The enterprise adoption of artificial intelligence is expanding swiftly, yet newly released empirical research highlights structural bottlenecks and strategic dependencies. According to a comprehensive survey conducted by the National Bureau of Economic Research (NBER Working Paper 34836), covering nearly 9,000 C-level executives across seven countries, 78 percent of companies now utilize AI. Financial commitments have risen correspondingly: US firms plan to spend an average of 2,000 dollars per employee on AI in 2026, totaling roughly 300 billion dollars. However, capital expenditure remains heavily concentrated, with just two percent of companies accounting for more than 50 percent of all AI investments.
Despite this capital influx, immediate productivity returns remain modest. Executives surveyed in the NBER study reported that AI has contributed an efficiency gain of merely 0.2 percentage points per year so far, with expectations rising to 0.6 percentage points annually over the next three years. Overall headcounts show no aggregate net job destruction, but substantial reallocation is underway internally. More than 40 percent of firms reported significant restructuring and job churn across individual departments, offsetting hiring and cuts to leave net employment figures neutral.
Meanwhile, the technical execution layer is experiencing a rapid geographic shift. Research from the Friedrich Naumann Foundation, analyzing OpenRouter transaction data, reveals that European software engineers are pivoting rapidly toward Chinese model infrastructure. By late summer 2026, 61 percent of billed API tokens in Germany routed through Chinese models such as DeepSeek and Qwen, while US models dropped to 37 percent. Across the European Union, the share of Chinese models climbed from 15 percent to 58 percent within twelve months. Germany represents the second-largest market for Chinese AI inference outside China, with 60.5 trillion tokens processed, driven primarily by open-source licensing and API costs up to 33 times lower than proprietary US equivalents.
This technical shift underlines growing concerns regarding operational dependency and digital sovereignty. A survey published by the German digital association Bitkom indicates that 96 percent of domestic enterprises rely fundamentally on imported digital core technologies. In the hypothetical scenario of an immediate embargo on foreign cloud services, software and AI models, companies estimated that German business operations could sustain themselves for an average of just 16 months. Furthermore, 56 percent of firms explicitly designated AI applications as a critical factor of foreign dependency.
These systemic vulnerabilities coincide with widening governance deficits inside corporate IT structures. In a global study of chief information officers released by Thoughtworks, 88 percent of IT leaders admitted that AI adoption across business departments is outrunning formal compliance and governance rules. In addition, 89 percent of CIOs noted that they must dedicate more time and responsibility to reorganizing work processes and workforce structures than to managing conventional IT infrastructure. The economic integration of AI is increasingly turning into an organizational challenge, coupled with acute external technology dependencies.