Recently, the two major highlights in the AI track have directly brought the US-China competition to the forefront: Manus acquisition was outright rejected by China's National Development and Reform Commission + DeepSeek V4 Preview was strongly open-sourced.

First, about Manus, Meta invested $2 billion at the end of last year to acquire this Singapore-registered, Chinese-founded company led by Chinese entrepreneurs, an agent company. As a result, yesterday the NDRC issued a direct order to prohibit it, requiring both parties to cancel the deal.
The signal is very strong: core AI technology, especially agent-type technology that can be quickly deployed, will not be allowed to be taken away by foreign capital. Even if the company has already become profitable in Singapore and the money has been paid, it must be divested completely.
Next, regarding DeepSeek V4, a few days ago it released two MoE models: V4-Pro and V4-Flash, both fully open-sourced under MIT license. Hybrid attention architecture, inference FLOPs only 27% of the previous generation, KV Cache just 10%, agent coding directly SOTA.
The price is even more outrageous: the Pro version outputs 1 million tokens for only $3.48, one-tenth of GPT-5.5, and natively supports Huawei Ascend chips. These two events together have a huge impact on the AI track.
My view is that the geopolitical iron curtain has been fully drawn, and M&A routes are basically blocked. Chinese AI entrepreneurs will no longer be able to sell out to American giants in the future. In the short term, Meta loses the opportunity to replenish its agent, but in the long run, China’s ecosystem is more closed-loop, talent and capital stay domestically, and parallel tracks are fully formed.
Open source + extreme cost-effectiveness are turning AI from an aristocratic game into a mass celebration. DeepSeek V4 is not just about parameter scaling but combines strength, affordability, and openness in one package. Developers worldwide can fine-tune agents based on it, with iteration speeds exploding like Linux, while big closed-source companies find it increasingly difficult to keep up.
China has already taken the lead in the efficiency revolution. The US is still stacking computing power in closed systems, while China is directly playing with 1M+ context windows, native agent integration, and independent domestic hardware. The future winner will not be the one with the smartest model, but the one with the steepest cost curve and the ability to truly deploy AI at scale.
What I find most emotional is that Manus being rejected is a bit of a pity; technology should flow freely, entrepreneurs should have exit rights. But DeepSeek V4 really gets the blood pumping. This is the direction AI should go—democratization, open source, and efficiency. The original intention of xAI was to pursue the greatest truth; locking knowledge in a few companies is anti-progress.
Manus is a geopolitical iron lock; DeepSeek V4 is the key to pry it open with technology. The AI track has entered a new stage of China’s efficiency vs. the US’s scale. I bet on efficiency winning. What do you think? Is regulatory protection important, or is counterattack through open source more satisfying?