⚡ Nvidia Challenged: Huawei Ignites AI Race!

Welcome to AI Entrepreneurs

This issue covers several important developments: DeepSeek is back in South Korea after a pause due to privacy concerns, marking a key moment for AI rules in Asia. Anthropic’s Claude Code can now automate 79% of coding tasks. Huawei’s Ascend 910D is competing with Nvidia’s AI chips, and Alibaba’s Qwen3 is changing language models. Baidu’s Turbo models are very cost-effective, SCAI AI achieved a 95.2% score on the USMLE exam, and Luma AI’s Ray2 Keyframes improve the accuracy of cinematic videos.

DEEPSEEK

DeepSeek Returns: AI App Back in South Korea After Suspension

Chinese AI service DeepSeek is once again available for download in South Korea, marking a return after a two-month suspension over data protection concerns. According to Reuters, South Korea’s Personal Information Protection Commission flagged DeepSeek for unauthorized data transfers at its January launch. Regulatory scrutiny led to a revised privacy policy, allowing users to opt out of data transfers to China and the U.S.

Details:

• Privacy Overhaul: South Korea’s Personal Information Protection Commission flagged DeepSeek for unauthorized data transfers when it launched in January. The company has since updated its policy to comply with regulations.

• App Stores Restored: As of April 28, DeepSeek is back on Apple’s App Store and Google Play Store, giving South Korean users access once again.

• Regulators Watching: According to Reuters, South Korea’s privacy watchdog stated that DeepSeek voluntarily reinstated the app after reflecting at least some of its recommendations, but scrutiny remains.

Why It’s Big: DeepSeek’s return highlights South Korea’s tightening grip on data privacy, setting new expectations for AI companies operating in the region. The AI-powered app’s compliance adjustments could be a litmus test for other foreign tech firms eyeing expansion in South Korea’s competitive AI market.

ANTHROPIC

Claude Code: How Anthropic’s AI Is Reshaping Software Development

Anthropic's latest Economic Index reveals how Claude Code is quietly revolutionizing the way developers build software. With automation rates soaring and startups leading adoption, Claude is pushing the future of coding into overdrive.

Details:

• Automation Overdrive: 79% of Claude Code interactions are pure automation, compared to just 49% on regular Claude.ai. Specialized agents are moving from "helping humans" to "doing the work."

• Rise of “Vibe Coding”: Web development dominates. JavaScript, HTML, and CSS lead the charge, as developers let Claude handle front-end builds with minimal input.

• Startups Sprint Ahead: 33% of Claude Code’s early adopters are startups, far outpacing enterprise adoption. Agile teams are seizing AI’s competitive edge before bigger corporations catch up.

Why It’s Big: Claude Code isn’t just an AI assistant—it’s becoming an autonomous coder. This shift signals a massive acceleration in software development speed, reshaping how apps are built, startups scale, and innovation happens.

HUAWEI TECHNOLOGIES

Ascend 910D — Huawei’s Bold Rival to Nvidia

Huawei is making waves in the AI chip industry with the Ascend 910D, a processor designed to challenge Nvidia's dominance in advanced AI hardware. According to the Bloomberg, Huawei has approached several Chinese tech companies to test the feasibility of this chip, aiming to establish it as a powerful domestic alternative amidst U.S. export restrictions.

Details:

• Strategic Testing: Huawei is collaborating with Chinese tech firms to rigorously test the Ascend 910D’s feasibility, with initial samples expected by late May.

• Performance Goals: The chip aims to match or exceed Nvidia's benchmarks, positioning itself as a key domestic alternative for AI workloads.

Why It’s Big: The Ascend 910D reflects Huawei’s determination to innovate and compete on a global scale, reshaping the dynamics of the AI chip market. This milestone underscores China’s resilience and growing influence in the semiconductor industry.

ALIBABA

Qwen3 — Transforming Language Models with Innovation

Qwen Team introduces Qwen3, the newest addition to their lineup of large language models, redefining benchmarks in coding, math, and multilingual capabilities. With advanced design and scalability, Qwen3 stands out among the top-tier models in the AI landscape.

Details:

• Performance Excellence: Qwen3-235B-A22B features 235 billion total parameters and excels in benchmarks against models like DeepSeek-R1 and Gemini-2.5-Pro. Even the compact Qwen3-30B-A3B delivers impressive results with fewer activated parameters.

• Open Weights: Two MoE models (Qwen3-235B-A22B, Qwen3-30B-A3B) and six dense models are open-weighted under Apache 2.0, enabling researchers and developers worldwide to innovate freely.

• Multilingual Mastery: Supporting 119 languages and dialects, Qwen3 paves the way for global applications in diverse industries.

Why It’s Big: Qwen3 combines hybrid thinking modes, pre-trained datasets of unprecedented scale, and optimized agentic capabilities to deliver unmatched flexibility and precision. With its open-source commitment, Qwen3 empowers the next generation of AI-driven solutions across research, development, and production ecosystems.

BAIDU

Baidu's Turbo Models — Redefining AI Affordability and Speed

Baidu has unveiled its latest AI breakthroughs, ERNIE 4.5 Turbo and X1 Turbo models, designed to challenge industry giants like OpenAI and DeepSeek. These models deliver unmatched efficiency and multimodal capabilities at a fraction of the cost, setting a new benchmark for accessibility.

Details:

• Faster Multimodality: ERNIE 4.5 Turbo excels in logical reasoning, coding, and multimodal tasks, costing just 11 cents and 44 cents per million I/O tokens—0.2% of GPT-4.5’s cost.

• Deep Reasoning: X1 Turbo focuses on advanced reasoning with pricing at 14 cents per million input tokens and 55 cents per million output tokens, outperforming DeepSeek R1 and V3 models.

• Developer-Ready: Both models lower barriers for developers and will soon be open-sourced to foster broader adoption.

Why It’s Big: Baidu’s Turbo models are reshaping the competitive landscape by offering unprecedented performance and cost-efficiency. This move positions Baidu as a key player driving innovation and democratizing AI capabilities for developers worldwide.

AI HEALTH

AI Masters Medical Exams

University at Buffalo’s SCAI AI scores an unprecedented 95.2% on USMLE Step 3, surpassing GPT4 Omni (90.5%). Processing 13 million medical facts with semantic reasoning in seconds, it eliminates bias by excluding clinical notes, redefining physician support.

Why It Matters:
SCAI’s clinical reasoning empowers doctors, enhancing diagnostics and expanding specialty care access.

More Inside: AI detects lung cancer early, streamlines behavioral health, protects kids in ERs, and predicts kidney cancer treatment success.

Subscribe → AIHealthTechInsider.com

AI CREATIVITY

Master Keyframes in Luma AI Ray2

In this tutorial, you’ll learn how to use Luma AI’s new Keyframes feature to create precise, cinematic AI videos by setting starting and ending frames.

Step-by-step:

• 1. Access Dream Machine

• 2. Set Your Keyframes

• 3. Add a Text Prompt (Optional)

• 4. Generate and Review Your Video

Example:

• Starting Frame: A park in autumn with golden leaves falling.

• Ending Frame: The same park covered in winter snow.