Everyone Has Given Up on AI Safety, Now What?

One of the most overlooked aspects of AI acceleration is the rapidly declining cost of both training and inference. Just a few years ago, training a state-of-the-art AI model required billions of dollars in compute resources. Today, open-source models can be fine-tuned on consumer GPUs for a fraction of the cost.

Not only that, but API-based access to the most powerful AI systems is becoming cheaper by the month. What used to cost hundreds or thousands of dollars to generate high-quality text, images, and video now costs pennies, or nothing at all. Companies are aggressively cutting prices to compete, and new innovations in efficiency (like better quantization and hardware acceleration) are making it easier for anyone, anywhere, to harness powerful AI tools.

If AI was once the domain of elite research labs, it’s now a commodity. This has massive implications for security, job markets, and the balance of power globally.

China’s AI Revolution Despite GPU Restrictions

The United States has worked hard to limit China's access to advanced GPUs, imposing export restrictions on cutting-edge AI chips like NVIDIA’s A100 and H100. The hope was that by restricting hardware, China’s AI ambitions would be slowed down.

That strategy has failed spectacularly. Chinese research labs and companies have figured out how to build and train capable AI models even with less powerful GPUs or source-restricted GPUs through other channels. Some are using massive clusters of older hardware, while others have optimized software stacks to squeeze every bit of power out of limited compute. China has also ramped up its own domestic chip production, and while it still lags behind NVIDIA in raw performance, the gap is closing fast.

The result? The AI arms race is now fully global. The idea that the West could “contain” AI development was always naive, but now it’s outright laughable.

The genie is out of the bottle, and there is no putting it back. AI safety has become an afterthought in the race for more powerful systems. Governments are largely playing catch-up, bad actors are already taking full advantage of AI, and entire industries are being upended in real time.

There are a few possible scenarios for the near future:

Regulatory Crackdowns (Too Little, Too Late?) Governments may eventually introduce strict AI regulations, but by the time they do, AI development will have moved far beyond their ability to meaningfully control it. AI models are already open-source, and training techniques are widely known. Regulations will likely come in the form of restrictions on AI-generated misinformation, security requirements, and possibly licensing requirements for AI developers, but enforcement will be challenging.

The AI Bubble Bursts (Or It Doesn’t) Some argue that AI will eventually hit diminishing returns, and the hype will die down. However, even if we hit a plateau in model capabilities, the existing technology is already disruptive enough to permanently change the economic landscape. There’s no “going back” to a world where AI isn’t omnipresent in business and security concerns.

Acceleration to AGI (And Then?) Many AI companies are openly working toward artificial general intelligence (AGI), a system that can reason and learn like a human. If they succeed, all bets are off. A superintelligent AI could reshape civilization in ways we can’t predict, for better or worse.

AI safety, as a mainstream concern, is dead. What remains is an AI arms race between corporations, governments, and bad actors, with little oversight and rapidly diminishing costs. We are now in uncharted territory, where AI is becoming a fundamental force shaping society at every level.

So, what happens next? No one knows for sure. But what’s clear is that we are past the point of slowing down.