This is an in-depth financial analysis that dissectes the power struggles behind the current "slowing down of AI" debate in the industry. As an economist and financial journalist, I will break down the complex technical jargon and political metaphors into plain language, revealing that this seemingly about "safety" debate is actually about how to divide the profits in the AI market.
Summary of Key Points
The core argument of this article is quite sharp: Leading AI companies, such as Anthropic (Dario Amodei) and OpenAI (Sam Altman), are using the guise of "AI being too dangerous and needing slower regulation" to actually build barriers through legislation and standard-setting, keeping competitors (especially open-source models and Chinese AI labs) out and thus monopolizing the profits in the AI industry.
The author cites venture capital expert Peter Thiel's metaphor of "hypersonic weapons" to point out that those with absolute power can either become dictatorial "devils" or completely altruistic "angels," with no middle ground. The current AI giants, while claiming to be concerned about loss of control, are actually acting like "devils" by restricting competition to maintain their monopoly. The essence of this struggle is not about technical safety but about the competition for global AI market dominance and profit distribution between the American computing power sector (chips/cloud) and the model sector (applications/algorithms), as well as between Chinese and American AI forces.
---
Detailed Analysis
1. The Truth Behind "Slowing Down": A Monopoly Shield Under the Mask of Safety
First, let's understand what Dario Amodei (CEO of Anthropic) and Sam Altman (CEO of OpenAI) really mean by their "AI slowdown" proposal.
On the surface, they say, "AI is too dangerous and could get out of control, so we need to slow down and introduce regulation." This sounds responsible, like the actions of an "angel."
However, if you look at the specific proposals they make, you can see the "devilish" logic behind them:
- Third-party on-site evaluations: This means only giants with the money, legal teams, and large security departments, like OpenAI and Anthropic, can afford such compliance costs. Small companies cannot afford it.
- Limits on growth and global coordination: This essentially defines who can participate in the AI market.
- Chips restrictions targeting China: This continues to strangle competition.
In simple terms: It's like opening restaurants. Before, anyone could open a restaurant, and competition was fierce. Now, the two largest chain restaurants (OpenAI and Anthropic) suddenly say, "Food safety is too important; we need to set the strictest hygiene standards and have government inspectors check every day." As a result, small restaurants, unable to afford the inspection and renovation costs, go out of business, leaving only the two giants standing. By setting high entry barriers, they exclude potential competitors (especially open-source model teams using cheaper methods).
2. The "Hidden Winners" of Computing Power: Why Do Nvidia and Cloud Providers Support Open Source?
There is an counterintuitive economic aspect to this. Many think that the US's restrictions on open-source models are aimed at suppressing China, but the article points out that American computing power giants (such as Nvidia, AWS, Azure) actually benefit greatly from open-source models, especially Chinese ones.
Why? Let's do the math:
- Current situation: The cost of global AI operations is high, but revenue is limited.
- Open source vs. closed source: Open-source models (like Chinese models like Qwen and Kimi) are much cheaper, about 25% of the cost of closed-source models (like GPT-4).
- Where does the money go? Although open-source models are cheaper, they are used extensively. These users, although using Chinese models, need to run them on American cloud services (AWS, Azure) and buy Nvidia chips.
- Closed-source model: OpenAI spends $10 billion on training and makes $12 billion in revenue, with most of the profit staying with the model company.
- Open-source model: The open-source models may earn only $3 billion (or less, as many are free or low-cost), but this money does not all go to Chinese labs. A large portion goes to American cloud providers and chip manufacturers.
In simple terms: It's like selling water and bottles:
- Closed-source models are like selling "brand-name mineral water" at high prices, with limited sales.
- Open-source models are like selling "tap water" at low prices, which is consumed in large quantities.
- Nvidia and cloud providers are selling the "bottles" and "pipes."
- If everyone buys expensive brand water (closed-source), sales of bottles will be low.
- If everyone uses cheap tap water (open-source), the demand for bottles and pipes will increase dramatically.
- Therefore, American computing power giants want open-source models to be strong because it means more GPUs are bought and more cloud services are used. They don't care who trains the models; they care about the "traffic" flowing through their systems.
Thus, restricting open-source models actually harms the interests of the American computing power sector. This is why Nvidia CEO Jensen Huang and the US AI coordinator David Sacks oppose banning open source.
3. Two Paths Leading to the Same Goal: Consolidating American AI Hegemony
The article divides the AI camps into two seemingly opposing groups, but their goals are the same:
- Open-source camp (represented by David Sacks and Jensen Huang):
- Argument: Support open source and oppose excessive regulation.
- Logic: Open-source models create a huge demand for computing power, making American chips and cloud services the global infrastructure. As long as the world uses American chips for open-source models, the US controls the core of AI.
- Metaphor: I want to build the best highway (computing power); whatever car you drive (model), you have to pay to use it.
- Slowdown camp (represented by Amodei and Sam Altman):
- Argument: Strengthen regulation, limit open source, and protect closed-source models.
- Logic: By establishing safety standards, only a few selected giants can release models, preventing open-source models from catching up and maintaining high profits in the model sector.
- Metaphor: I want to set the strictest driving tests; only my driving schools (giant labs) can produce qualified drivers; others are not allowed on the road.
Key Insight: Although the methods differ, the end goal is the same: both camps aim to consolidate America's dominance in the entire AI stack (chips, cloud, models, applications, rules).
- The open-source camp fears that strict regulation will make American AI companies (like Harvey) unable to afford open-source models and cause business to move overseas.
- The slowdown camp fears that strong open-source models will dilute their profits and threaten their monopoly.
- Common point: Both camps assume the US must be in control and see China as a rival to be kept at bay. Regardless of the winner, the US maintains its position, with profits distributed between the "sword sellers" (computing power) and the "gold miners" (model developers).
4. The Dialectics of "Angels and Devils": No Pure Good or Evil, Only a Balance of Interests
The article concludes with Peter Thiel's view, highlighting the paradox of this debate:
- The angelic side: Amodei and others do point out the potential risks of AI (such as loss of control and job displacement), calling for caution, which is morally noble and meets public expectations of technology benefiting society.
- The demonic side: They use this caution as a political tool to exclude rivals and consolidate their monopoly, running their models on the most expensive infrastructure while asking others (open-source developers and Chinese labs) to slow down.
In simple terms: It's like a doctor (AI giant) telling patients (the public/market), "This new drug (AI) has serious side effects; for safety, only my top hospitals can prescribe it, and I will charge high fees; other clinics are not allowed." From a safety perspective, this is right, as misuse of the drug is dangerous. From an interest perspective, it's using information asymmetry and power to monopolize resources and raise prices.
Conclusion:
In today's AI landscape, there are no pure "angels" or "devils." Pure openness (no restrictions) would lead to technological chaos and unsustainable R&D investment, while complete closure (total control) would lose public trust and hinder technological progress. The giants have chosen a balance of "selective openness" and "restricted safety," using legislation and standards to turn "safety" into an entry barrier, thus limiting competition to a few players.
For the general public, it's important to understand this: When you hear AI giants calling for a slowdown, don't just see their concern for the future of humanity; also recognize the commercial logic behind their efforts to lock in market share and profits through rule-making. This debate is ostensibly about technical ethics but is essentially an economic war over global computing power and data sovereignty.