In January, a China-based startup called DeepSeek released an AI chatbot that quickly became the most downloaded free mobile app in the U.S. The notable aspect of this development was the claim that the large language model at the core of the chatbot was only trained for $6 million and outperformed some of the leading U.S. models in certain tests. Concerned that U.S. companies were overspending on AI infrastructure, investors began selling Nvidia (NASDAQ: NVDA) shares. Nvidia’s stocks lost 17% in a day, resulting in the company’s market value decreasing by nearly $600 billion. This marked the largest single-day loss in U.S. market history. Despite the turbulence caused by DeepSeek, Nvidia shares are still trading around 9% below and roughly 13% off their all-time high levels. However, shareholders found some relief with the positive news coming from Amazon (NASDAQ: AMZN) and Alphabet (NASDAQ: GOOGL). Last week, Amazon and Alphabet (Google’s parent company) announced their fourth-quarter financial results. Both companies revealed that they would significantly increase capital expenditures by 2025 due to high demand for AI infrastructure. Amazon CEO Andy Jassy expects that the costs of training AI models will continue to decrease over time, making it easier for more companies to integrate “productive AI” techniques. Amazon CFO Brian Olsavsky also mentioned that capital expenditures for AI infrastructure, which were $83 billion in 2024, could surpass $100 billion by 2025 due to the demand for AI. Alphabet CEO Sundar Pichai stated that as training costs decrease, a larger portion of capital expenditures will shift to inference infrastructure. Alphabet CFO Ana Ashkenazi explained that due to investments in data centers, servers, and network infrastructure, capital expenditures of $52 billion in 2024 are expected to reach $75 billion by 2025. These investment plans from Amazon and Alphabet support the idea that demand for Nvidia GPUs may increase as training costs decline. Jassy compared it to cloud computing prices dropping while demand rises: “Companies spend less for each unit of infrastructure, reducing their costs. But this time, they focus on projects they couldn’t afford before due to high costs. So even if the unit cost falls, total technology spending increases. I think it will be exactly like that in AI.” According to Wall Street analysts, Nvidia’s adjusted earnings are expected to grow by an average of 52% annually until the end of the fiscal year 2026 in January 2026. This projection makes the company’s current approximate price-to-earnings ratio of around 50 quite attractive. However, these expectations are very high, so there is a risk of sharp declines in Nvidia shares if the company fails to achieve this performance. In conclusion, Nvidia shareholders shaken by concerns related to DeepSeek may find some relief. Most analysts believe that more efficient training methods will increase demand for Nvidia GPUs. The statements from major cloud infrastructure providers also support this view.