Futures
Access hundreds of perpetual contracts
TradFi
Gold
One platform for global traditional assets
Options
Hot
Trade European-style vanilla options
Unified Account
Maximize your capital efficiency
Demo Trading
Introduction to Futures Trading
Learn the basics of futures trading
Futures Events
Join events to earn rewards
Demo Trading
Use virtual funds to practice risk-free trading
Launch
CandyDrop
Collect candies to earn airdrops
Launchpool
Quick staking, earn potential new tokens
HODLer Airdrop
Hold GT and get massive airdrops for free
Launchpad
Be early to the next big token project
Alpha Points
Trade on-chain assets and earn airdrops
Futures Points
Earn futures points and claim airdrop rewards
More
Algorithms and Autonomy: How China's Industry Achieves a Breakthrough in Artificial Intelligence
Eight years ago, the operations of the giant telecommunications company ZTE suddenly halted after an absolute U.S. ban. Today, in March 2026, China is making steady strides toward building a fully independent AI system that doesn’t rely on Nvidia or foreign technologies. This shift isn’t just about chips; it’s a true revolution in algorithms and strategy.
From Ban to Response: The ZTE Lesson and Tough Beginnings
On April 16, 2018, the U.S. Department of Commerce issued a comprehensive ban on ZTE Communications, which employed 80,000 people and generated annual revenues exceeding one trillion yuan. Without Qualcomm chips, base stations stopped functioning. Without Google’s Android license, phones lost a usable operating system. The company paid a heavy price: $1.4 billion in fines, plus net losses of 7 billion yuan in 2018 alone.
Former ZTE CEO wrote in an internal memo: “We live in a complex industry heavily dependent on global supply chains.” This phrase reflected resignation and reliance.
But eight years later, the scenario has changed dramatically.
Algorithms as the Solution: From CUDA to Technical Independence
The real challenge facing Chinese AI companies isn’t the chips themselves, but something called CUDA—a computing platform developed by Nvidia in 2006. This platform controls 90% of the global AI training chip market and forms the basis of nearly every AI framework, from TensorFlow to PyTorch.
By 2025, Nvidia had built a closed ecosystem: 4.5 million developers, 3,000 approved applications, 40,000 active companies. Over 90% of AI developers worldwide are tied to Nvidia’s system.
The real challenge isn’t just finding an alternative chip—it’s rebuilding an entire ecosystem of algorithms, tools, and software environment from scratch.
China’s response wasn’t direct. Instead of trying to compete with Nvidia on its turf, Chinese companies chose a completely different path: advanced algorithms.
Algorithm Revolution: Mixture of Experts Models
From late 2024 through 2025, a collective shift occurred in Chinese AI: the adoption of Mixture of Experts (MoE) models. Instead of activating a huge monolithic model, it’s divided into smaller experts, activating only the necessary parts for each task.
DeepSeek’s V3 model illustrates this concept: 671 billion parameters in total, but only 37 billion (5.5%) are activated during inference. Training cost: $5.576 million using 2,048 H800 units over 58 days. Compare this to $78 million to train GPT-4.
The result? Massive quantitative improvements in economic efficiency:
DeepSeek is 25 to 75 times cheaper than Claude.
This huge price gap has caused real chaos in global developer markets. In February 2026, Chinese model usage on OpenRouter surged 127% in just three weeks, surpassing the U.S. for the first time. A year earlier, Chinese models accounted for only 2%. After a year, it reached 60%.
A Quantum Leap: From Inference to Training
But reducing inference costs isn’t the full solution. The real challenge lies in training—which requires enormous computing power, not just auxiliary.
This is where domestic chips come into play.
In 2025, a new production line was launched in Jiangsu, stretching 148 meters, built in just 180 days. The foundation: Loongson 3C6000 processors (fully domestically designed) and T100 AI cards from Taichu Yuanqi. Production rate: one server every 5 minutes, aiming for 100,000 units annually.
Most importantly, these chips are already handling real, massive training tasks.
In January 2026, Zhipu AI, in collaboration with Huawei, launched the GLM-Image model—the first advanced image generation model trained entirely on Chinese-made chips. Then in February, China Telecom trained the “Towers” model (trillions of parameters) on purely Chinese computing hardware.
This signals one thing: domestic chips have moved from “inference” to “training”—a huge leap forward.
Software Environment: Ascend System and Ongoing Development
Behind these achievements is Huawei’s Ascend system—a local software ecosystem alternative to CUDA.
By the end of 2025:
At MWC on March 2, 2026, Huawei launched the new SuperPoD architecture for international markets. The FP16 processing power of Ascend 910B reached Nvidia’s A100 level—no longer “unsolvable,” but “usable” and “user-friendly.”
Energy: The Unmatched Advantage
The scene becomes even more complex when considering energy.
Early 2026, Virginia canceled approval for new data centers. Georgia followed (until 2027), then Illinois and Michigan. The reason: electricity.
In 2024, U.S. data centers consumed 183 TWh (4% of total electricity). By 2030, projected at 426 TWh (12%). Arm’s CEO predicts AI data centers alone will consume 20-25% of U.S. electricity by 2030.
The U.S. power grid is strained. By 2033, the U.S. will face a 175 GW energy shortfall (enough for 130 million homes). Prices have risen 267%.
China? A completely opposite scenario.
Annual Chinese production: 10.4 trillion kWh. U.S.: 4.2 trillion. China produces 2.5 times more than the U.S.
Additionally, China’s domestic consumption is only 15% of total electricity (U.S.: 36%), meaning massive industrial energy is available for redirection.
Electricity costs: US at $0.12–0.15 per kWh; western China at $0.03—one-quarter to one-fifth of American prices.
Tokens Instead of Factories: New Export Strategy
While the U.S. faces an energy crisis, China quietly goes abroad—but this time, not with products or factories, but with tokens.
Tokens, the basic unit of information for AI models, have become a new digital commodity. Produced in Chinese computing factories, then transmitted via submarine cables worldwide.
Distribution of DeepSeek users:
Supporting 37 languages, it’s popular in emerging markets like Brazil. 26,000 global companies have accounts. 3,200 institutions have deployed the enterprise version.
In 2025, 58% of new AI startups relied on DeepSeek. In China, market share reached 89%. In sanctioned countries, between 40-60%.
The Japanese Lesson: Independence vs. Dependence
In 1986, Japan signed the U.S.-Japan Semiconductor Agreement under intense American pressure. Terms: open 20% of the Japanese market to U.S. chips, ban export prices below cost, impose 100% penalties on exports.
By 1988, Japan held 51% of the global semiconductor market. Six of the top ten companies—NEC second, Toshiba third.
But after signing? Everything changed. The U.S. exerted comprehensive pressure, supporting Samsung and SK Hynix to flood the Japanese market with low prices. Japan’s share of the DRAM market plummeted from 80% to 10%. By 2017, only 7% of the integrated circuit market remained.
The Japanese lesson: accept being the best product in a global system dominated by others, but not build an independent system. When the wave receded, they had only production left.
History Repeats—But with a Different Scenario
Today, China faces a similar crossroads—but with a different choice.
Three rounds of chip restrictions (October 2022, October 2023, December 2024) with ongoing escalation. CUDA barriers remain high.
But this time, the path is entirely different:
Each step builds an independent industrial system that Japan never achieved.
Financial Reports Reveal the Truth: “War Tax”
On February 27, 2026, three Chinese chip companies published their financial reports on the same day:
Half fire, half water.
Fire: Market appetite. The gap left by Huang Renshen (Nvidia’s CEO), with 95% market share, is gradually being filled.
Water: Massive losses—not mismanagement, but a “war tax.” Heavy R&D investments, software support, engineers solving translation issues one after another.
These losses are the true price of building genuine independence.
Summary: From “Can We Survive?” to “What Is the Acceptable Cost?”
Eight years ago, the question was “Can we survive?”
Today, it’s “What price must we pay to survive?”
Changing the question itself signifies progress.
Through advanced algorithms, not just chips. By building a truly independent ecosystem, not just a better product. By long-term investment in local computing power, not reliance on external supplies.
China’s AI industry isn’t in surrender like ZTE was eight years ago. It’s in fierce battle, pushing from the front lines. But this time, there’s a real exit route.