China This Week: Big Chip Fund, Property Aid Spreads — DeepSeek Is Freaking Out the AI World
Introduction
This week in China, two major developments made global headlines: the relaunch of its Big Fund to support domestic semiconductor manufacturing, and an expansion of property sector aid to stabilize the real estate market. But among insiders in the tech industry, another story is creating far more buzz: the rise of DeepSeek, China’s rapidly advancing open-source AI project.
While much of the world is focused on the financial and policy ramifications of China’s economic maneuvers, the AI world is on edge. Why? Because DeepSeek is doing what few thought possible — challenging OpenAI, Anthropic, and Google in both performance and accessibility.
1. The Big Fund Returns: Semiconductor Sovereignty
The China Integrated Circuit Industry Investment Fund, also known as the Big Fund, has relaunched with a renewed focus:
$47 billion (USD equivalent) allocated to national chip initiatives
Priority for AI accelerator chips, GPU alternatives, and fabrication tools
Targeted partnerships with Huawei, SMIC, and Loongson
This fund is widely seen as China’s response to Western export controls on NVIDIA and AMD chips. More importantly, it serves as strategic infrastructure for AI projects like DeepSeek.
“To build sovereign AI, we need sovereign chips.” – Ministry of Industry and Information Technology (MIIT)
2. Property Aid Spreads: Economic Stability First
China also expanded policies to support the real estate sector:
Lowered mortgage rates for first-time buyers
Liquidity injections into major urban property developers
Pilot home purchase subsidies in 10 major cities
These efforts aim to restore consumer confidence and stabilize housing prices. However, they are also freeing up space for tech sector investments, as capital shifts from distressed real estate toward growth industries like AI.
“The economy is rebalancing — from bricks to bytes.” – Economist at Peking University
3. DeepSeek’s Disruptive Momentum
Now let’s turn to the story that’s sending shockwaves through global AI circles: DeepSeek.
In just 12 months, DeepSeek has gone from an experimental project to a fully functional LLM ecosystem that:
Releases open-source weights
Offers a scalable API at 1/50th the cost of GPT-4
Outperforms GPT-4.5 in math and coding benchmarks
Can be deployed offline, on local hardware
Benchmarks That Frighten the Giants
| Task | GPT-4.5 | DeepSeek V3 |
|---|---|---|
| GSM8K (math) | 88.8% | 89.3% |
| HumanEval (code) | 67.0% | 65.2% |
| MMLU (knowledge) | 86.4% | 87.1% |
Cost Comparison (Output Token per 1M)
| Provider | Cost |
| OpenAI (GPT-4) | ~$30–60 |
| Anthropic (Claude 3.5) | ~$60+ |
| DeepSeek V3 | ~$1.12 |
“DeepSeek isn’t just competitive — it’s disruptive.” – AI Analyst, Hugging Face
4. Why the AI World Is Anxious
Several factors are causing ripple effects:
a. Open Weights
DeepSeek’s weights are freely available via Hugging Face and GitHub. This gives:
Researchers access to state-of-the-art models
Enterprises freedom to deploy AI on-prem
Regulators fewer concerns about data sovereignty
b. Local Deployments
DeepSeek supports:
LM Studio
GGUF format
Ollama / WebUI integrations
This means developers in Asia, Europe, Latin America, and even the U.S. are adopting DeepSeek for:
Cost savings
Privacy compliance
Avoiding censorship filters
c. Accelerated Model Evolution
In 2025 alone, DeepSeek released:
R1 (671B parameters)
V3 (MoE architecture, 37B active)
V3-0324 update (faster reasoning, better design output)
“They’re innovating faster than the West expected.” – Former OpenAI engineer
5. Reactions From Silicon Valley
While public comments have been muted, private Slack channels and Discord servers tell a different story:
Developers:
Testing DeepSeek’s API side-by-side with GPT-4
Impressed by speed and token context
Founders:
Considering DeepSeek for internal tools
Evaluating DeepSeek’s stability for chatbots and assistants
VCs:
Funding Western alternatives like Mistral, Groq, and xAI
Concerned about OpenAI’s pricing inflexibility
“DeepSeek is what GPT-J and LLaMA could have been if they had hardware and state backing.” – ML Researcher
6. Government Responses
U.S.:
Export controls on AI chips remain in place
Increased funding for open-source AI research (NSF, DARPA)
Calls for AI safety alignment continue
China:
Backing DeepSeek through infrastructure, not censorship
Positioning DeepSeek as a flagship of soft power tech diplomacy
Pitching DeepSeek API to Belt and Road countries
“AI will be to the 2020s what 5G was to the 2010s.” – Foreign Policy Institute
7. What’s Next for DeepSeek?
Roadmap Highlights:
DeepSeek V4 (Multimodal, Arabic + Spanish support)
DeepSeek Cloud SDK (for enterprise integration)
Education bundle for university AI labs
Fine-tune-as-a-Service platform
“DeepSeek is becoming an ecosystem, not just a model.” – Chinese AI commentator
Conclusion: The Week China Shook AI Again
While investors were watching the return of China’s Big Fund and the expansion of real estate aid, the most important shift may have happened in software.
DeepSeek’s rise is more than a tech story — it’s a signal. The world’s next AI superpower may no longer be hypothetical. It may already be training its next trillion-parameter model in the cloud.
“Everyone was watching chip fabs and housing markets — but AI just became China’s new growth engine.”
