Has GPT-5 Marked a Shift in the AI Race?

GPT-5 shift in the AI race: OpenAI’s much-anticipated GPT-5 failed to bring about a revolutionary breakthrough and seems to have hit a bottleneck. The focus in the market is now shifting to how existing technologies can be used to create broader business value at both product and service levels.

Last week, when OpenAI released its new GPT-5 model, it was expected to be another highlight for the company. Sam Altman had described GPT-5 as “a significant step on the path to AGI.” However, the release quickly sparked disappointment. Users took to social media to share examples of the model’s basic mistakes, such as mislabeling a map of the United States. Experienced users also expressed frustration with its performance and changes in “personality,” pointing to mediocre results on benchmark tests.

This may not have been OpenAI’s original intention, but the launch of GPT-5 clearly demonstrates that the nature of the AI race has changed. Even if it does not deliver extraordinary progress in AGI or so-called superintelligence, it could still lead to more innovation in the products and services built upon AI models.

Table of Contents

Has Generative AI Reached Its Current Limit?

The controversy surrounding GPT-5 has sparked a sharp question in Silicon Valley. After hundreds of billions of dollars in investment, has generative AI reached its limit?

This doubt not only challenges the foundation of OpenAI’s $500 billion valuation but also prompts the wider industry to re-examine the trajectory of AI development.

Despite the skepticism, enthusiasm for AI applications in the capital market and across industries remains strong. Investors appear to value the tangible growth of AI in commercial applications more than the distant promise of AGI. This shift suggests that the next phase of the AI race will focus less on pushing the limits of model capabilities and more on delivering pragmatic and cost-effective products.

The Gap Between Expectations and Reality

Over the past three years, AI researchers, users, and investors have become accustomed to rapid progress in the field. The release of GPT-5 disrupted this momentum. Technical glitches made its performance appear clumsy, leading to widespread complaints and even claims that it was inferior to GPT-4.

Sam Altman admitted the launch was “bumpy,” explaining that a malfunction in the underlying “auto-switcher” caused the system to fall back to a weaker model.

Thomas Wolf, co-founder and chief science officer at Hugging Face, noted:

“People were expecting to discover something completely new from GPT-5, but we didn’t see that.”

This disappointment was amplified by the industry’s earlier optimism. Before the launch, bold predictions about AGI’s imminent arrival dominated the conversation. Altman himself had speculated that AGI could appear as early as the Trump presidency.

Gary Marcus, professor emeritus of psychology and neuroscience at New York University and a well-known AI critic, summarized the sentiment:

“GPT-5 is a symbol of this whole path to AGI through scaling, but it didn’t work.”

At the same time, the competitive landscape has quietly shifted. Google, Anthropic, DeepSeek, and Elon Musk’s xAI have all made strides in narrowing the gap with OpenAI. The days of OpenAI’s undisputed dominance may be coming to an end.

The Law of Scale Encounters a Bottleneck

The reason behind GPT-5’s underperformance lies in the limits of the “scaling laws” that have powered the development of large language models. For years, companies such as OpenAI and Anthropic followed a simple formula: feed more data, apply more computing power, and produce bigger models.

But this approach is running into two serious challenges.

Data Exhaustion

AI companies have nearly tapped out all the free training data available online. While deals with publishers and copyright holders are being pursued, it is uncertain if these sources can provide enough material to fuel further breakthroughs.

Computing and Economic Limits

Training and running models like GPT-5 demand vast computing resources and energy. Estimates suggest that GPT-5’s training required hundreds of thousands of Nvidia’s latest processors. Sam Altman himself admitted this week that while underlying models are still improving, products like ChatGPT are “not going to get any better.”

The Specter of an AI Winter

Some researchers fear the signs of slowing progress could signal a new “AI winter,” similar to the collapse of the 1980s. Stuart Russell, a computer science professor at UC Berkeley, warned that expectations may once again be outpacing results.

He described the situation as “a game of musical chairs, with everyone vying to avoid being the last one holding the AI baby.”

Russell emphasized that inflated expectations can quickly erode investor confidence. If capital decides the AI bubble is unsustainable, “they’ll rush out the door as quickly as possible, and things could collapse very, very, very quickly.”

Yet for now, money continues to flow into AI. According to Bain & Company and Crunchbase, AI projects account for 33% of all global venture capital this year.

From AGI Dreams to Productization

The AI race may not be stalling but rather shifting direction. Instead of chasing AGI at any cost, companies are beginning to focus on turning AI into practical, profitable products.

Princeton researcher Sayash Kapoor observed that AI companies are “slowly coming to terms with the fact that they are building the infrastructure for their products.” His team found that GPT-5, while not groundbreaking, offers cost efficiency and faster performance, qualities that could support more useful applications.

Meta’s Chief Scientist, Yann LeCun, echoed this view, arguing that text-only LLMs are facing diminishing returns, but multimodal “world models” trained on video and other inputs may hold far greater potential.

This strategic pivot is already visible. OpenAI, for example, now deploys “frontline engineers” to client companies to help integrate models into real-world workflows. As Kapoor pointed out, “If companies think they’re on the verge of automating all human work, they won’t do it.”

Investors Bet on Applications

While experts debate AI’s long-term trajectory, investors appear confident in its short-term potential. AI-related stocks and startups continue to climb, with Nvidia’s market capitalization nearing $4.4 trillion and SoftBank enjoying a surge of more than 50% in the past month.

Investor enthusiasm is no longer tied to AGI hype but to tangible business growth. ChatGPT alone has generated an estimated $12 billion in annual recurring revenue. Coatue Management partner David Schneider compared its impact to Google, noting that OpenAI’s products have become “a verb.”

Many venture capitalists believe the current wave of models still has untapped commercial value. Peter Deng of Felicis argued that businesses are “just scratching the surface” of AI’s potential in consumer and enterprise markets.

As Thomas Wolf of Hugging Face put it, even without AGI breakthroughs, “there are still a lot of cool things that can be created.” For the market, this pragmatic perspective may be the most important signal of all.