AI Building AI: AMD's Bold Vision to Supercharge Development Efficiency by 20x

Introduction: A Turning Point in AI Evolution

At the GDS2025 World Digital Summit, one of the most significant developments in artificial intelligence was put on display by Sharon Zhou, Corporate Vice President at AMD (Advanced Micro Devices). In her keynote, Zhou revealed a seismic shift in AI development: the idea of AI creating AI. According to Zhou, this transformation could lead to a 20x increase in development efficiency, drastically reducing the cost and time required to build high-performance AI systems.

With global interest surging around generative technologies like ChatGPT and Midjourney, Zhou’s statement not only captured the imagination of technologists and businesses but also raised pressing questions around ethics, regulation, and accessibility in the fast-moving world of AI.

the Original

At the GDS2025 World Digital Summit hosted by Nikkei, Sharon Zhou, Corporate VP of AMD, delivered a thought-provoking presentation on the future of artificial intelligence. Her central claim: AI itself will soon be capable of developing more advanced AI, leading to a 20x increase in development efficiency. This self-improving loop could enable the production of high-performance models at a much lower cost, breaking current barriers to entry in the AI space.

Zhou highlighted the escalating cost of running state-of-the-art AI models, suggesting that the ability to automate AI development is key to achieving scalability and affordability. Her remarks are timely, considering the global rise of generative AI tools that can create text, images, and even code with minimal human input.

In particular, tools like ChatGPT (text generation) and Midjourney (image generation) have triggered explosive growth in the field, pushing countries and industries to rush the implementation of international regulations, especially concerning intellectual property and AI ethics. The increasing reliance on generative AI means that legal frameworks and industry standards must evolve rapidly to keep pace with innovation.

What Undercode Say:

Zhous bold projection isnt just

If AI systems can iteratively build and refine their successors, we’re talking about a self-reinforcing cycle of intelligence amplification. That means shorter development cycles, faster model iteration, and potentially, the democratization of powerful AI for startups and smaller nations. But it also introduces unprecedented risks—chief among them being opacity, bias amplification, and loss of human oversight.

Furthermore, the timing is critical. The AI field is already grappling with runaway compute costs, energy consumption concerns, and training data saturation. Zhou’s thesis offers a workaround—using AI to optimize itself, minimizing waste, and enhancing performance across multiple dimensions.

This has implications for national policy too. If AI design becomes cheaper and faster, AI supremacy will depend less on capital and more on strategic know-how and regulatory agility. Countries that adapt to these changes quickly will hold significant leverage.

However, with great power comes the ethical dilemma of control. Who is responsible for the behavior of an AI created by another AI? How do we audit decision-making pipelines that we didn’t directly program? And perhaps most worryingly, how do we prevent AI-generated AI from evolving unintended capabilities?

AMD is signaling that hardware must evolve in tandem with software innovation. Their message: Let us give you the silicon brains—AI will handle the rest. If that pans out, we’re entering a world where human involvement in AI development may no longer be necessary—a double-edged sword that could either liberate creativity or invite chaos.

🔍 Fact Checker Results

✅ Sharon Zhou is indeed Corporate Vice President at AMD and spoke at the GDS2025 Digital Summit.
✅ The claim of a 20x efficiency boost refers to AI developing AI and is consistent with trends in AutoML and neural architecture search.
❌ No current model independently creates new AI without human oversight—this is still theoretical.

📊 Prediction

By 2027, we expect to see mainstream adoption of AI-assisted AI development platforms in enterprise-level environments. Companies like AMD, NVIDIA, and Google will compete to provide hardware-software ecosystems that automate everything from training optimization to deployment. At the same time, regulatory frameworks will likely lag behind, forcing the tech sector to self-regulate or face backlash.

One wildcard: if AI can create AI, then geopolitical AI races may intensify—nations with the best generative platforms could outpace competitors not by talent, but by automation scale.