"[Is it the same level as Nvidia?] No, but it's pretty close and it's getting better every generation." — Naveen Rao (Discussing Huawei’s Ascend line) [14:14]
"It forced the DeepSeek folks to think very creatively and write a lot of low-level optimized software to make it happen... they had to be efficient because they only had so many chips available." — Naveen Rao (On how US sanctions spurred architectural innovation) [18:02]
Disclaimer: Orignal content owned by or sourced from third parties. It does not represent the views of 'Nuggets' platform or it's team. AI is used extensively across this platform including for summaries. Accuracy is not guaranteed, there can be mistakes. Any info or content on this platform is not a financial, legal, or investment advice. Do your own research. Refer for complete disclosures:- Terms of Use · Full Disclaimer
"China's growth in electricity is tremendous... they're basically ahead of us by two to 3x on the amount of power that they're building." — Narrator (Highlighting the divergence in energy infrastructure) [08:55]
"The risk isn't that China wins at the cutting edge; it's that over time it proliferates everywhere else." — Narrator (On China's global "turnkey" AI export strategy) [12:08]
"A 'DeepSeek hardware moment' would be if China came out with hardware that was vastly cheaper to buy and operate and gave you 90% of the performance." — Naveen Rao (Defining the potential tipping point) [31:51]
Executive Summary
This video explores China's strategic pivot from AI software development to the indigenization of AI hardware, a move catalyzed by US export restrictions. By leveraging state incentives, a massive domestic energy build-out, and creative engineering workarounds like multi-patterning, Chinese firms are closing the performance gap with Nvidia. The core thesis suggests that while the US leads in "frontier" technology, China is positioned to dominate the global "middle market" by exporting a turnkey AI stack—including models, software, and chips—to countries with lower budgets and fewer constraints.
Key Takeaways
Hardware Indigenization: China is moving beyond the DeepSeek software moment to a hardware-focused "Four Dragons" era, aiming to replace Nvidia with domestic alternatives like Huawei and More Threads [02:04].
Efficiency Through Scarcity: Limited access to Nvidia chips forced Chinese labs to develop highly optimized software and more efficient training paradigms [05:32].
Brute-Force Engineering: Lacking access to EUV lithography, China uses multi-patterning to achieve 5nm performance, sacrificing yield and cost for domestic self-reliance [16:42].
Energy as a Bottleneck: AI progress is tied to power; China's central planning allows for a rapid expansion of energy grids that currently outpaces the US by 2x to 3x [09:07].
The Global "Prius" Strategy: China is targeting the global market with "good enough" AI—the Prius to the US's Ferrari—which is more economical for emerging economies [30:42].
Unconventional Computing: The next race in AI hardware may be in non-digital or analog computing to overcome the physical limits of Moore’s Law [34:23].
Detailed Summary by Topic
The "Four Dragons" and the New IPO Wave
China's IPO market is seeing a revival driven by four primary chip firms: More Threads, MetaX, Byron, and Enflame. These companies, along with the established champion Huawei, are being bankrolled by the CCP to build a homegrown Nvidia-style ecosystem. More Threads, founded by a former Nvidia executive, saw its valuation jump 400% after its debut, while Enflame (backed by Tencent) is valued at approximately $3 billion. This surge is supported by a reported $70 billion in new government incentives [02:04].
Engineering Around Sanctions: The Brute Force Strategy
Since ASML cannot deliver EUV machines to China, SMIC is using a technique called multi-patterning. This involves applying masks multiple times with higher wavelengths of light to build 5 nanometer chips. While this increases complexity and decreases yields compared to TSMC's process, it allows China to maintain a presence near the leading edge. Additionally, Chinese labs "brute force" AI quality by stringing together thousands of slightly less efficient HuaweiAscend chips to match the performance of fewer NvidiaH100s [16:42].
Energy Infrastructure and Central Planning
Power is the ultimate bottleneck for scaling compute. Naveen Rao notes that while the US leads in total data center capacity, China's energy growth trend is a "straight line up," outpacing the US by 8x over the last 20 years. Central planning allows the CCP to redirect power from factories to AI data centers, a level of macro-optimization difficult to achieve in the West due to public pushback on electric bills and water usage [08:55].
Global Market Mind Share and the "Turnkey" System
China is not just selling chips; they are exporting a "turnkey" system that includes the model, service, and hardware in a single deployment. This strategy, described as a "Trojan Horse," locks in long-term dependence in markets across Africa, Russia, and even parts of Europe. Many global companies are opting for Chinese open-source models like DeepSeek because they are "good enough" for business tasks like document summarization [10:31].
The Future: Unconventional ComputingNaveen Rao's new venture, Unconventional AI, is focusing on rethinking computer foundations to solve the energy crisis. Digital computation is reaching its limits as Moore's Law slows down. The next phase of competition involves using the intrinsic physics of semiconductors to create systems that are 10x to 100x more efficient. Research papers from Chinese academia suggest they are heavily invested in this "unconventional" path as well [32:43].
Data & Figures
Data Point
Value
Context
CCP Chip Incentives
$70 billion
Reported incentives prepared by the CCP as of Dec 2025 [04:15]
Existing Chip Fund
$50 billion
Pre-existing fund dedicated to domestic semiconductors [04:21]
Enflame Valuation
$3 billion
Reported valuation for the Tencent-backed chip challenger [02:44]
Nvidia Production
4-5 million
Projected data center chips Nvidia will produce this year [15:03]
Huawei Production
200,000
Estimated production scale for Huawei’s chips []
Stories & Anecdotes
The Costco Strategy:Aaron Jen compares China’s approach to shopping at Costco. You don't get the "frontier" tech, but you get "good enough" technology one or two generations behind, available in massive bulk to compensate for individual inefficiency [07:46].
The Prius vs. Ferrari Analogy: The market is bifurcating. The US builds the "Ferrari" (bleeding-edge), while the rest of the world has an appetite for the "Prius" (economical and efficient for daily tasks). China is successfully capturing the "Prius" market [29:00].
The DeepSeek Wake-up Call:DeepSeek proved that chip limits merely change strategy. Because they couldn't "buy more chips," they innovated at the software level, leading to a model that surprised Western leaders with its efficiency [05:32].
References & Recommendations
Companies & Tools:
Unconventional AI - Naveen Rao’s startup focusing on energy-efficient AI hardware [12:33].
More Threads - Known as "China's Nvidia," founded by former Nvidia execs [02:20].
SMIC (Smick) - The primary Chinese foundry competing with TSMC [14:09].
DeepSeek - The open-source model that upended the landscape with its cost-efficiency [06:31].
CUDA - Nvidia’s software moat that Chinese firms are attempting to bypass [32:24].
People Referenced:
Naveen Rao (Founder of Unconventional AI) - Expert on hardware-software co-design [12:40].
Aaron Jen (Founder of Hydro Host) - Specialist in AI data center services [07:46].
Jensen Huang - CEO of Nvidia, mentioned regarding his desire to stay in the Chinese market [24:26].
Speakers & Credentials
Deirdre Bosa: Host of The Take on CNBC, specializing in technology and the AI industry.
Naveen Rao: Guest and expert. Founder of Unconventional AI, formerly VP of AI at Intel, and key executive at Data Bricks [12:40].
Actionable Next Steps
Monitor "Unconventional Computing" Benchmarks: Keep a close eye on research readouts from Unconventional AI and Chinese academia regarding analog computing [34:41].
Evaluate Turnkey AI Stacks: Assess the cost-benefit of Chinese "turnkey" AI systems compared to fragmented Western solutions [10:44].
Track Energy Policy: Watch for US legislative movements regarding data center power permissions vs. China's central planning [09:32].
Prioritize Software Optimization: Follow the DeepSeek methodology of "doing more with less" rather than relying solely on brute-force compute [18:02].
Jul 13, 2026
Yanis Varoufakis | Closing Keynote | Thursday 18th June 2026 | Web3 Foundation
"Politics is who does what to whom... who has the power to do to make you do stuff." Yanis Varoufakis 00:02:36 https://youtu.be/WZeuKyUs9hM?t=2m36s "We have created machines and machinery—network machines—that are not produced means of pro…