As 2025 began, an obscure Chinese AI startup called DeepSeek suddenly grabbed global attention with a breakthrough that upended assumptions in the tech world. In a matter of months, DeepSeek went from unknown to a frontrunner in artificial intelligence, sparking excitement, fear, and swift responses around the world.

🚀 January Launch: DeepSeek rolled out a free chatbot in Hangzhou, designed …

As 2025 began, an obscure Chinese AI startup called DeepSeek suddenly grabbed global attention with a breakthrough that upended assumptions in the tech world. In a matter of months, DeepSeek went from unknown to a frontrunner in artificial intelligence, sparking excitement, fear, and swift responses around the world.

That pulled the trigger on the AI race between China and the West.

🚀 January Launch: DeepSeek rolled out a free chatbot in Hangzhou, designed to rival OpenAI’s ChatGPT at a fraction of the cost.

💸 Budget & Chips: Trained on export-restricted Nvidia chips with under $6 million in funding, yet still delivered performance on par with Western peers.

📱 Rapid Success: Within days, DeepSeek’s assistant overtook ChatGPT in App Store downloads.

🧩 Open-weight Approach: Unlike Western rivals, DeepSeek released its model parameters, almost like open source.

⚡ Signal Effect: The breakthrough proved that cutting-edge AI can be built faster and cheaper, sparking a new global AI race.

A low‑cost Chinese AI model suddenly threatened both Big Tech’s dominance and the high‑end chip business.

🇨🇳 China: Dozens of city and provincial governments adopted DeepSeek R1 in public services, while EV giant BYD integrated it into driving-assist features.

🇸🇦 Saudi Arabia: Runs on Aramco Digital’s local data centers, fitting the kingdom’s plan to become a regional AI hub.

🚗 Tesla: Added DeepSeek for conversational voice in its China models, alongside ByteDance’s Doubao for commands.

📈 Record Popularity: The DeepSeek app briefly overtook ChatGPT as the most-downloaded free app on Apple’s App Store.

🤖 Cerebras Systems: Launched a system with DeepSeek R1 for U.S. enterprise clients seeking more cost-effective computing power.

🇨🇿 Czech Republic: Full ban in government agencies over data security concerns and possible access by Chinese authorities.

🇩🇪 Germany: The data protection authority asked Apple and Google to remove DeepSeek from app stores, citing illegal data transfers to China and far-reaching access rights.

🇮🇹 Italy: Nationwide block ordered by the privacy watchdog after unanswered questions about data use.

💼 Microsoft: Reportedly told staff not to use DeepSeek internally, citing risks around data security and propaganda.

🍎 Apple & 🔎 Google: Received takedown requests from German regulators to remove the app from their platforms.

By late 2025, the question is whether it was a short-lived shock or the start of lasting change.

In the span of a year, DeepSeek has evolved from a no-name startup into a global AI earthquake, one that has shaken up markets, bridged the East-West tech gap, and forced both companies and governments to react. Despite facing bans in some countries, the wave unleashed by DeepSeek continues to build.

Across Asia, AI has moved from buzzword to daily reality. Private companies, financial institutions, and governments know that AI is changing the world and will continue to do so.

That’s why all of these players are investing significant budgets, time, and attention in AI deployments.

We will detail this in the next section, THE BIGGER PICTURE.

Asia is racing ahead with open‑source models and sovereign stacks – from Alibaba’s Qwen 3 coding engine and Baidu’s rebuilt search to Seoul’s consortia and industry tools like iAorta and Qiyuan. Investors are backing generative commerce, agentic outsourcing deals such as Capgemini’s US$3.3 billion WNS buy, and regional exports, while Beijing’s AI+ plan calls for ubiquitous AI and global cooperation; next tests include South Korea’s national model launch and whether Baidu’s search overhaul pays off.

Models multiply: Chinese giants are now playing the role of AI fast‑fashion houses. Alibaba’s open‑source Qwen 3 coding model is marketed as matching the craftsmanship of U.S. flagships, and Baidu’s search has been redesigned around its Ernie engine, taking queries into a multimodal showroom. Even the Chinese Academy of Sciences is experimenting off the beaten path with a brain‑inspired “SpikingBrain,” hinting at new styles beyond the transformer silhouette.

Sovereign stacks: Seoul is building its own AI wardrobe. Five consortia anchored by SK Telecom, LG, and Naver are stitching together a national open‑source model from locally spun memory, chips, and cloud. The goal is a home-grown outfit that meets domestic needs while also being tailored for export to customers who prefer not to purchase from American or Chinese suppliers.

Vertical adoption: AI is moving off the runway and into real workwear. Tools like Alibaba’s iAorta analyse routine CT scans for acute aortic syndrome, while Tencent’s Qiyuan assembles ICU patient summaries in seconds. Hundreds of similar models are being piloted across finance, manufacturing, and education — signifying that industry‑specific AI is leaving the lab and clocking in.

Search to sales: Baidu’s generative revamp turns its search bar into a shopfront, raising the stakes on session length, answer share, and in‑result advertising. Investors, meanwhile, are backing the next big exports — from new open‑weight models to South Korea’s forthcoming sovereign AI — and Hong Kong’s exchanges have reopened to multibillion‑dollar AI listings after the DeepSeek shock.

Agents up for hire: Consulting and outsourcing firms are pivoting from call centres to code. Capgemini’s US$3.3 billion purchase of WNS is a bet that generative and agentic AI can re‑engineer back‑office tasks, pricing services on performance rather than headcount. Expect more deals where the service contract looks as much like a machine lease as a staffing invoice.

From labs to ledgers: Vertical AI like iAorta and Qiyuan is moving from pilot to practice; the challenge now is budgets and billing — hospitals and insurers must agree on reimbursements. Cost matters, too: Chinese open‑weight models from firms like Alibaba and Zhipu are reducing total cost of ownership and speeding integration, while Meta’s more cautious stance may nudge developers towards Asia’s lower‑priced alternatives.

Blueprints and backbones: Beijing’s “AI+” roadmap imagines an economy humming with AI at every level. It stresses connected compute, open ecosystems and shared standards, aiming to make resources more accessible and regulation more predictable for local developers.

Diplomacy by code: At a Shanghai forum, Premier Li Qiang called for an international AI cooperation organisation and warned against the technology becoming an exclusive playground. He offered China’s know‑how to the Global South — a gesture that contrasts with Western debates over restricting open source and could sway procurement choices in emerging markets.

The scoreboard ahead: The next quarter’s milestones include South Korea’s national model debut, the performance of Baidu’s search revamp, and whether Southeast Asian pilots convert proofs of concept into production deals. Progress on domestic AI chips will show if local silicon can shoulder mainstream training and inference workloads.

Asia is building a multi-stack AI economy: sovereign models, domestic chips, and open-weight ecosystems tied to real verticals. For Western firms, the opportunity is big — but it requires localization, energy-efficient deployment, and comfort operating across parallel standards.

Nvidia’s Jensen Huang says the world will pour US$3–4 trillion into AI infrastructure by the end of this decade. Think power, chips, data centers, networks, and the software layer that keeps it all humming. Set that as the horizon line.

Now look at the sprint in front of us: China disclosed 1,509 large AI models, more than 40% of the global total, alongside 5,100 AI firms and a listed-company engine driving ~70% of domestic AI revenue.

💸 Capex goes big. China’s AI spending for 2025 is projected at US$84–98B, nearly 50% more than last year. Alibaba pledged US$52B over three years for cloud and AI. Across Asia-Pacific, data centre investment hit US$180B this year, almost double 2024. Singapore, Malaysia, and Indonesia are in the spotlight — and already wondering if there is enough electricity and water to keep up the speed.

🧩 Models with breadth. China’s 1,500+ models aren’t just leaderboard entries. Ant has one for finance, RockAI built one for offline devices, NetEase made one for mining, and Alibaba is trying smart glasses. From banks to bulldozers to face-wear, the strategy is: cover all bases, fast.

🎟️ Access programs. Beijing introduced compute vouchers cutting training costs by up to 80% for SMEs. In short: cheap tickets to the AI race to empower more players. Startups, universities, and mid-tier firms now get a seat in the stadium instead of watching from outside, which will increase the AI spendings.

🔧 Chips go trillion. Industry leaders agreed that AI is the main driver pushing semiconductors toward a US$1 trillion market by 2030. Already, 39% of chipmaking equipment spending was AI-related in 2024, a figure expected to hit 55% by 2030. From cars (chip value per vehicle jumping from $922 in 2024 to $1,401 by 2030) to advanced packaging like CoWoS wafers that cost more than a car itself, AI is reshaping demand across the supply chain.

Capital flows

Retail investors. Mainland traders have put US$26B into Hong Kong AI ETFs this year. Call it “AI FOMO” with Chinese characteristics (AI错失恐惧症).

Private equity. Hermitage Capital plans US$500M for AI and robotics over three years. Their founder compared it to 1995 and the early internet. If he’s right, we’re only at the dial-up stage.

Regional examples

Hong Kong. 38% of financial institutions already use generative AI, versus 26% globally. The Cyberport AI Supercomputing Centre hit 90% utilization within months. A city once seen as a tech latecomer is catching up fast.

Southeast Asia. Malaysia has slowed approvals for new data centres to conserve power and water. Singapore still attracts hyperscalers but now puts efficiency front and center. Translation: growth is welcome, but bring your own electricity.

Why this matters

🌏 Asia’s share of AI investment and model output is growing rapidly.

🖥️ Domestic chip and system development reduces reliance on foreign suppliers.

📜 Policies are extending access to smaller firms and shaping international standards.

🏙️ Regional hubs beyond China, such as Singapore and Hong Kong, are becoming important nodes in the global AI network.

Further Read: China Daily, Asia News Network, SCMP

In 2025, if you want to sound smart, you have to use the word “AI” in every pitch, business dinner, or even on dates. But AI is so much more than just a buzzword. At asiabits, we’re sure it will change our lives for the better. It already has. With a bit of AI help, we’ve pulled together some use cases that are already changing how we commute, learn, and shop.

As the name says, you just hop into a car that takes you from A to B. No small talk about the weather, no rants about how bad “those up there” supposedly treat us.

🚖 China’s robotaxi surge: At least 19 Chinese cities are running robotaxi or robobus trials, and seven allow operators to run without a human safety driver. Baidu’s Apollo Go plans to deploy 1,000 robotaxis in Wuhan by year‑end and to operate in 100 cities by 2030.

🌴 Desert rides: Baidu, Pony.ai, and WeRide struck deals with Dubai’s Roads & Transport Authority to run autonomous taxis in Dubai and Abu Dhabi from 2025, with 25% of trips targeted to be driverless by 2030. WeRide is also expanding into Saudi Arabia, where officials aim for 15% driverless trips by 2030.

🗺️ Mapping & micro routes: In Japan, Waymo began mapping Tokyo in April 2025, sending 25 Jaguar I‑PACE cars on local roads with partners Nihon Kotsu and Go. Singapore’s Land Transport Authority granted pilot robotaxi services to Grab + WeRide and Pony.ai + ComfortDelGro, with 12‑km shuttle routes in Punggol slated for 2026.

Analysts project the robotaxi market will surge to roughly US$45.7 billion by 2030 with Asia-Pacific as the biggest region.

School’s out? Not quite. AI is rewriting the curriculum.

🍎 Homework by chatbot: In April 2025, China declared that AI would be woven into teaching, textbooks, and curricula at all levels, aiming to foster independent thinking, problem‑solving, and cooperation.

📚 Test prep goes AI: Seoul‑based Riiid runs profitable AI tutors for the TOEIC exam in Japan and South Korea, licences the tech to education firms and plans to deploy it in public schools to offer real‑time feedback and reduce inequality.

🚗 Driver’s-ed gets virtual: At Beijing’s Eastern Pioneer driving school, more than 600 cars use onboard AI instructors and VR simulators, boosting pass rates and safety.

The global AI‑in‑education market is expected to soar from about US$5.9 billion in 2024 to roughly US$32.3 billion by 2030, suggesting more capital and innovation heading to classrooms.

🧑‍🎤 Digital divas: TopSocial India’s virtual model Kyra has more than 240,000 Instagram followers and even walked in a metaverse fashion show.

🧏‍♀️ Forever 22: South Korea’s avatar Rozy boasts over 160,000 followers and more than 100 sponsorship deals, with earnings forecast above 1 billion won (US$716,000).

🛍️ Brands love bots: In China, hyper‑real digital performers like Yuri are starring in campaigns for companies such as The North Face, showing how synthetic influencers are entering the US$11 billion influencer market.

The global virtual‑influencer market is projected to jump from about US$6.1 billion in 2024 to roughly US$45.9 billion by 2030, suggesting brands will continue pouring money into AI‑powered marketing.

🧠 From Software Distributor to Tech Mega-Investor: Masayoshi Son founded SoftBank in 1981, built the Vision Fund into the world’s largest tech investor from 2017, placed early bets on Alibaba and Arm, and is now steering SoftBank squarely toward AI.

After tough years, the Vision Fund returned to profit with $4.6 billion in the fiscal year to March 2024 and just recently hit a record high after a quarterly profit of $3.7 billion beat expectations due to massive AI bets. Son’s playbook focuses on a few key platform bets and the ability to adjust the portfolio in response to changing facts rapidly.

🤖 From Village Kid to AI Pioneer: Born in a small village in Guangdong, he founded the AI startup DeepSeek. In January 2025, he launched the DeepSeek-R1 model, trained on 2,048 GPUs for just $5.6 million. The accompanying free app took the U.S. App Store by storm, knocking ChatGPT from the top spot.

With a personal fortune of $1 billion (Forbes #2933), the 40-year-old ranks among China’s most influential tech visionaries and was already invited to expert roundtables by Premier Li Qiang and Xi Jinping in 2025.

🇰🇷 🤖 AI Express for Growth: Seoul puts ₩150 trillion on the table

South Korea plans a major five-year push across 30 AI and deep-tech projects. The core is 15 AI initiatives, from humanoid robots and autonomous mobility to AI-driven factories and faster drug approvals. Another 15 bolster materials, energy and new industries like SiC power chips, superconductors, graphene, offshore wind and SMRs. It’s funded by a National Growth Fund worth ₩150 trillion (about US$108B). Half comes from KDB, the rest from pension funds, financial institutions and citizens.

AI Expressway. Not asphalt, but a fast lane of data centers, GPU capacity and quicker approvals, anchored by the new Ulsan datacenter. The goal is real productivity gains and stronger GDP growth.

In short: plan, compute, capital — and the economy shifts up a gear.

🇨🇳 🏎️ AI turbo, speed limit on: China is driving AI on two tracks.

Xi warns provinces against over-investing in EVs and AI. Overcapacity, price wars, and local debt are already dragging on the economy. Instead of pouring more concrete for data centers, Beijing plans a state-orchestrated network to pool excess compute and route it where it’s needed. Power should flow to the work.

The State Council’s “AI Plus” sets the long route:

New rules add the guardrails. Mandatory labels for AI content since September 2025, and strict facial-recognition rules since June 1, 2025.

Nvidia’s Jensen Huang called China’s open-source AI a “catalyst for global progress” and praised the country’s rapid innovation.

Left-lane mindset. We love speed and signs. Keep the balance, then hit the gas in the AI race.

🇯🇵 💸 SoftBank’s AI diet: fewer scouts, bigger bets.

Masayoshi Son is cutting roughly 20% of the Vision Fund staff as he redirects people and cash to large AI bets. SoftBank will move staff onto a few flagship AI projects. The biggest is “Stargate,” a proposed $500 billion buildout of data centers in the US that OpenAI would use. The company says the aim is to back a small number of large, long-horizon AI investments that can create durable returns.

On the balance sheet, SoftBank rebuilt a roughly US$4.8 billion Nvidia stake by June, added TSMC and Oracle, and put about US$2 billion into Intel, while selling chunks of T-Mobile and trimming Deutsche Telekom to free up capital.
The stock ripped to records this summer on the “all-in on AI” story and achieved a profit rebound.

Reality Check. Stargate and the Japan joint venture face delays. Many companies still see weak returns from generative AI. OpenAI’s monetization path is uncertain, and SoftBank may scale back funding if milestones slip. Son is narrowing the focus to chips, compute and models, backed by a large cash cushion, to turn the AI story into cash flow.

🤖 Robots Race and Stumble: China’s first “robo Olympics” in August drew more than 500 humanoid robots from 16 countries to Beijing’s National Speed Skating Oval.

They sprinted the 100 m, played basketball and sorted medicine, alternating between jerky tumbles and glimpses of real power. A five‑a‑side football match looked like slapstick as seven‑year‑old‑sized robots fell over one another, yet a domestic champion still managed to complete a 1,500 m race in under seven minutes and even shrugged off crashing into a human.

The spectacle was playful, but behind it is a serious push: China has put humanoids at the centre of its national strategy, announced a huge fund for robotics and AI start‑ups, and already buys more industrial robots than any other country.

For Asia, the games were a sign of how eager the region is to learn from each stumble.

See you next time in your inbox,