On May 28th, Anthropic, the developer of the AI model Claude, announced the completion of a $650 billion Series H financing round, with a post-money valuation reaching $965 billion, surpassing its competitor OpenAI ($852 billion) to become the world's highest-valued private AI company, once again highlighting the global capital frenzy for AI.
As trillion-dollar giants engage in close combat over the computing power infrastructure, what opportunities remain for ordinary startup teams at the application layer? What direction will the actual division of labor in the AI industry between China and the US take? With these questions, Jenny Yang, Founder and CEO of Starlabs Consulting, recently interviewed Steve Hoffman (Steven S. Hoffman), Founder and CEO of the top US startup incubator Founder Space, known as the 'Startup Guru' of Silicon Valley.
Hoffman is a serial entrepreneur and venture capitalist, as well as a bestselling author of several highly acclaimed works such as 'Make Elephants Fly,' 'Survival Rules,' and 'The Five Forces of Innovation.' He is also a globally sought-after keynote speaker and has long served as a strategic advisor to governments, well-known companies, and incubators worldwide.
As a seasoned venture capitalist who has deeply mentored thousands of startups globally, Hoffman offers an exceptionally calm, candid, and far-sighted business deconstruction of the current AI craze.
The following are excerpts from Jenny Yang's interview with Hoffman:
The True Inflection Point for Autonomous Agents Could Arrive Within 2 Years
Jenny Yang: You just finished a trip to China. Please share your overall impression of China's AI technology, AI companies, and the current state of AI applications. What differentiated roles do you think Silicon Valley and China will play in the next phase of AI competition?
Hoffman: My overall impression is that China is moving forward at an incredible pace, extremely rapidly. The Chinese startups I spoke with are integrating AI into every aspect: payments, logistics, customer service, human resources, marketing, sales, procurement, manufacturing, and so on.
At the same time, I believe Silicon Valley will continue to lead fundamental research in frontier large models. The concentration of computing power, top-tier talent, and capital in the US remains unparalleled for now. However, China will excel in application and commercialization. Chinese companies are exceptionally adept at scaling a technology at a breathtaking speed and transforming it into commercial products with real users and real scenarios. This pragmatic attitude and efficient execution are precisely China's strengths.
China also possesses formidable top-tier AI labs, including Moonshot AI, Alibaba, ByteDance, and DeepSeek. These labs will play the role of extremely sharp 'fast followers,' closely tracking their American counterparts. While their capital may not be exceptionally abundant, they always find innovative ways to push costs to the extreme, thereby driving the global expansion of their platforms.
Furthermore, China holds absolute dominance in the robotics field. Globally, there is no other place that simultaneously possesses such a complete supply chain, infrastructure, and talent pool to support the mass production of robots. The next stage of the AI race is not winner-takes-all. Silicon Valley will continue to build the most powerful technology engines, while China will construct the best commercial ecosystems and robotics hardware. Both are equally important.
Jenny Yang: Do you think AI has borders? Against the backdrop of increasingly stringent global data sovereignty and AI regulatory policies, are you more optimistic about companies deeply focused on their home markets or those that are 'Global from Day 1'?
Hoffman: Technologically speaking, AI has no borders; but in reality, global regulatory policies are rapidly drawing boundaries. Data sovereignty laws, national security reviews, model export restrictions... these are reshaping the global compliance framework.
Some founders, seeing this trend, conclude that they should focus deeply on a single home market. I understand this logic, but I absolutely do not agree with it.
I firmly support 'Global from Day 1' for a simple reason: companies that aim to establish themselves locally first and expand overseas later almost invariably run into trouble. Because distribution channels differ between countries, compliance requirements vary, and even brand positioning needs to be rebuilt from scratch, which is not only costly but also slow.
Global-first companies, from day one, build modular and highly adaptable systems. They architect their foundations to address regulatory differences upfront, rather than patching problems later. They can attract international teams that understand diverse markets, translating into lasting structural advantages.
Admittedly, compliance is becoming more difficult, and companies need to incorporate localized compliance systems. But the key to breaking through is building flexible architectures, not settling for a corner. Market opportunities are global, and so should be the ambition of every tech entrepreneur.
Jenny Yang: You've pointed out that we are still in the very early stages of the AI revolution, and the explosion of autonomous agents will completely颠覆现有商业范式 (overturn the existing business paradigm). From your observations, how far are we from that day? Facing the challenge of structural unemployment triggered by AI, what preparations can we make in business models or systems?
Hoffman: That day is close—closer than most people think, but further away than media hype suggests. Autonomous agents capable of handling independent, clearly defined specific tasks have already emerged, such as automated customer service, code review, data analysis, research summarization—these are no longer demos but are already commercially deployed.
The true inflection point—where different agents can self-coordinate, handle ambiguous multi-step goals, and operate across systems unattended—is probably 2 to 4 years away, maybe even sooner.
When that wave truly hits, labor replacement will be stark and real, not alarmist talk.
The solution is definitely not to slow down AI's pace, but to ensure social mechanisms keep up with the speed of AI technological iteration.
- On the business model side: The smartest founders are designing companies around 'Human-AI Collaboration' rather than 'pure automation.' Their models have humans responsible for judgment, creative output, and accountability, while agents handle workload and efficiency. This model is more resilient and better for team development.
- On the policy side: We need to honestly confront issues like vocational retraining, social security systems, and educational reform. This time, it's not just low-skill jobs being replaced, but lawyers, analysts, copywriters, consultants, and nearly all knowledge-intensive positions. This fundamentally changes the underlying logic of social governance.
Jenny Yang: You've noted that traditional 'Humans as a Service' (HaaS) businesses like consulting and brokerage, due to high marginal costs, struggle to achieve true scale. Now, AI is massively replacing and automating professional intellectual services. Does this mean AI-driven knowledge services will break the curse of HaaS businesses being difficult to scale?
Hoffman: Traditional consulting has always faced a dilemma: to grow, you must add people; adding people increases costs, compressing profit margins and stalling scale. This is the inherent trap of the HaaS model.
But AI fundamentally changes this underlying equation. Today, a senior consultant fully equipped with AI agents can provide analysis that previously required a small team, meaning the marginal cost of adding a new client plummets. This is unprecedented.
So yes, AI-driven knowledge services finally have the potential to break the scalability curse. But the prerequisite is that companies are willing to restructure their organizations accordingly. Future companies that thrive in this transformation won't just treat AI as an efficiency tool but will completely reshape their entire business system around the AI foundational base.
Startups Should Focus on Scenario Innovation
Jenny Yang: Regarding open source vs. closed source, from Founder Space and a venture capital perspective, are you more inclined to support applications deeply tied to the closed ecosystems of giants, or independent projects built on open-source ecosystems? Why?
Hoffman: In the US, I favor applications built on the ecosystems of major cloud providers (including AWS, Azure, and Google Cloud). These platforms offer mature distribution channels, enterprise-grade trust, and deep integration capabilities necessary for scaling. Developing on these large platforms inherits many native advantages: security compliance, stability commitments, and global infrastructure support. Open source is exciting, but 'excitement' doesn't win enterprise deals.
But China is different. The cloud ecosystem there is primarily shaped by Alibaba Cloud, Tencent Cloud, and Huawei Cloud. The policy and regulatory environment dictates which platforms companies can choose. In China, open-source models like DeepSeek are gaining significant market traction because they allow Chinese companies to run autonomously without relying on external overseas infrastructure. In this context, open source is not just a philosophy but a strategic necessity.
Therefore, the correct answer entirely depends on where you are building your product and who you are selling it to.
Jenny Yang: With computing power and algorithms monopolized by giants, how can early-stage AI startup teams effectively identify and capture demand pain points that have genuine scalable commercial potential and are not easily crushed by giants?
Hoffman: Tech giants will undoubtedly commoditize general-purpose foundational technologies. If what your startup does is something OpenAI, Anthropic, Google, or Microsoft could launch as a new feature within six months, it's not a business—it's just a feature on their product roadmap.
To survive in such a fiercely competitive environment, startups must focus on niche, specialized, and deeply contextual domains. For example: a workflow requiring sharp understanding of a specific industry, a compliance solution relying on specialized expertise that foundational models lack, or a customer relationship that takes years to build trust.
Vertical depth in a niche is the startup's defensive moat. The more a solution relies on the hands-on experience of industry experts (surgeons, supply chain managers, insurance actuaries, etc.), the harder it is for industry giants to replicate quickly.
Ultimately, speed is the most important moat for an early-stage company. Your iteration speed must exceed the pace at which giants can internally approve competing projects and budgets. By the time those giants react, agile startups have already built their brand and solidified market leadership—meaning you have a growing user base, proprietary data, and a mature product that truly fits the market.
Jenny Yang: With the development of generative AI, AI forgery and fraudulent information are proliferating. From the perspective of cybersecurity and anti-AI fraud, does this present a highly promising赛道 (sector/field) for entrepreneurs?
Hoffman: Yes. Today, creating synthetic media has no barrier to entry. Voice cloning, deepfake videos, realistic AI phishing emails, etc., are becoming an increasingly severe nightmare.
The defense mechanisms in the cybersecurity industry lag significantly behind attack methods. This pain point is a market opportunity. Detection tools, traceability verification, digital watermarking, identity authentication—all these areas hold tremendous entrepreneurial potential. Enterprises and government agencies need such solutions, and the financial industry especially does, as they suffer monetary losses from various AI fraud activities.
But it's important to note that detection models can only defend against known attack types. Therefore, from the start, startups must confront this adversarial nature and design products with continuous learning and dynamic iteration capabilities.
If a startup team has deep expertise in both generative AI and cybersecurity, they have the opportunity to build a multi-billion-dollar company addressing the industry-wide problem of rampant deepfake technology.
Web3 + AI Could Be a Trap
Jenny Yang: In today's AI era filled with technological anxiety and capital狂热 (frenzy), what fundamental mindsets, different from before, do you believe a founder leading a team to build the next unicorn must possess?
Hoffman: Forget everything you thought you knew about 'moats.' In the current industry environment, a product from 18 months ago might already be obsolete. Entrepreneurs who make it to the end already recognize this.
First, replace feature thinking with systems thinking. The next unicorn will not be built around a clever prompt. It will be built on a network of agents, data flywheels, and multi-party integration systems, growing through long-term compound effects.
Second, stay tightly focused on real user needs. AI significantly increases development efficiency but can also easily lead products astray from practical applications into self-indulgence. Excellent entrepreneurs always focus on the core user needs. Deviating from direction with blind iteration ultimately becomes internal friction.
Third, recruit highly adaptable talent. Skills hot today may be obsolete in two years. Companies need to build continuous learning teams, not just execution teams.
Fourth, do not fear the technology. Many entrepreneurs view AI as an inscrutable black box. You must understand it well enough to accurately know what it can and cannot do. This understanding itself is your competitive advantage.
Jenny Yang: In the past, you mentioned that blockchain has been overhyped for many enterprise applications beyond cryptocurrency, while AI is the truly universal foundation touching all industries. Today, many Web3 companies are trying to integrate AI with Web3. Do you think 'Web3 + AI' is a promising entrepreneurial direction?
Hoffman: I'll be blunt: Web3 has real value, but primarily for those already in the crypto space. Applications like decentralized finance, asset tokenization, and cross-border settlement without intermediaries are significant for that specific group. But this group represents a small fraction of the global economy.
For ordinary business clients, small and medium merchants, and the general public, it's different. I don't believe Web3 can drive substantive change in the mainstream market. I've never been an advocate, and developments over the past few years haven't changed my view.
Most consumers and businesses simply don't need blockchain to achieve their commercial goals. They need reliable products, excellent user experience, and reasonable prices. Web3 adds friction, increases complexity, and brings regulatory risks. For ordinary consumers and users, Web3 doesn't deliver what they actually need.
In contrast, AI is the true universal underlying technology. It can reach every industry. Almost every industry can use pattern recognition, automation, and intelligent decision-making to solve real problems. This is a fundamentally different value proposition.
Forcibly combining Web3 and AI doesn't multiply their value; it just adds complexity. For most founders, this is not an opportunity but a trap. Of course, AI might help those already deeply embedded in the Web3 ecosystem, but for the broader mass market, it won't materially change user adoption rates or the industry's development trajectory.
Jenny Yang: We noted that you announced an ambitious non-profit plan in early 2026—to establish research centers at 10 universities worldwide aimed at training future leaders on how to make AI reflect core human values. Could you share the current progress of this plan? What principles of 'responsible innovation' do you hope to convey to future AI entrepreneurs through these centers?
Hoffman: Our vision is to establish dedicated research centers at ten universities globally. We are still a considerable distance from that goal.
We are currently in the very early stages, with most of our effort focused on fundraising. Because before we can truly execute, we must ensure we have the necessary resources. Building substantive, sustainable programs within universities requires real financial investment.
What drives us forward is a simple belief: Every young person entering the workforce today will spend their entire career in a world where AI is integrated into every product, every service, every business. However, most of them are not prepared for this monumental shift. Our research centers aim to change that.
We want the next generation of entrepreneurs not only to know how to use AI to build products but also to know how to build AI products that align with human values; to learn to anticipate the various secondary and derivative impacts of technology deployment; and to pursue responsible technological innovation while maintaining ambition.
That is our mission, and we are moving towards that goal.
