Unique Research · Insight · May 23, 2024
In November last year, Unique Capital partner Wu Di organized a Silicon Valley study tour, taking nearly 20 AI founders to OpenAI, NVIDIA and other leading AI companies up close (see prior recap: On the ground for Silicon Valley’s AI drama and the night before Gemini). At Unique Research’s 2024 AIGC Application Development Summit, Wu Di sat down with tour companion Kong Rong, Co-Dean of TF Securities’ Global Forward-Looking Industry Research Institute, for a conversation on China–U.S. AI differences from a firsthand view.
Edited transcript below.
Silicon Valley study tour — conversation
Wu Di, Partner, Unique Capital
Dean Kong brings more than a decade of securities research, with a focus on AI. Under her leadership, TF’s institute has ranked in the top three for overseas-market New Fortune research three years running — a strong record.
In early December we traveled together to the U.S. for an AI study tour — Silicon Valley and New York — visiting Meta, NVIDIA, Google, plus local and overseas AI startups. We also visited leading labs, including Rutgers University’s AI and robotics labs (among the top three nationally), and Stanford, where we spoke with members of the student innovation and entrepreneurship group. Dean Kong, please share your impressions of the trip.
Kong Rong, Co-Dean, TF Global Forward-Looking Industry Research Institute
The trip left a deep mark. I had been to the U.S. more than once last year, but going with Unique Capital changed my overall feel. I gained a much deeper read on AI-related companies — both tech firms and relatively traditional ones.
On AI as a whole, two observations stood out. First, Silicon Valley — and the U.S. more broadly — has innovation activity centered on AI. From engineers at big tech to Stanford CS students, almost everyone is building around AI. “In full swing” is no exaggeration for the startup atmosphere; it is intensifying. That is the entrepreneurial-environment read.
Second, the investment lens. After Silicon Valley Bank’s failure, many asked whether capital support for AI was still adequate. In deep conversations with top Silicon Valley VCs and other investors, we found everyone is investing around AI and sees it as one of the most important opportunities of the next decade — not only for AI itself, but for how it reshapes other technology fields.
Apple’s Vision Pro and spatial computing are closely tied to AI. Robots will become more intelligent and able to learn — again, AI. Industry end-markets are also affected: many traditional sectors can raise competitiveness and market share through AI. Those are impressions from founders, industry, and investors.
Wu Di
There is a joke in China’s venture circle lately: founders no longer go “to VC,” and VCs no longer go “to LP” — a snapshot of a muted, pessimistic mood. Our U.S. trip felt very different. About 40% of U.S. venture capital is concentrated in Silicon Valley.
Watching how investors put capital to work, we saw many founders living reasonably well, with a workable life–work balance, and relatively accessible funding. Early deals often use a SAFE structure: investors do not take equity at the first check, but get a discounted entry in the next priced round. That is very different from China’s traditional venture pattern — Silicon Valley’s investing culture is founder-led.
Against that backdrop, how do China’s and the U.S.’s AI startup environments differ today?
Kong Rong
On industry trends, U.S. AI startups have a strong innate edge in B2B, rooted in mature SaaS and B2B infrastructure built over many years.
Attention to AI hardware is also rising this year. From the latest CES, AI is not limited to PCs — on-device phone AI and AI wearables drew heavy interest.
One AI-hardware founder on our trip noted that Chinese companies may hold distinctive advantages here — especially supply chain and cost control. So even where China–U.S. AI paths diverge, China has real openings in hardware and related niches.
Wu Di
On hardware: I am involved in an AI-hardware project focused on companionship. China does have strengths in hardware R&D and supply chain. This year, AI × hardware has become a trend you cannot ignore — rabbit is a recent high-profile example.
Dean Kong is right that we need not undersell ourselves. In many verticals — digital humans, for example — China’s AI is not far behind the U.S. on R&D, operations, or commercial maturity. Domestic markets are not weaker than overseas ones.
I also see that this AI wave often runs with very small teams. Silicon Valley’s talent density is high — almost anyone you pull aside is a top-school PhD — but China is not weak on commercialization, market size, or application. Those are my observations.
Kong Rong
Many eyes are on GPT-5 in 2024. It may improve understanding — including video understanding and video generation. What new application scenes open after GPT-5, from now through next year, is worth watching.
In recent interviews, Sam Altman has framed AGI as the core change over the next 25 years. Many once thought AGI needed a decade, or doubted it outright. With compute and performance rising and foundation models evolving, some now expect AGI in 3–5 years; another camp puts roughly a 50% chance of AGI by 2028.
On a medium horizon, model iteration and AI progress are likely to keep accelerating — not hit a wall after GPT-5. Beyond GPT-5, people are watching AGI itself. 2025 may bring milestone breakthroughs; 2024 can be seen as the first year of the AGI era.
Wu Di
On the Silicon Valley tour, the common view was AGI in about three years — roughly a 3–5 year window. Meta’s Mark Zuckerberg has stressed this recently as well: AGI expected to reach 1 billion users and eventually open source, with compute prepared — GPU count already at 600,000 — for an “AI for everyone” goal.
What does the path to AGI look like?
Kong Rong
We are likely in a mid-stage on the road to AGI — a medium- to long-term focus. Better AGI and AI will unlock more applications and real deployment. Some also worry about risks to humanity along the way.
So “superalignment” has become a central topic for core labs and for figures such as Ilya and that school of thought. Beyond application landing, we should also look further out. In 2024–2025, maturity and adoption of AI video tools will matter more.
From a 2024 vantage: image generation surged over prior years; 2023 saw rapid music generation; in 2024 video tools will move even faster — not only animation. As consistency and reliability improve, real-person video and clip length should rise, and product iteration will speed up. From a secondary-creation angle, we are constructive on AI video tools’ application outlook.
I am personally constructive on AI social products. North America’s DJ AI drew heavy Twitter attention recently — not only for anime/Avatar-style interaction, but for very realistic audio and music. AGI is a medium- to long-term goal; we also watch products that can land in 2024–2025, including video and social.
B2B also has high certainty — a point we heard from top U.S. VCs. Many hope for a consumer breakout app, but those are partly random in timing. In industry deployment, combining user data and concrete scenes yields more deterministic opportunities — scenes that may not be novel, but with clearer commercial value. For this year and next, that is worth sharing.
Wu Di
How do you see multimodal?
Kong Rong
Late last year, fragments about GPT-5 multimodal capability circulated on Twitter; Sam Altman later clarified. He did speak to multimodal tech, which may show more potential in a future GPT-5-class model.
From this year on, multimodal momentum should keep strengthening and drive progress for several years. The key watch-item is multimodal fused with real products — how enhanced video lands in products, and how product design blends AI agents with multimodal. Those developments deserve close attention.
Overseas study-tour program materials — CN imagery attach on ingest (1 image).