Sovereign AI is a critical geopolitical imperative for nations to avoid foreign dependency, a trend that NVIDIA is successfully commercializing to diversify its customer base.
China, particularly through Alibaba's Qwen and DeepSeek, has surpassed Meta's Llama to become the undisputed leader in the open-source AI movement, excelling in vision and world modeling.
The overall AI market is not in a bubble due to the strong financial fundamentals of major tech companies, but specific areas like AI for materials science are experiencing "irrational exuberance" from VCs.
Despite the rise of powerful open-source alternatives, the market will ultimately favor high-quality, convenient closed-source models, mirroring the iOS vs. Android dynamic.
Europe is undergoing a necessary and significant re-militarization, driven by waning U.S. security guarantees and Russian aggression, which is reversing decades of underinvestment in its defense industrial base.
▶The Geopolitics of AI and Defense
Benaich analyzes the intersection of technology and national security, focusing on the "sovereign AI" trend where nations build domestic AI infrastructure to avoid foreign control, as exemplified by US oversight of exported F-35 jets. He links this to a broader European re-militarization in response to waning U.S. support and Russian aggression.
Investors should recognize that national security interests, not just commercial applications, are becoming a primary driver of multi-billion dollar investments in AI hardware and defense technology, creating new, state-sponsored markets.
▶AI Market Structure and Value Accrual
He details a market shift where foundational model providers (like OpenAI) are now capturing more value than the application-layer companies built on top of them, a reversal of the earlier dynamic. He also observes that while the overall AI sector is not in a bubble, specific sub-sectors like materials science are overfunded relative to their commercial infrastructure.
The power dynamic in the AI stack is consolidating towards model providers, suggesting that defensibility for application companies must come from proprietary data and superior user experience, not just access to a third-party model.
▶The Open vs. Closed Source AI RaceApr 2026
Benaich tracks the leadership in open-source AI, noting that Chinese firms like Alibaba (Qwen) and DeepSeek have taken the mantle from Meta (Llama). However, he maintains that the highest-performing models remain closed-source and predicts that, like iOS vs. Android, convenience and quality will lead most users to prefer closed systems long-term.
While open-source provides a competitive check and a platform for innovation, the ultimate market winners may be closed-source players who can offer a more polished, integrated, and performant product, creating a potential winner-take-most scenario.
▶The Economics of AI ModelsApr 2026
Benaich breaks down the costs and profitability of different AI modalities, noting that text and audio are relatively cheap while video and world models are extremely expensive. He highlights that the most efficient model providers are achieving high gross margins (60-80%), and the process for training large models has become a more efficient, repeatable "recipe."
The high cost of training and serving advanced models (video, world models) creates a significant barrier to entry, concentrating power among a few well-capitalized labs and cloud providers, while the commoditization of cheaper modalities opens opportunities for niche application developers.