under the Trump administration. The latter is pursuing a tough line in geopolitical competition with China, with AI playing a key role in the struggle for economic, military and political supremacy.
Despite US restrictions, China appears to still have access to NVIDIA chips - both the Hopper architecture (H100, H200) and the latest Blackwell systems, which have been shipping since the last quarter. These are no longer just individual chips, but complete server systems in which NVIDIA achieves a larger share of sales. A conspicuously high proportion of these systems were invoiced to companies in Singapore before being shipped on to other countries. It is possible that China is also being supplied in this way. This poses a major challenge for the US government, as it is becoming increasingly difficult to control the actual end users of the AI systems.
The Trump administration is likely to further tighten export restrictions on AI technologies. At the same time, these measures are strengthening China's efforts to develop its own chips. In the area of older semiconductor technologies (14-28 nm), China is already largely independent of imports and produces chips for the mass market, not least for the entertainment and automotive industries. It is estimated that China is about 35% independent of imports in chip production. However, in the top class of modern semiconductors (2-5 nm), which are essential for AI and high-performance computing, China is still years behind the USA. However, the example of DeepSeek shows that technological backlogs can be made up faster than expected.
Technological progress and strategic importance for China
China is pressing ahead with its technological development. A 2023 analysis by the Australian Strategic Policy Institute shows that the country now leads in patent registration in 57 out of 64 technology areas - a significant increase from just three areas in 2007. President Xi Jinping is also focusing specifically on strengthening the technology sector: at a recent meeting with executives from companies such as Alibaba, Xiaomi, Huawei, Tencent, DeepSeek and BYD, the central role of the industry in China's economic future was emphasized.
According to Bloomberg Economics, China's high-tech sector reached a volume of 20.7 trillion yuan in 2024, which corresponds to around 15.4% of GDP. This share is expected to rise to 18.3% by 2026 - meaning that the high-tech industry could replace the real estate industry as the most important economic sector. If segments such as electric vehicles, batteries and solar energy are included, the share could even rise to 23%.
Xiaomi is attracting attention in the electric vehicle market with attractively priced models based on luxury brands. The company also ranks third in the global smartphone market after Apple and Samsung and operates the world's largest AIoT network with over 861 million connected devices - a system that combines artificial intelligence (AI) with the Internet of Things (IoT) to enable intelligent and automated device interactions. Xiaomi's CFO explained that the company is currently still working with NVIDIA chips, but is already looking for alternatives - a strategic necessity given the geopolitical tensions.
Massive capital flows into the AI sector
Investment in artificial intelligence continues to rise rapidly. Hyperscalers such as Amazon, Google and Microsoft in particular are driving the expansion of AI infrastructure with high capital expenditure. Their CapEx expenditure is estimated to be around USD 320 billion in 2025 - an increase of over 40% compared to the previous year. A significant proportion of these funds will flow into the expansion of the AI infrastructure and the further development of more powerful AI systems.
AI is also increasingly seen as a key strategic technology at a political level. Together with leading software companies, the Trump administration has announced the 500 billion dollar "Stargate" project, while the European Union is providing its own impetus with the 200 billion dollar "AI Act". In France, around 100 billion dollars will flow into AI projects via private investors and Meta is planning to build a 200 billion dollar AI campus. Despite all the concerns about a possible slowdown, the investment momentum in the AI sector remains undiminished.
Increasing efficiency as a growth driver: DeepSeek and the Jevons Paradox
The Jevons paradox, named after the British economist William Stanley Jevons, describes the apparent paradox that technological advances in resource efficiency do not necessarily lead to lower consumption, but on the contrary can increase demand.
A classic example of this is server virtualization, which enabled more efficient use of computing capacity in the early 2000s - and paradoxically led to a further increase in demand for more powerful servers, including chips. The emergence of cloud computing had a similar impact: The ability to switch server instances on and off as required optimized the use of resources, which further boosted demand for hardware.
A similar effect could now occur in the field of artificial intelligence. The AI model developed by DeepSeek promises a more efficient use of computing power. Instead of reducing hardware requirements, this increase in efficiency could even increase the demand for specialized AI hardware in the long term - in line with the Jevons paradox.
Technological advances: AI hardware and software on the rise
The development towards general artificial intelligence (AGI) is based on both advances in hardware and innovative software approaches. On the hardware side, NVIDIA's Hopper architecture has increased AI training performance by a factor of 10. The newer Blackwell architecture goes one step further: it enables a 15 to 30-fold improvement in AI inference, i.e. the processing of queries, and at the same time reduces energy consumption by a factor of 25.
Significant progress has also been made in the area of software. Tests in disciplines such as natural sciences, mathematics, programming and languages have shown that AI models can already operate at doctoral level. In these clearly defined, rule-based task areas, artificial intelligence is thus approaching human performance.
Nevertheless, significant limitations remain: In areas such as situational adaptation, independent learning, transfer performance and empathy, AI still lags far behind human intelligence.
Agentic AI and physical AI: progress in automation and robotics
Agent-based artificial intelligence (agentic AI) is becoming increasingly important. AI agents are taking on ever more complex tasks - from automated responses to typical customer inquiries and order processing to the optimization of business processes. Leading AI developers such as OpenAI, Google Gemini, Perplexity and DeepSeek have now developed specialized deep search agents that can solve challenging problems with high precision. The quality of generative AI systems has improved significantly in recent years and is now classified as consistently powerful to excellent.
Another growing field of innovation is physical AI, also known as embodied AI. Here, artificial intelligence is integrated into physical objects so that they can interact with the laws of the real world - for example in self-driving cars or robotic systems. This technology, whose development began more than two decades ago, has recently made great progress. Robotaxis are now in use in several cities in the USA, China and the Gulf region, and will soon be in Singapore too. The technological maturity of these systems is increasing rapidly and the leading providers are showing continuous improvements in safety and functionality.
Technological progress in the field of robotics is also considerable. Leading manufacturers assume that humanoid household robots could be ready for the market in the next two years. Humanoid robots are already being used in automotive production and logistics centers, where they optimize efficiency and work processes. The combination of physical AI and advanced robotics could play a central role in automation in the coming years.
Corrective phase with long-term potential
The AI market is currently in a correction phase triggered by uncertainties about the profitability of high infrastructure investments, geopolitical tensions and tightened US export restrictions. At the same time, technological development continues unabated. Advances in AI hardware and software, particularly in the areas of agentic AI and physical AI, are driving innovation and opening up new market potential.
Despite temporary setbacks, the long-term outlook for the sector remains positive. Considerable investments continue to be made in the development of more efficient AI models and powerful hardware. In addition, AI-supported applications have already established themselves in many industries and are continuing to grow.
The current correction has also led to some market leaders in AI chips, chip manufacturing, memory technologies, network equipment, data center construction, AI platforms and software solutions appearing attractively valued.
Legal notice
Marketing advertisement: All information published herein is for your information only and does not constitute investment advice or any other recommendation. The statements contained in this document reflect the current assessment of DJE Kapital AG. These may change at any time without prior notice. All statements made have been made with care in accordance with the state of knowledge at the time of preparation. However, no guarantee and no liability can be assumed for the correctness and completeness.