Most people assume that more resources = more innovation.
Bigger budgets. Faster chips. Larger teams.
But what if that assumption is wrong?
This past couple weeks, two very different approaches to AI collided, sending shockwaves through the industry:
Project Stargate – A $500 billion mega-initiative from OpenAI, SoftBank, and Oracle to build the world’s most advanced AI infrastructure.
DeepSeek – A Chinese AI startup, blocked from using the latest NVIDIA chips, forced to find a radically different way to train models more efficiently.
Days after Stargate’s grand unveiling, DeepSeek released its latest AI model—built with a fraction of the resources. The result? A $600 billion market drawdown as investors began questioning whether we’re throwing too much money at AI rather than thinking smarter.
This is a lesson in constraint-driven innovation.
Sanctions as a Catalyst for Innovation
In 2024, the U.S. government escalated restrictions on China’s access to advanced semiconductor technology, banning exports of NVIDIA’s latest AI chips to Chinese firms. This move was part of a broader geo-economic strategy aimed at slowing China’s progress in military and AI capabilities, particularly in areas related to autonomous weapons, surveillance, and cybersecurity threats.
The Biden administration, citing national security concerns, feared that unrestricted AI development in China could strengthen:
🛡️ Military applications – AI-powered warfare, cyber operations, and autonomous weapons.
🛡️ Mass surveillance – Advanced facial recognition and predictive policing.
🛡️ Economic competition – China gaining dominance in AI-driven industries like biotech, finance, and supply chain automation.
As a result, the U.S. Department of Commerce banned NVIDIA from selling high-end AI chips like the A100 and H100 to China, forcing Chinese AI firms—including DeepSeek—to find alternative ways to develop large-scale AI models.
What many saw as a crippling setback became a catalyst for innovation.
Instead of following the typical “bigger is better” approach, DeepSeek engineered a radically different way to train AI models more efficiently—without relying on the latest U.S.-made hardware.
A disaster for China’s AI development?
Not for DeepSeek. Instead of following the typical “bigger is better” model, they engineered a way to train AI more efficiently than their Western counterparts.
Their secret? Rewriting the playbook on model efficiency.
According to the DeepSeek-V3 Technical Report:
⚙️ Smarter Processing (MLA): Instead of processing all data equally, this method prioritizes key information, reducing wasted computation and speeding up responses.
⚙️ Targeted Intelligence (MoE): Instead of activating the entire AI model, this method only uses the parts necessary for a specific task, making it faster and more cost-efficient.
⚙️ Efficient Computing: Rather than depending on expensive, top-tier hardware, DeepSeek fine-tuned its process to work seamlessly on less powerful, more affordable chips.
Their model isn’t just surviving under constraints—it’s excelling, with performance comparable to OpenAI’s GPT-4 and fully open-source for anyone to use freely. A dramatically different approach than what's happening in the U.S.
A Stark Contrast in AI Strategy
In early 2025, Project Stargate was unveiled—a half-trillion-dollar bet that AI’s future depends on sheer scale. OpenAI, SoftBank, and Oracle announced plans to:
🌍 Build AI superclusters across the U.S. and Europe
🌍 Secure long-term NVIDIA chip access to keep models growing exponentially
🌍 Outspend competitors to maintain leadership, driven by the belief that more compute equates to better performance and continued dominance in AI.
Their strategy underscores a fundamental contrast in AI development philosophies: DeepSeek optimized under constraints, while the West doubled down on raw computational power.
Just days later, DeepSeek dropped a model built under extreme constraints—and the market reacted.
📊 $600 billion in AI-related market losses.
📊 Investors started questioning if scaling AI this aggressively is even necessary.
📊 Smaller, more efficient models suddenly looked like the smarter bet.
Since then, however, the market has partially rebounded as investors reassess the landscape, likely because major players couldn't shift their positions quickly enough. Next quarter's announcements will be telling. While concerns about AI spending persist, major players have reiterated their commitment to large-scale models, stabilizing the sector.
What If There Were No Sanctions?
Now, let’s consider an alternative scenario. What if DeepSeek had access to NVIDIA’s top-tier GPUs?
Would they have developed these innovations? Probably not.
Instead, like OpenAI and SoftBank, they might have:
🔍 Scaled up infrastructure instead of optimizing algorithms.
🔍 Followed the “bigger models = better performance” mentality.
🔍 Made incremental improvements instead of fundamental breakthroughs.
This is why constraints matter.
Without restrictions, the easiest path is often wasteful, slow, and unimaginative.
The Lesson: Scarcity Breeds Creativity
This story isn’t just about AI. It’s a fundamental principle of innovation:
When you have unlimited resources, the default instinct is to spend your way to progress.
When you have constraints, you’re forced to innovate your way forward.
This is why:
🚀 NASA put a man on the moon with computers less powerful than a modern calculator.
🚀 Apple engineers optimized the iPhone battery because they couldn’t just make it bigger.
🚀 Startups disrupt industries because they have to—there’s no billion-dollar safety net.
DeepSeek’s story is proof that being forced to do more with less can lead to true breakthroughs.
How to Leverage Constraints in Your Business
If you’re building products, leading a team, or running a company, here’s how to turn limitations into advantages:
🎯 Encourage Creative Problem-Solving → Treat constraints as an innovation challenge, not a setback.
🎯 Focus on Core Strengths → Prioritize the one thing that will drive the biggest impact.
🎯 Move Fast & Iterate → Small, efficient experiments beat slow, bloated projects every time.
Final Thought: Maybe We’re Spending Too Much on AI
If DeepSeek can match OpenAI’s capabilities with 1/100th the budget, maybe AI’s future isn’t about who can spend the most—but who can think the smartest.
🌍 What’s a constraint that forced YOU to get creative?
Drop a comment—I’d love to hear your thoughts. 👇
Warmly,
Shaune Reynolds
Professor of Innovation at UNC Greensboro (Bryan School of Business & Economics)
Constraints Advocate & Consultant