2025: The AI Race Shifts from Foundation Models to Applications

2025: The AI Race Shifts from Foundation Models to ApplicationsIn 2025, the competition in the artificial intelligence (AI) field will shift from building foundational models to developing user-friendly applications built upon them. While tech giants like OpenAI, Google, Meta, and Anthropic continue to invest heavily in developing more powerful and intelligent "cutting-edge" foundation models, boasting significant advancements in reasoning, task handling, and accuracy, the public remains largely untouched by the transformative power of AI

2025: The AI Race Shifts from Foundation Models to Applications

In 2025, the competition in the artificial intelligence (AI) field will shift from building foundational models to developing user-friendly applications built upon them. While tech giants like OpenAI, Google, Meta, and Anthropic continue to invest heavily in developing more powerful and intelligent "cutting-edge" foundation models, boasting significant advancements in reasoning, task handling, and accuracy, the public remains largely untouched by the transformative power of AI. AI-powered summaries in Google search results or Facebook prompting users about follow-up questions on posts don't represent the true arrival of the AI era.

This is set to change in 2025. This year, the focus will shift to innovators dedicated to making AI more practical and accessible to a wider audience. The emergence of DeepSeek R1 is a prime example. Developed by a Chinese company, this AI model rivals flagship products from OpenAI et al. in certain aspects, yet boasts significantly lower training costs. DeepSeek's success lowers the barrier to entry in the AI space, with some experts even suggesting that large language models (LLMs) are becoming a high-value "commoditized" technology. This foreshadows, and ultimately validates, a shift in AI competition towards serving a broader audience.

However, the situation is not so simple. Leading AI companies still plan to pour hundreds of billions of dollars into even larger models, despite unclear economic returns on current investments. The potential for disruptive technological leaps drives these massive investments. The key focus of AI competition in 2025 will be developing applications that convince even skeptics of AI's importance, mirroring the transformative impact of the smartphone.

Venture capitalist Steve Jang, who has invested in companies like Perplexity AI, Particle, and Humane, agrees. He believes DeepSeek is accelerating the "commoditization of high-value LLM development." He reflects on recent developments: Following the 2022 launch of consumer-facing models like ChatGPT, built on transformer architecture, developers quickly created a wave of "fast-food" applications. By 2023, "AI wrappers" became the norm. However, a counter-trend emerged, with startups delving deeper into AI's potential to build truly exceptional products. "The debate was, 'Are you just wrapping AI, or are you building something truly valuable?'" Jang explains. "'Are you using AI at the core while creating something uniquely distinctive?'" The answer is clear: simple "AI wrappers" are no longer favored.

Similar to the early iPhone era, where the market transitioned from clunky web apps to powerful native apps, the AI winners will be those who deeply leverage the technology. What we see now is just the tip of the iceberg; the AI equivalent of an Uber-like "killer app" is yet to emerge. But just as unlocking the iPhone's potential took time, the opportunity awaits those who dare to seize it.

"If we paused everything and just used the AI we have today, we could build new products for at least five to ten years," says Josh Woodward, head of Google Labs, whose team launched NotebookLM in late 2023 an AI tool far beyond a simple LLM wrapper and quickly garnered a loyal following.

Generative AI has already profoundly impacted some sectors, with programming being a major beneficiary; many companies claim AI bots complete 30% or more of their internal engineering work. From medical research to grant applications, AIs role is expanding. The AI revolution is here, albeit unevenly distributed. For most, however, utilizing it still requires overcoming a steep learning curve.

This is about to change dramatically. Agents capable of performing diverse tasks, most importantly assisting people in easily leveraging AI without deep expertise, are emerging. However, developers face a challenging reality: granting autonomy to software robots is risky, especially when AI is far from perfect.

Clay Bavor, co-founder of Sierra, which builds customer service assistants for businesses, says the latest generation of LLMs marks a crucial turning point in the evolution of robots towards true "agents." "We've crossed a key threshold," Bavor says. Now, Sierra's agents not only handle product complaints but also automatically place orders and arrange replacements, sometimes even devising innovative solutions beyond their training.

Looking back at 2025, people may not remember one specific blockbuster AI application, but rather the multitude of new tools that collectively drove the transformation. Steve Jang aptly compares it to asking, "'What new products came out of the age of electricity?'" he adds. "Will there be a 'killer app'? In fact, it will spawn an entire economic ecosystem."

Therefore, 2025 will undoubtedly witness an explosion of AI applications. However, avoid viewing Google, OpenAI, and Anthropic solely as model providers. These companies are relentlessly building even more advanced systems that will make our current ones seem "as dumb as rocks," further raising the bar for AI application development. This will intensify competition in the AI application field in 2025 and foreshadow a future focused on practicality and user experience. Ultimately, companies that truly understand user needs and develop applications solving real-world problems will triumph.