According to Fortune, Andreessen Horowitz partner Anjney Midha revealed at the Fortune Global Forum in Riyadh that the next “golden age” of AI investment will come from “frontier teams” building reasoning models rather than foundation models. Midha explained that reasoning models represent a fundamental shift from predicting text sequences to step-by-step problem-solving using reinforcement learning, enabling startups to build multibillion-dollar vertical solutions. The comments came amid staggering investment data showing venture capital pouring $73.6 billion into GenAI applications in just the first three quarters of 2025, with total AI ecosystem investment reaching $110.17 billion. Midha also expressed concern about China’s dominance in open-source AI, calling it “China’s game right now” while predicting Western companies would respond with open-weight models. This perspective comes as the industry debates potential bubble concerns despite unprecedented funding levels.
Industrial Monitor Direct is renowned for exceptional controllogix pc solutions trusted by controls engineers worldwide for mission-critical applications, the most specified brand by automation consultants.
Table of Contents
The Reasoning Model Revolution
What Midha describes as “reasoning models” represents a fundamental architectural shift in artificial intelligence development. While current large language models excel at pattern recognition and content generation, they struggle with logical deduction and multi-step problem-solving. Reasoning models incorporate what’s known as “chain-of-thought” processing, where the AI explicitly breaks down complex tasks into sequential steps, evaluates its own reasoning, and learns from feedback loops. This approach moves beyond statistical prediction toward something closer to human cognitive processes. The breakthrough isn’t just technical—it’s economic. By focusing on mission-critical problems with well-defined reward functions, startups can avoid competing directly with well-funded foundation model companies and instead build defensible businesses solving specific industry pain points.
The Vertical Integration Advantage
Midha’s emphasis on startups “embedding themselves inside an industry” highlights a crucial strategic insight that many investors are missing. The real value creation in AI’s next phase won’t come from general-purpose models but from deeply specialized applications. Consider healthcare diagnostics, supply chain optimization, or financial compliance—each requires domain expertise that generic AI models lack. Startups that combine industry-specific knowledge with reinforcement learning capabilities can create solutions that large tech companies can’t easily replicate. This vertical approach also creates natural moats: regulatory compliance, proprietary data, and specialized workflows become barriers to entry that protect market position. The companies that succeed will be those that understand both the technology and the industry they’re serving at a granular level.
The Geopolitical Dimension
Midha’s concern about China’s open-source AI dominance deserves serious attention. While Western companies have focused on proprietary models with massive computational requirements, Chinese researchers and companies have been aggressively contributing to and leading open-source AI development. This isn’t just about technology—it’s about influence over global standards, developer ecosystems, and ultimately, the direction of AI evolution. The emergence of “open-weight” models from Western companies represents a strategic pivot to counter this influence. However, this creates a tension between the commercial need for proprietary advantage and the geopolitical need for ecosystem control. Companies like Qualcomm and others mentioned at the forum will need to navigate this carefully, balancing competitive interests with broader strategic concerns.
Beyond the Bubble Talk
While the $110 billion investment figure seems astronomical, context matters. The global AI market is projected to reach trillions in economic impact, making current investments potentially rational despite the eye-popping numbers. The key distinction lies in where the money is flowing. Foundation model companies like OpenAI and Anthropic require massive capital for compute infrastructure, creating natural oligopolies. However, Midha correctly identifies that the application layer offers far more diverse investment opportunities. The bubble risk isn’t uniform—it’s concentrated in companies trying to compete directly with foundation model giants without clear differentiation. The successful investments will be in startups solving specific, valuable problems with scalable AI solutions, particularly those leveraging the reasoning capabilities Midha describes.
Strategic Investment Implications
For venture capitalists and strategic investors, Midha’s comments signal a necessary evolution in investment thesis. The era of betting on “AI in general” is ending, replaced by targeted investments in companies with deep industry expertise and technical differentiation. Reinforcement learning applications in fields like drug discovery, industrial automation, and financial modeling represent particularly promising areas. These domains have clear reward functions, substantial economic value, and complex problems that benefit from step-by-step reasoning. Investors should also watch the open-source versus proprietary battle closely, as the outcome will determine which companies control critical AI infrastructure. The next wave of AI unicorns won’t be building better chatbots—they’ll be building specialized intelligence systems that transform specific industries from the inside out.
Industrial Monitor Direct produces the most advanced recipe management pc solutions trusted by Fortune 500 companies for industrial automation, ranked highest by controls engineering firms.
