Nasdaq recovered from DeepSeek shock on US AI dominance hope

·       But China’s rapid progress in AI and related ecosystems including high-end chips maybe posing a serious threat to US global tech dominance

·       Wall Street Futures, Gold, and USD waved on a hawkish hold by Fed statement followed by a less hawkish Q&A by Chair Powell to balance

·       Fed may not cut more than 50 bps in 2025; the next Fed cut may be in June’25 followed by Dec’25

·       Powell made it quite clear Trump’s big policy change narrative on tariffs, immigration, fiscal (tax cuts), and also various regulations may impact the US labor market and price stability

·       Thus the  Fed has to wait to assess Trump’s actual implemented policy to model its impact on US price stability and employment

·       The Fed will not heed any narrative from any elected politician including the President of the US on monetary policy stance

·       The Fed believes in institutional independence and is not in a hurry to cut rates further

On Monday, Wall Street Futures slid on the US-Columbia trade war drama and also the US-China tech war. Nvidia and other AI chip-related stocks slumped as China’s increasingly popular generative AI DeepSeek is providing chat GPT-like services with less powerful processors. Chinese AI company DeepSeek's capabilities propelled it to the number one spot in the Apple App Store. US stock futures tumbled early Monday driven by growing concerns about the US's dominance in AI. Global markets crashed led by US techs on cheap China AI ecosystem panic.

Chinese startup DeepSeek's latest AI model praised for its cost-effectiveness and ability to operate on less-advanced chips, garnered significant attention as its AI Assistant overtook ChatGPT to become the top-rated free app on the US Apple App Store. Japan chip stocks also tumbled as DeepSeek's challenge to U.S. AI dominance raises worries for Asian tech firms. DeepSeek generated excitement with tech analysts claiming the company's AI model outperforms the world's best Chatbots at a tenth of the US rivals such as Chat GPT. China's AI Breakthrough may also lead to weaker dollar as US global tech dominance may be seriously challenged in the coming days. Also, Trump tariffs could raise prices on technology like laptops/PC, smartphones, and AI.

DeepSeek, ChatGPT, and Perplexcity are all generative AI models, but they have distinct differences in their architecture, functionality, and applications. DeepSeek is a large language model developed by Meta AI that focuses on generating text summaries of long-form content, such as articles, research papers, and books. It uses a transformer-based architecture and is trained on a massive dataset of text from the internet.  

ChatGPT is a conversational AI model developed by Meta AI that can engage in natural-sounding conversations with humans. It uses a transformer-based architecture and is trained on a massive dataset of text from the internet. Perplexity is a generative AI model developed by the University of California, Berkeley that focuses on generating text that is both coherent and diverse. It uses a transformer-based architecture and is trained on a dataset of text from the internet.

DeepSeek is likely optimized for enterprise use, with competitive pricing for specialized tasks (e.g., coding, and data analysis). It May offer volume discounts or tailored solutions and targets cost-sensitive industries (e.g., China’s tech sector), while focusing on efficiency in specific domains, reducing compute overhead. The premium version is cheaper compared to ChatGPT and Perplexcity.

DeepSeek excels in specialized domains (coding, math, Chinese-language tasks) and is optimized for technical accuracy and efficiency in narrow applications, but may be less robust in creative or open-ended tasks compared to general-purpose models. DeepSeek excels in technical tasks such as coding and reasoning, often providing faster responses in programming-related queries. It is particularly noted for its efficiency in generating code in languages like Python and Java.

ChatGPT may be the best-in-class for versatility, creativity, and coherence (especially GPT-4). It’s strong in dialogue, content generation, and complex problem-solving, but prone to hallucinations; and struggles with real-time data without plugins. ChatGPT is recognized for its versatility in creative writing and conversational tasks, offering features like memory functionality that allows it to remember details from past interactions. This makes it more suitable for general-purpose narrative generation and engaging dialogues.

Development Costs:

 DeepSeek's development costs are reported to be around $5.5 to $6 million, utilizing less powerful hardware and fewer resources compared to ChatGPT. In contrast, ChatGPT has seen expenditures exceeding $100 million for its development and ongoing enhancements. Overall, the DeepSeek ecosystem requires less powerful chips and lower energy requirements than both ChatGPT and Perplexcity as it uses a unique algo and feeding/trained model.

In summary, DeepSeek stands out for its cost-effectiveness, particularly appealing to developers and users looking for an affordable solution without compromising on technical performance. In contrast, ChatGPT offers a richer set of features that cater to creative applications but come at a higher cost.

DeepSeek, ChatGPT, and Perplexity AI are prominent generative AI models, each with unique strengths in terms of cost-effectiveness and quality. DeepSeek has garnered attention for developing AI models at a fraction of the cost of its competitors. For instance, its R1 model was developed for approximately $6 million, compared to over $100 million for OpenAI's GPT-4. This efficiency is achieved through the use of less powerful GPUs and optimized training (data feeding) processes.

In evaluations, DeepSeek performs well in basic tasks such as solving physics, math, and coding problems and providing cheerful responses. However, it may struggle with real-time queries and large data processing. Additionally, it implements stringent guardrails to comply with Chinese government requirements, resulting in blocked responses for politically sensitive topics.

OpenAI' GPT-4, which powers ChatGPT, required a substantial investment exceeding $100 million for development. The operational costs are also higher due to the use of advanced hardware and extensive computational resources. ChatGPT is renowned for its comprehensive and coherent responses across a wide range of topics. It excels in real-time data processing and maintains response integrity, making it a preferred choice for unbiased and straightforward general queries.

Development of Generative AI Chips in China

China's pursuit of generative AI technology is significantly influenced by its efforts to develop advanced AI chips, particularly in light of restrictions imposed by the US on semiconductor exports. This has led to innovative strategies and rapid advancements within the Chinese tech landscape.

Domestic AI Chip Production: In response to U.S. export bans, China is accelerating the development of its domestic AI chip technology to reduce reliance on foreign chips. Chinese companies like Baidu, Cambricon, and HiSilicon have introduced high-performance AI chips, including Baidu's Kunlun series, Cambricon's MLU series, and HiSilicon's Ascend series. These chips have demonstrated superior energy efficiency compared to some foreign counterparts. The Chinese government has introduced substantial subsidies, amounting to approximately $48 billion in 2024, to bolster its semiconductor industry and support companies like Baidu, Cambricon, and HiSilicon in producing high-performance AI chips.

Innovative Approaches:

Chinese AI startups like DeepSeek have emerged as significant players in the AI sector, developing models that rival those from major U.S. companies at a fraction of the cost. For instance, DeepSeek's R1 model reportedly required only $6 million in computing power for training, showcasing efficient use of available resources and innovative training techniques. This model's success highlights how constraints can lead to creative solutions in chip utilization and model training.

RISC-V Architecture Adoption:

To achieve self-sufficiency in chip production, China is increasingly adopting the RISC-V architecture for its AI chips. This shift aims to overcome challenges posed by the dominance of established architectures like x86 and ARM, facilitating the development of custom solutions tailored to specific AI applications.

Collaborative Training Techniques:

Chinese researchers have developed methods to train generative AI models across multiple data centers using various GPU architectures. This approach allows them to maximize the utility of limited high-performance chips while integrating lower-end GPUs, thus maintaining progress despite sanctions that limit access to cutting-edge technology.

Impact on Global AI Landscape China's advancements in generative AI and chip technology are reshaping the global landscape.

Competitive Models: The emergence of models like Yi-Lightning from 01.AI indicates that Chinese firms are capable of developing competitive generative AI solutions without relying on the extensive resources typical of their Western counterparts.

Patent Leadership: China leads globally in generative AI patents, reflecting its active engagement in research and development (R&D) within this domain. However, challenges remain in terms of the quality and global competitiveness of these patents.

Market Dynamics: With over 50 companies developing AI chips and a growing domestic market for these technologies, China's focus on self-reliance is likely to disrupt existing market dynamics and challenge established players in the U.S. and beyond.

Also, Nvidia is not the only company making chips for AI workloads. AMD MI300X and Intel Gaudi 3 chips rival Nvidia's performance. Companies like Google, Amazon, and Microsoft are developing in-house custom silicon, purpose-built for AI workloads. However, the wide adoption of Nvidia's CUDA platform among developers makes it harder for companies to switch.

DeepSeek's Breakthrough: The Chinese AI startup DeepSeek has developed models that rival leading American counterparts at a fraction of the cost. By employing innovative technical approaches, such as efficient pre-training and reinforcement learning, DeepSeek trained its models for approximately $5.6 million, compared to the $100 million to $1 billion range typical in the U.S. This achievement has disrupted the AI industry and challenged the dominance of U.S. companies.

Investment Growth: China's investment in generative AI is projected to surge with an estimated 86% compound annual growth rate over the next five years. This growth is driven by a focus on technological self-sufficiency, encompassing applications to chips, and a strong emphasis on locally developed technology.

Challenges:

Despite these advancements, Chinese companies face challenges in catching up to global leaders. For instance, Semiconductor Manufacturing International Corporation (SMIC), a key player in China's chip manufacturing, only began large-scale production of 14 nm chips in 2022 and remains one to two generations behind leading competitors.

In summary, China's development of generative AI chips is progressing rapidly, driven by substantial investments and a strategic focus on self-reliance. While significant achievements have been made, ongoing efforts are necessary to overcome existing challenges and achieve parity with global leaders in AI chip technology. China's development of generative AI chips is a critical component of its national strategy to achieve technological self-sufficiency and global leadership in artificial intelligence. Driven by geopolitical tensions (e.g., U.S. export controls on advanced semiconductors) and the need to support its booming AI industry, China is aggressively investing in domestic AI chip design and manufacturing.

Geopolitical Pressures: U.S. sanctions (e.g., restrictions on NVIDIA's A100/H100 GPUs) have accelerated China's push for semiconductor independence.

AI Industry Growth: China’s generative AI sector (e.g., Baidu’s ERNIE, Alibaba’s Tongyi Qianwen, and DeepSeek) requires high-performance chips to train and deploy large language models (LLMs).

National Policy: Initiatives like "China Standards 2035" prioritize AI and semiconductor R&D to reduce reliance on foreign technology.

Domestic Chip Designers

Huawei (Ascend Series): Ascend 910 AI accelerator competes with NVIDIA’s A100 in training LLMs. Used in Huawei’s cloud infrastructure and partnered with domestic AI firms.

Integrated with Huawei’s MindSpore AI framework for software-hardware optimization-Biren Technology: BR100 GPU targets generative AI workloads, claiming 2x NVIDIA A100 performance in some benchmarks. But faces challenges due to U.S. sanctions limiting access to TSMC’s advanced nodes.

Horizon Robotics: Focuses on edge AI chips (e.g., Journey series) for autonomous vehicles and robotics, supporting lightweight generative models. Cambricon: Specializes in neuromorphic chips (e.g., MLU series) for AI inference, used in data centers and IoT devices.

Foundries & Manufacturing

SMIC (Semiconductor Manufacturing International Corp): Produces 7nm chips (e.g., for Bitcoin miners) but lags behind TSMC/Samsung in yield and efficiency for AI-specific designs. U.S. restrictions limit access to EUV lithography machines, hindering sub-7nm progress. Hygon JV with AMD (pre-sanction) producing x86 CPUs and DCU accelerators for data centers.

Software Ecosystem: China is developing homegrown AI frameworks (e.g., Baidu’s PaddlePaddle, Huawei’s MindSpore) to optimize performance on domestic chips and reduce dependency on CUDA (NVIDIA’s platform).

Challenges

Advanced Manufacturing: Lack of EUV lithography machines restricts China to 7nm+ nodes, limiting chip performance and energy efficiency.

IP & Ecosystem Gaps: NVIDIA’s CUDA ecosystem and software libraries remain dominant; Chinese alternatives are still maturing.

Talent Shortage: Competition for semiconductor engineers and AI researchers, exacerbated by global recruitment barriers.

Supply Chain Vulnerabilities: Reliance on imported equipment (e.g., ASML’s EUV) and materials (e.g., high-purity silicon).

Government & Industry Collaboration

National IC Fund: A $50B+ state-backed fund to subsidize semiconductor R&D and production.

Regional Clusters: Hubs like Shanghai’s Zhangjiang Hi-Tech Park and Beijing’s AI zones foster chip-AI synergy.

Military-Civil Fusion: Dual-use AI chip research (e.g., for defense applications) is prioritized under national security mandates.

Implications for Generative AI

Cost Efficiency: Domestic chips could lower compute costs for Chinese AI firms (e.g., DeepSeek, Baidu), enabling scalable LLM training.

Performance Trade-offs: Current Chinese GPUs lag behind NVIDIA’s H100 in raw performance but are closing the gap for specific workloads.

Localized Innovation: Hybrid architectures (e.g., integrating optical computing) and software-hardware co-design are emerging to bypass hardware limitations.

Global Impact

Market Competition: Chinese chips (e.g., Ascend, BR100) may undercut NVIDIA/AMD in price for emerging markets (e.g., Southeast Asia, Africa).

Decoupling Risks: A bifurcated AI chip ecosystem could emerge, with China and the U.S. developing parallel standards.

Export Controls: China may restrict its AI chip and rare earth materials exports in retaliation, as seen with gallium and germanium export curbs in 2023.

Future Trends

RISC-V Adoption: China is investing heavily in RISC-V open-source architectures to bypass ARM/x86 licensing restrictions.

Chipset Technology: Modular designs (e.g., Hygon’s DCU) could help compensate for process node disadvantages.

Neuromorphic & Optical Chips: Research into brain-inspired and photonic computing for energy-efficient AI acceleration.

Conclusion

China’s generative AI chip development is a high-stakes race combining state-backed investment, geopolitical urgency, and innovation under constraints. While still trailing U.S. leaders in cutting-edge performance, China’s progress in alternative architectures and software hardware integration could make it a formidable player in the global AI hardware market by 2030. Success hinges on overcoming manufacturing bottlenecks and building a robust software ecosystem. DeepSeek AI shows amazing and impressive performance despite working with less-advanced chips.

China's development of generative AI chips is characterized by a strategic pivot towards domestic production, innovative training methodologies, and an emphasis on self-sufficiency. These efforts not only aim to counteract U.S. export restrictions but also position China as a formidable competitor in the global AI landscape, with significant implications for the future China has been making significant strides in the development of generative AI chips, aiming to reduce reliance on foreign technology and establish a self-sufficient AI ecosystem. This effort has been particularly pronounced following U.S. export restrictions on advanced chips to China. Also, Chinese AI applications now have bigger aims — they're looking beyond Chabot.

On the Trump tantrum front, there were also some reports:

·       Trump to Impose Tariffs and Sanctions on Countries Refusing to Accept Migrant Returns

·       Trump wants to collaborate with Congress on tax cuts

·       Trump aims to maintain low taxation by lowering taxes

·       Trump to Keep Promise on No Tax on Tips and Social Security

·       Trump plans to cut taxes to 15% for American-made products

·       Trump to Impose Tariffs on Foreign Computer Chips Soon

·       Trump suggests building plants in the US to avoid tariffs

·       Trump to Impose Tariffs on Steel, Aluminum and Copper

·       US to Impose Tariffs on Foreign Computer Chips

·       US to Impose Tariffs on Semiconductor Imports

·       Trump: We will impose tariffs (sanctions) on individuals who intend to cause harm

·       Trump warns that the release of DeepSeek AI should serve as a wake-up call, although it’s a good competition

·       Trump says DeepSeek is a positive - a 'wake up' call

·       Trump: Chinese startup DeepSeek should be a wakeup call

·       Trump vows to produce all defense systems in the US

·       Will place tariffs on computer chips, pharmaceuticals

·       Trump to Use Campaign Funds for House and Senate Races

·       White House Halts Federal Government Grants and Loans (including US states)

Market impact:

Monday's market plunge shattered a market narrative that had prevailed since Trump's re-election in November: America-first, tech-fueled hyper-bullishness that saw a clear upward path for risky assets propelled by deregulation, tax cuts, and even government sponsorship of AI investment. Stocks fell on both sides of the Atlantic as well as the Pacific.

Monday's market mayhem was precipitated by the emergence of DeepSeek's latest AI model to the top of the Apple AppStore. The S&P 500 plummeted 1.5%, while the Nasdaq 100 fell 3%. A carefully regarded index of chipmakers has dropped the most since March 2020. Nvidia, the poster child for the AI frenzy, fell 17%, marking the worst market-cap drop for a single stock in market history.

But late Monday, a Nvidia Spokesperson said: “DeepSeek’ work shows how new models can be created. Leveraging widely available models and computers that are fully export control compliant.”

Also, OpenAI announced the launch of ChatGPT for government agencies in the U.S. Subsequently, Wall Street Futures recovered to some extent Tuesday led by Nvidia as Trump’s tech war policy with China may help the US tech sector in the medium term. Also, Boeing helped amid upbeat guidance despite a terrible report card. On Tuesday, Wall Street was also buoyed by a report US Treasury Secretary Bessent pushes for a gradual 2.5% universal US tariffs plan.

Weekly-Technical trading levels: DJ-30, NQ-100, SPX-500, and Gold

Looking ahead, whatever the fundamental narrative, technically Dow Future (CMP: 42200) now has to sustain over 41800-41600 for any recovery and rally to 43000/43350-43800/45000* and 45500 in the coming days; otherwise sustaining below 41600, DJ-30 may further fall to 41200/40600-40400/40000 in the coming days.

Similarly, NQ-100 Future (20990) has to sustain over 20700-21000 for recovery and rally to 21500/21900-22250/222500 and further 22700-23000/23300 in the coming days; otherwise, sustaining below 20700, NQ-100 may further fall to 20500/20300-20100/19250 in the coming days.

Technically, SPX-500 (CMP: 5865), now has to sustain over 5950 for any further recovery/rally to 6025/6050-6150/6200 and 6350/6500 in the coming days; otherwise, sustaining below 5925-5900, SPX-500 may further fall to 5800-5775 and 5700/5600-5475 in the coming days.

Also, technically Gold (CMP: 2690) has to sustain over 2705 for a further rally to 2725 and further 2740/2750-2775/2795 and 2815 in the coming days; otherwise sustaining below 2700-2685 may again fall to 2655/2620-2605/2600 and 2595/2575-2535/2435 in the coming days (depending upon Fed rate cuts, Gaza/Ukraine war trajectory); Gold surged almost 75% in the last one year since Gaza war started back in October’23. Now it may retrace to $2500-2400 levels if Trump indeed can mediate both the Gaza and Ukraine war ceasefire by early 2025.