By Michael Spilg

The SCSP AI Expo for National Competitiveness was held in Washington DC in May 2024. There was a ton of public-private sector innovation on display with a hyper-urgent, enthusiastic demand for speed communicated by high-level presenters, including Senate Majority leader Chuck Schumer, Secretary of Commerce Gina Raimondo, Palantir CEO Alex Karp, and former Google CEO Eric Schmidt. Schumer emphasized how innovation is the North Star for the country’s AI policy roadmap, which is expected to be unveiled later this summer. Raimondo further elaborated on the need for innovation, but that it needs to be balanced with safety. However, she acknowledged that there would be significant delays if regulatory guardrails were applied too heavily and too early in the “early innings” of the AI race.

Nvidia remains the undisputed leader of AI. At the SIEPR Economic Summit (March 2024), CEO Jensen Huang predicted that “AI is going to be everywhere.” During the Q125 Nvidia earnings call (May 2024), Huang declared “The next industrial revolution has begun.” Huang announced at Computex in Taiwan that the company is ramping up their development plans to release new chip systems from every two years to every year (June 2024). This is a major feat that illustrates how Nvidia plans to stay competitive and continue to be relevant in the future. During the last earnings call, it was announced the next version of the Hooper 200 GPUs would be shipped later this quarter. The demand for the Blackwell platform is expected to be double that of Hooper at launch, with shipments planned for Q2 and ramping up in Q325 (Lily Investment Research). Customers are expected to already have the Blackwell stood up in data centers by Q425 (Noah’s Arc Capital Management). The company’s next release will be Blackwell Ultra and the next generation chip, Rubin, will be available in 2026 (Huang, Computex).

Improved Performance of Blackwell Relative to Other GPUs (Nvidia GTC Conference)

In The Economist, Semiconductors - Think Different (May 2024), Cerebras CEO Andrew Feldman, addresses the main challenges with GPU performance issues, which lies in the data transfer between the GPU and memory chips. In addition to latency, another issue revolves around maximizing the performance of the GPU. Currently, chips queue data that needs to be processed. A better design would be to incorporate parallel processing, or the ability to process multiple workloads at the same time. Similarly, there are cases where the chip remains ideal since it is specialized to process specific tasks. A solution to this issue would be to design multimodality into the GPU so it can process different types of data to complete different tasks.

Nvidia has a strong lead in GPU chip development, around 70-95% of the AI chip space according to Mizuho Securities (CNBC), but with such a strong demand for AI chip production, other companies are trying to carve out their own share of the market and develop their own chips. Another American startup similar to Cerebras, Groq is attempting to reduce latency by specializing their in-house chips as LPMs (language processing units). MatX is also building chips that remove GPU circuitry unnecessary to LLM training to maximize speed and compute efficiency. Other startups, Hailo (Israel), Taalas (Canada), Tenstorrent (US), and Graphcore (UK) are also designing specialized AI chips.

A main challenge for developers is to design chips to perform the necessary specialities. Since chips can take two to three years to develop and reach production, the function of the chip may not be relevant by the time it is produced and reaches the market since the development of AI products continues to exponentially develop. Big tech companies, such as Google and Microsoft, customers of Nvidia, have also been investing heavily in AI start-ups like Anthropic (Claude) and OpenAI (ChatGPT and Dall-E) respectively, and have developed their own AI models like Meta (Llama3) and Google (Gemini). This has incentivized big tech companies to also invest in building their own AI chips like Google’s TPUs (Tensor Processing Units) and OpenAI. Other industry leaders, like AMD and Intel are struggling to maintain their current market share in the chip space as Nvidia continues to dominate.

At a panel discussion on the end of the first day of the SCSP AI Expo, Felipe Million from OpenAI referenced Amara’s Law to describe the hype around AI. Developed by futurologist Roy Amara, it claims that the potential impact of new technology is overestimated in the short-term, but underestimated in the long-term. At the current rates, we may see sooner than later how significant and impactful the development and integration of artificial intelligence will be across all industries and society. It’s an exciting time to witness the beginning of the next industrial revolution.

Michael Spilg has degrees in Mechanical Engineering from Colorado State University and International Relations from, the University of Illinois at Urbana-Champaign. He currently works as a Senior Systems Engineer in the medical device industry. His interests are in technology, geopolitics, and diplomacy.

References

Feiner, L. (2024, June 2). Nvidia dominates the AI chip market, but there's rising competition. CNBC.www.cnbc.com/2024/06/02/nvidia-dominates-the-ai-chip-market-but-theres-rising-competition-.html

Huang, J. (2024, March 7). Stanford SIEPR Economic Summit. Stanford Institute for Economic Policy Research, Stanford, CA. www.youtube.com/watch?v=cEg8cOx7UZk

Huang, J. (2024, March 20). Nvidia GTC Conference. San Jose, CA. www.youtube.com/watch?v=Y2F8yisiS6E&list=PLZHnYvH1qtOYPPHRaHf9yPQkIcGpIUpdL

Huang, J. (2024, May 22). Nvidia 1st Quarter FY25 Financial Results, Santa Clara, CA. www.investor.nvidia.com/events-and-presentations/events-and-presentations/event-details/2024/NVIDIA-1st-Quarter-FY25-Financial-Results--2024-UOmEun3udV/default.aspx

Huang, J. (2024, June 2). Computex. National Taiwan University, Taipei, Taiwan. www.youtube.com/watch?v=pKXDVsWZmUU

Lily Investment Research. (2024, May 28). These 3 Things Should Push Nvidia Even Higher. Seeking Alpha.

Noah’s Arc Capital Management. (2024, May 28). Nvidia Q1: Chips Are Too Expensive. Seeking Alpha.

Raimondo, G. (2024, May 7). SCSP AI Expo for National Competitiveness. Washington DC, USA.

Schumer C. (2024, May 8). SCSP AI Expo for National Competitiveness. Washington DC, USA.

The Economist. (2024, May 25). Semiconductors - Think Different.

Additional Resources:

Eason, L. (2024, May 31). Nvidia Prospects: From An Engineer POV. Seeking Alpha.

Chevallier, A., Dalsace, F., & Barsoux, J.-L. (2024). The art of asking smarter questions. Harvard Business Review.

Kaku, M. (2023). Quantum Supremacy: How the Quantum Computer Revolution Will Change Everything. Doubleday.

Loeffler, J. (2024, January 15). What is an NPU: The new AI chips explained. TechRadar. https://www.techradar.com/computing/cpu/what-is-an-npu

Miller, C. (2022). Chip War: The Fight for the World's Most Critical Technology. Scribner.

The Asian Investor. (2024, May 24). Nvidia: Profit Explosion And Stock Split Are Game-Changers. Seeking Alpha.

Zourmapanos, Y. (2024, May 28). Nvidia: The Next Bull Run Begins. Seeking Alpha.