Digital AI Closes, Physical AI Opens

Jeong Ji-hoon, General Partner at Asia2G, diagnosed that the digital AI flywheel is entering its final stages with Anthropic. In contrast, physical AI holds significant growth potential for the future, with new devices such as autonomous driving, humanoids, and robots expected to drive large-scale memory demand. This shift is projected to bring a new peak to the memory market around 2028.

Partner Jeong Ji-hoon emphasized the necessity of understanding the industry's direction from a longer perspective and investing accordingly, rather than predicting immediate stock price fluctuations. He stated that identifying technological bottlenecks is crucial, citing the example of predicting Intel and memory semiconductor shortages two years ago. He explained that such insights allow investors to seize opportunities early when signs appear.

Venture capital investment focuses on long-term growth rather than short-term stock fluctuations. Partner Jeong Ji-hoon stated that investing in early-stage startups typically involves a long-term commitment of 5 to 10 years. Companies in the pre-IPO stage are held for their growth potential, and stock trading is often restricted during this period.

Partner Jeong Ji-hoon explained that the flywheel model prioritizes growth through reinvestment over profit distribution. Amazon succeeded with a strategy of continuously reinvesting profits to enhance customer experience, expand its product range, and maintain thin margins. This contrasts with traditional investment evaluation methods centered on PER and PBR, focusing instead on increasing corporate value through continuous growth.

After winning platform competition, full-scale monetization begins, and invested capital must convert into service revenue and net profit. He noted that once capital is injected and the flywheel is complete, doubts about the industry disappear, enabling sustained growth.

With the launch of ChatGPT in November 2022, the AI flywheel began, and currently, it is divided into two flywheels: one centered on digital AI and another on physical AI, explained Partner Jeong Ji-hoon. The digital AI flywheel has entered its completion phase through a hyperscale loop, and after NVIDIA's GPU bottleneck, the market's focus is shifting to inference and service infrastructure.

Anthropic, despite being a latecomer, entered the market with a different strategy than OpenAI. Instead of expanding into the mass B2C market, it directly targeted the enterprise market and developed dedicated AI services. This allowed it to avoid being pushed out by scale competition and focus on profitable B2B efficiency, securing high profit margins compared to token prices and achieving profitability. This strategy demonstrates the importance of understanding enterprise customers' characteristics and providing tailored solutions.

Anthropic's Claude service, despite costing five times more than OpenAI for similar tasks, is used by many users. Partner Jeong Ji-hoon analyzed that this is because users recognize the premium value derived from actual performance and productivity improvements. Companies reinvest these profits to secure infrastructure and improve services, creating a continuous virtuous cycle of growth.

Partner Jeong Ji-hoon used McDonald's and In-N-Out Burger as an analogy to explain corporate strategy. While In-N-Out Burger focuses on quality control with a localized growth strategy, McDonald's leveraged its global real estate infrastructure to become the world's largest hamburger operator. This illustrates the strategic choice companies face between focusing on maintaining quality and expanding market dominance through economies of scale.

Partner Jeong Ji-hoon pointed out that there are areas difficult to predict with general technical analysis alone, emphasizing the importance of understanding the industrial cycle structure. He explained that maximizing profits can be achieved through cycle investing, where investments are made when momentum events are measured and evidence is confirmed. This suggests that the ability to read market trends and signs of change plays a key role in investment success.

The digital AI market is already dominated by global players. In this situation, Partner Jeong Ji-hoon advised focusing on infrastructure technologies that critically contribute to overall performance improvement, i.e., companies that have been relatively less recognized. Such technologies play a crucial role behind the scenes and can generate significant value with market growth.

Meta replaced only its network switches with Broadcom's optical switch equipment while maintaining its existing NVIDIA GPU infrastructure. Partner Jeong Ji-hoon stated that this replacement alone yielded benchmarking results showing an efficiency improvement of almost 40%. This maximized GPU utilization, resulting in a performance improvement akin to adding 40% more GPUs. Thus, resolving bottlenecks through infrastructure improvements can generate greater economic value than simply increasing the number of GPUs.

Partner Jeong Ji-hoon predicted that in terms of future value, the physical AI loop definitely holds higher potential. While digital AI already has players and market themes largely formed, and revenue is starting to materialize, physical AI is the future yet to come, possessing greater potential. He advised preparing for the shift towards physical AI after digital AI.

The current memory semiconductor market faces concerns about a peak after next year due to digital AI optimization. However, Partner Jeong Ji-hoon predicted that transitioning to physical AI will increase image and video data, requiring significantly more memory. He explained that if physical AI progresses faster than expected, generating immense memory demand from autonomous vehicles, the memory market could experience continuous growth 'without valleys,' unlike traditional cycles. This suggests that a double-loop structure could become a new growth engine for the memory market.

Partner Jeong Ji-hoon saw the power sector as the most promising in terms of consistent growth potential. This is because the explosive increase in AI data centers is causing persistent shortages in almost all areas related to power and energy. In particular, the demand for ESS (Energy Storage System) batteries is bound to continue increasing, and in the long term, the hydrogen energy sector is also expected to improve significantly. He stated that he is investing heavily in hydrogen energy from a long-term investment perspective of over 5 years. This implies that it is time to focus on policy changes and optimization technologies for energy security.

Partner Jeong Ji-hoon pointed out that Solaseado in Haenam, Jeollanam-do, has large-scale solar power infrastructure but lacks places to consume the generated electricity. He also explained that Gangwon Province only utilizes 10% of its generation capacity, and transmission network expansion is essential due to geographical factors like the Taebaek Mountains. These cases demonstrate that even with large-scale power generation facilities, they can be rendered useless if not supported by infrastructure that efficiently supplies and consumes the power.

Regarding some Wall Street AI investment bubble theories and cross-trading criticisms, Partner Jeong Ji-hoon interpreted AI infrastructure investment as a 'token factory' business model that produces tokens. He explained that data center construction can be approached from the perspective of securing profitability similar to residential facility sales, arguing that the AI infrastructure business can be seen as a real asset-backed business with analog collateral bonds, like SpaceX's successful rental income case. This perspective suggests that the AI investment boom is not merely a bubble but a sustainable, real asset-based business.

He explained that if they fail to grasp core business rights during this period, they will eventually have to yield their throne, hence their desperate investments. While it might appear as investment overheating, he presented it as an essential survival strategy for securing long-term market leadership.