AGI Race: Big Tech's Hype vs. Scientific Reality
The pursuit of artificial general intelligence (AGI) and superintelligence has become the defining technological race of our era, with global tech giants pouring unprecedented resources into these ambitious goals. OpenAI, a frontrunner in the field, recently unveiled its GPT-5 model, describing it as a “significant step on the path to AGI,” the theoretical state where an AI system can perform any human job with high autonomy. Yet, even OpenAI CEO Sam Altman, while touting GPT-5’s advancements, acknowledged crucial limitations, noting the model is “missing something quite important,” particularly its inability to learn continuously post-launch. This caveat underscores a critical distinction: current systems, while impressive, have yet to achieve the self-sustaining autonomy required for full-time human-level work.
OpenAI is not alone in this high-stakes competition. Meta CEO Mark Zuckerberg recently declared that the development of superintelligence—a theoretical state where an AI system far exceeds human cognitive abilities—is “now in sight.” Google’s AI unit, for its part, outlined its own path to AGI by revealing an unreleased model designed to train AIs within convincing simulations of the real world. Meanwhile, Anthropic, another significant player, announced an upgrade to its Claude Opus 4 model, further intensifying the competitive landscape.
Despite the intellectual and financial fervor, the scientific underpinnings of this quest remain debated. Benedict Evans, a seasoned tech analyst, characterizes AGI as “a thought experiment as much as it is a technology.” He points to a fundamental lack of a theoretical model explaining why current generative AI systems work so effectively, let alone what it would take for them to reach AGI. Evans draws an analogy to the early space race, suggesting it’s akin to building an Apollo program without fully understanding gravity or rocket mechanics, hoping that simply making a rocket bigger will suffice. He describes much of the current discourse as “vibes-based,” with AI scientists often relying on personal intuition rather than concrete scientific understanding.
However, not all experts share this level of skepticism. Aaron Rosenberg, a partner at venture capital firm Radical Ventures and former head of strategy at Google’s AI unit DeepMind, offers a more optimistic, albeit narrower, definition of AGI. He believes that if AGI is defined as achieving at least 80th percentile human-level performance across 80% of economically relevant digital tasks, it could be within reach within the next five years. Conversely, Matt Murphy, a partner at VC firm Menlo Ventures, emphasizes that the definition of AGI itself is a “moving target,” implying the race will evolve and continue for years as the bar is consistently raised.
Even without achieving AGI, the financial success of current generative AI systems is undeniable. OpenAI’s annual recurring revenue has reportedly surged to $13 billion, with projections to exceed $20 billion by year-end. The company is also in talks for a share sale that could value it at an astonishing $500 billion, surpassing even SpaceX. Yet, this commercial success and the grand pronouncements about superintelligence raise concerns among some experts. David Bader, director of the Institute for Data Science at the New Jersey Institute of Technology, views such claims as more reflective of competitive positioning than genuine technical breakthroughs. He warns against confusing marketing narratives with actual advances, stressing the need to focus on immediate concerns like ensuring current systems are reliable, transparent, and free of bias, rather than chasing distant, ill-defined goals.
The race is also intensely global, with China emerging as a formidable contender. DeepSeek, a Chinese firm, recently unveiled its R1 model, which boasts reasoning capabilities comparable to OpenAI’s leading work. Major corporations, including Saudi Aramco, are already integrating DeepSeek’s AI technology, reporting significant improvements in efficiency. Artificial Analysis, a company that ranks AI models, notes that six of the top 20 models on its leaderboard are Chinese, developed by companies such as DeepSeek, Zhipu AI, Alibaba, and MiniMax. In the rapidly evolving field of video generation models, six of the top ten, including the current leader, ByteDance’s Seedance, are also Chinese. Microsoft President Brad Smith has underscored the strategic importance of global adoption, stating that the country whose AI technology is most broadly adopted worldwide will ultimately win the AI race, drawing parallels to the difficulty of supplanting Huawei’s leadership in 5G.
Regardless of the ongoing debates about the feasibility and timeline of superintelligent systems, immense sums of money and top talent are being poured into this global competition. The world’s two largest economies, the US and China, are locked in an intense technological contest that shows no signs of slowing. As Rosenberg observes, just five years ago, suggesting AGI was on the horizon would have been considered “blasphemous”; today, it is increasingly becoming a consensus view that the path is indeed clear.