Silicon Valley spent billions trying to build walled gardens around artificial intelligence. Meanwhile, developers in Beijing chose a completely different path. They gave their models away for free.
It worked. Read more on a connected subject: this related article.
If you think soft power is still about Hollywood movies, pop music, or exchange programs, you're looking at a world that doesn't exist anymore. Soft power in 2026 runs on software repositories, open weight models, and the code powering global applications. China isn't trying to win hearts and minds through traditional media. They're doing it by becoming the default technical infrastructure for the rest of the planet.
This strategy shift happened fast. While Western tech giants locked their best models behind expensive API paywalls and subscription tiers, Chinese research labs and tech giants flooded the global developer community with open-source alternatives. Small startups, academic researchers, and engineers across Latin America, Southeast Asia, and Africa suddenly had access to state-of-the-art tools without paying a dime to American companies. More reporting by The Verge delves into comparable views on the subject.
That isn't charity. It's a calculated move to own the underlying standards of global technology.
The Open Source Trojan Horse
For years, the consensus among Western executives was simple: proprietary models would always maintain a huge leads over open alternatives. Open source was viewed as a sandbox for hobbyists or academic experiments.
That assumption was wrong.
When platforms like Alibaba released open-weights models like the Qwen series, they changed the math entirely. Developers didn't just test these models; they built businesses on them. Modern tech teams want control over their stack. They want to host models on their own servers, fine-tune them on proprietary datasets, and avoid getting squeezed by subscription fees or sudden platform API policy changes.
Open models offer independence. And when developers want independence, they turn away from closed ecosystems.
Think about what happens when millions of engineers worldwide build their apps on top of open models originating from Chinese institutions. The default architectures, tokenization methods, and technical standards quietly align with Beijing's tech ecosystem. Over time, switching away becomes nearly impossible.
That's real influence. It isn't forced. People choose it because it solves an immediate practical problem.
What Western Strategists Kept Missing
Washington spent years focusing on hardware restrictions. Export controls on high-end chips, restrictions on manufacturing equipment, strict trade barriers—these dominated the news cycle. The goal was simple: choke off the compute needed to train top-tier systems.
It turned out to be an incomplete playbook.
Hardware limits forced Chinese researchers to innovate on efficiency rather than pure brute-force scale. They optimized algorithms, streamlined training architectures, and squeezed every drop of performance out of mid-tier hardware. When you can't throw a hundred thousand cutting-edge GPUs at a problem, you get creative with optimization.
The result? Exceptionally fast, lightweight models that run cheaply on accessible hardware.
That technical constraint accidentally created the exact product the global market actually wanted. Most businesses don't need a trillion-parameter giant to draft internal emails or summarize support tickets. They need smaller, cheap, efficient models that can run locally. By forcing its AI ecosystem to focus on optimization, trade restrictions inadvertently helped produce models that were perfectly suited for mass adoption in emerging markets.
Why the Global South Is Choosing Open Weights
If you run a software company in Nairobi, Jakarta, or São Paulo, paying high per-token costs to access cloud APIs hosted in Northern Virginia is a massive drag on your margins. Local currency fluctuations make foreign SaaS pricing a constant gamble.
Then comes an open-weights model that performs almost as well, costs nothing to license, and can run on modest local servers.
The choice isn't even hard. It's basic economics.
Proprietary API Route:
[User Data] ---> [US Cloud Servers] ---> [Pay-per-token Fee] ---> [Lock-in Risk]
Open-Weights Route:
[User Data] ---> [Local Infrastructure] ---> [Zero Licensing Fee] ---> [Full Control]
This dynamic is quietly reshaping geopolitical alignment. When a nation's digital infrastructure relies on open software origins, its technical talent pool becomes deeply familiar with those frameworks. Universities teach courses using those tools. Local startups build workflows around them.
Soft power used to mean getting people to watch your television shows or admire your political institutions. Today, it means getting their brightest engineers to build their careers using your technical foundational blocks.
When your code becomes the base layer of another country's economy, you don't need to persuade them. You're already built into their daily operations.
The Real Risks of the Open AI Strategy
This shift isn't without serious friction or concerns.
Open weights mean less direct control over how models get used after release. Once model weights are published online, anyone can download them, strip away safety guardrails, or fine-tune them for malicious tasks. Western critics point out that open distribution makes tracking misuse or preventing harmful outputs nearly impossible.
There's also the issue of embedded bias. Every model carries the implicit values, cultural assumptions, and content filters of its training data and fine-tuning guidelines. When open models originating from heavily regulated information environments get deployed globally, they bring specific content boundaries with them.
Yet for many global developers, those concerns take a back seat to sheer utility and price. A free tool that works today beats a restrictive, costly service every single time.
How to Navigate the New AI Soft Power Reality
If you're a founder, CTO, or product manager navigating this shifted environment, relying purely on one closed model provider is a dangerous strategy. You risk cost hikes, sudden API depreciation, and vendor lock-in.
To maintain real control over your technical stack today, implement these practical steps immediately:
- Audit your current API dependencies. Map out every feature in your product that relies on third-party proprietary endpoints. Identify what breaks if prices double or access gets throttled.
- Set up a local evaluation workflow. Test your core prompts and workflows against leading open-weights models locally using frameworks like Ollama or vLLM. Measure performance, latency, and actual compute cost.
- Own your fine-tuning pipeline. Gather and curate your internal domain data now. The true value in software moving forward isn't the base model itself, but your ability to adapt open models to your specific business tasks.
- Implement a hybrid architecture. Route simple, high-volume tasks to lightweight, locally hosted open models while reserving expensive proprietary models only for complex reasoning tasks that strictly require them.
The battle for technical influence won't be settled by policy speeches or trade sanctions. It is being decided every single day by developers choosing which models to download, fine-tune, and deploy into production.
