Nov 17, 2025 · 1h 6m · 20vc
AI Fund’s GP, Andrew Ng: LLMs as the Next Geopolitical Weapon & Do Margins Still Matter in AI? · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this deep-dive interview, AI pioneer Andrew Ng discusses the critical physical bottlenecks of AI scaling, the geopolitical soft-power struggle between the US and China, and the economic realities of building defensible, high-margin AI applications. Ng challenges common industry narratives around AI safety, AGI timelines, and the future of software engineering, advocating for a pragmatic focus on workflow restructuring and education.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 20.2% of the talking time here. How this is scored →
Harry as informed peer 5.9 Guest teaching 3.9 Guest disagreement 2.4 Harry pushing back 4.4
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 36:21 Andrew explicitly rejects Karpathy's agent timeline
Andrew directly contradicts Andrej Karpathy's widely cited claim that useful agents are a decade away, declaring that effective agentic workflows are already running across AI Fund's portfolio today.
Hardest push from Harry ▶ 32:48 Harry confronts Andrew with Replit and Lovable pass-through economicsHarry refuses Andrew's optimistic framing of application layer investing by presenting concrete unit economic metrics, citing how application startups pay up to 80% of their revenue straight to model providers like Anthropic.
Biggest teaching moment ▶ 48:59 Andrew reframes AI value creation beyond labor substitutionAndrew reframes Harry's core venture investment thesis that AI ROI requires replacing human labor budgets, demonstrating how true enterprise value is unlocked by redesigning workflows for speed and scale expansion.
Harry holds his own ▶ 32:48 Harry demonstrates deep knowledge of startup margin structuresHarry showcases sharp industry insight by citing exact pass-through margin figures for high-profile AI startups like Replit and Lovable to challenge theoretical venture returns.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| AI as the New Electricity & Infrastructure Bottlenecks | 5 | 3 | 1 | 2 | Harry sets a structured debate around data, compute, and algorithms as core bottlenecks. Andrew reframes the primary near-term constraints to physical infrastructure, specifically electricity grid permitting and semiconductor supply. | |
| Token Efficiency and the Rise of Vertical AI Applications | 3 | 4 | 1 | 1 | Andrew details token generation efficiency improvements and draws a parallel to early internet horizontal search versus vertical apps. Harry allows Andrew room to explain the theoretical transition to vertical AI assistants. | |
| The Maturity and Impact of AI Coding Assistants | 6 | 4 | 2 | 3 | Harry introduces an expert quote from Joelle Pineau comparing current coding tools to 2016 image generation, which Andrew politely corrects. Harry then presses Andrew on whether Trump's administration helped or hurt AI infrastructure. | |
| AI's Impact on Workforce Productivity and Job Displacement | 7 | 4 | 2 | 6 | Harry cross-references David Cahn's bottom 5% workforce replacement barometer against Cohere's 10x framing and directly pushes back on Andrew's optimism regarding job displacement and analyst headcount reductions. | |
| Junior Talent Pipelines and Changing CS Education | 6 | 5 | 2 | 5 | Harry challenges Andrew on a white-collar junior talent pipeline crisis where AI eliminates entry-level associate roles. Andrew breaks down developer talent into four distinct tiers and blames outdated university curricula. | |
| Compensation in AI & the Impact of Wealth on Productivity | 6 | 3 | 3 | 6 | Harry challenges $100M pay packages for single engineers, arguing that sudden extreme wealth harms productivity and focus. Andrew pushes back based on Silicon Valley culture, while Harry contrasts Karpathy's 2% GDP forecast against Masa Son's 5-6%. | |
| The Geopolitical Impact of Open-Weight Models | 7 | 5 | 2 | 6 | Harry offers a sharp counter-hypothesis regarding open-weight models, suggesting China supports open models to leverage its superior manufacturing scale over model IP. Andrew acknowledges this while highlighting soft power and geopolitical influence. | |
| Is There a US vs. China AI Race? | 6 | 3 | 3 | 5 | Harry challenges Western and American arrogance regarding China's AI capabilities, highlighting the vastly superior work ethic and execution velocity in Chinese tech hubs. Andrew agrees, citing whole-of-nation state commitments. | |
| Export Controls on Chips & the Backfire Effect | 5 | 4 | 3 | 2 | Andrew argues that US semiconductor export controls backfired by forcing China to accelerate domestic chip development. When Harry asks how Europe can catch up, Andrew bluntly criticizes European regulators for viewing regulation as a competitive advantage. | |
| Where to Invest: Infrastructure vs. The Application Layer | 8 | 4 | 3 | 7 | Harry confronts Andrew on application layer unit economics, pointing out terrible gross margins and citing specific pass-through numbers like Replit and Lovable sending 80% of revenue directly to Anthropic. Andrew defends the layer by comparing current dynamics to early VC-subsidized food delivery. | |
| Monolithic vs. Specialized Models | 6 | 4 | 4 | 4 | Harry brings up Andrej Karpathy's claim that useful AI agents are a decade away, prompting Andrew to explicitly disagree and share concrete operational examples from his portfolio like Gaia Dynamics for tariff compliance. | |
| Building for the Technology of the Future | 6 | 4 | 2 | 5 | Harry asks if Andrew takes a 'utopian view' that terrible AI margins will simply resolve themselves over time. Andrew explains how internal engineering techniques bend cost curves down faster than market token prices fall. | |
| The Myth of the One-Person Unicorn and the Reality of Data | 7 | 4 | 2 | 7 | Harry dismisses corporate data optimism by citing direct conversations with enterprise CEOs and hard bans on ChatGPT at institutions like JPMorgan and Goldman Sachs. Andrew responds by comparing enterprise adoption speed to the multi-decade transition to the cloud. | |
| The Timeline of AI Adoption and Outdated Career Advice | 4 | 4 | 2 | 1 | Andrew calls advice urging people not to learn to code some of the worst career advice ever given. Harry facilitates Andrew's explanation of coding as precise intent specification for AI systems. | |
| Funding the Next Decade of AI Progress | 7 | 5 | 3 | 6 | Harry challenges Andrew on venture economics, arguing that AI funds require a budget shift from human labor to software to justify expanded TAMs. Andrew educates on how AI value creation relies on velocity and scale expansion rather than simple cost savings. | |
| Stack Integration, Financial Froth, and the Hype Problem | 7 | 4 | 3 | 5 | Harry pushes Andrew on circular financing deals, stack integration, and Sequoia's $600B AI revenue shortfall calculation. Andrew expresses concern over complex debt instruments shifting risk and criticizes existential risk hype. | |
| Quick Fire Round & Conclusion | 5 | 3 | 2 | 4 | In the quick fire round, Harry presses Andrew on AI Fund's cap table terms and operational structure. Andrew explains their studio model of taking common stock sweat equity and co-founding companies. |
Statements from this episode (43)
Andrew Ng: No one in AI has ever felt they have enough compute
“In my career working in AI, I have yet to meet a single person that ever felt like they had enough compute.”
Andrew Ng: Open-weight AI models are a tremendous source of geopolitical influence
“I think that open-weight models is a tremendous source of geopolitical influence.”
Andrew Ng: China's all-nation industrial commitment is a powerful force
“When China's government makes an all-nation commitment, it's all industrial. Commitment. That's actually a very powerful force that I wouldn't underestimate.”
Andrew Ng: AI inference demand currently exceeds global hardware and power capacity
“I find that many companies have, really have excess demand, which is a very rare problem to have, but so many people want more OM inference, want more tokens generated, and we just don't have the semiconductors and data centers and electricity to meet the dema…”
Andrew Ng: ChatGPT leads horizontal AI, but Gemini has a distribution advantage
“ChatGPT seems to be the dominant player in the new, new gen horizontal information discovery. Although I think Gemini with this channel advantage through control of Android and Chrome, you know, is a serious player as well.”
Andrew Ng: AI coding tools foreshadow AI adoption in other job functions
“I often look at AI coding assistance as a harbinger for what might happen to other job functions as well, as the AI marketing tools become more efficient, as the AI recruiting tools become more efficient, the AI finance tools become more efficient. So I often …”
Andrew Ng: AI coding tools are more mature than 2016 image generators
“I think it's further along. I think in 2016 image generation wasn't super valuable.”
Andrew Ng: Claims that AI could cause human extinction are ridiculous
“There are a lot of hyped up AI safety narratives saying AI could lead to human extinction, which is kind of ridiculous statement.”
Andrew Ng: Trump and David Sacks did a good job clearing AI regulations
“I think Trump did a good job and then his whole team, David Sachs and Christian and so on, did a good job clearing out unnecessary regulations.”
Andrew Ng: One engineer can now build six-month projects in a weekend
“There are so many projects that used to take, you know, six engineers, half a year to build, that today I or one of my engineers can build in the weekend.”
Andrew Ng: Everyone across all job roles should learn to code
“I think everyone should learn to code.”
Andrew Ng: Artificial general intelligence is decades or longer away
“So this phantom AGI, someday with AI, they can do everything a human can do. I think we're very far away from that. I would say, you know, like decades away, maybe even longer.”
Andrew Ng refuses to hire experienced engineers who ignore AI tools
“Those people, I just don't hire people like that anymore, but there are people that, you know, did a comfortable job, they kept coding the old way, and they just did not learn AI.”
Andrew Ng: Businesses face a massive shortage of AI-fluent college graduates
“The fresh college grads know AI. We can't find enough of them. So many businesses would love to hire those fresh college grads.”
Andrew Ng: AI will make intelligence cheap and boost GDP growth
“I hope we can get much closer to five, six or more percent GDP growth. When looking at the future, it turns out one of the most expensive things in today's world is intelligence. This is why it's so expensive, at least in the US, to hire a highly skilled docto…”
Andrew Ng: China is leading the US in releasing open-weight AI models
“China especially has been really taking the lead or, well, Taking a lead or getting up there in terms of releasing tons of really good open way models. So I would say if not, it is kind of not whatever predicted, you know, a decade ago that China AI would end …”
Andrew Ng: Closed AI models and high salaries slow Western AI innovation
“When the US has more closed models, and when, you know, teams are trying to pay these hundred million dollar salaries to extract talent, then that circulation of knowledge becomes very slow, and it slows down the rate of American and European innovation.”
Andrew Ng: China builds a global AI user base via free models
“Open-weight models are a key part of the AI supply chain, and China releasing, you know, free low-cost or free models into that key part of supply chain means it's, Really starting to build up a lead, right, in, in build up a commanding user base”
Andrew Ng: AGI is hyped PR, AI will improve continuously for decades
“I feel like because of PR goals, AGI has been hyped up as a finish line, but I don't think it's a finish line. It's just, we'll have continually improving capabilities for, you know, decades to come.”
Andrew Ng: US chip export controls have backfired and accelerated Chinese innovation
“I think the export control on chips has largely backfired. The way that the US and I think that the way the US first put restrictions on Huawei and then later on, you know, exported NVIDIA and AMD and other semiconductors that really incentivized China. So bef…”
Andrew Ng: Being a leader in AI regulation is not a competitive advantage
“If I had one wish for the European regulators, I spoke with quite a few European regulators, I was hearing things like, we want to be leaders in regulating AI, and that's a competitive advantage. And with all due respect, that's not a competitive advantage. So…”
Andrew Ng: AI application capital recycles through model providers to Nvidia
“In fact, if you look a lot of the application layer investments. Sometimes it feels like, you know, firms are putting in a hundred million dollars so that they can pay Open Ananthropic. So the Open Ananthropic can pay NVIDIA, which is where all the money is, i…”
Andrew Ng: High capital efficiency hinders deploying billions in AI applications
“But the dilemma is you could do it in a very capital efficient way. So if someone wants to say, I want to put ten billion dollars to work. You know, yes, you can build ten billion dollars worth of data centers. We know how to spend that money. But how do you s…”
Harry Stebbings: AI coding startups pay 80% of pass-through revenue to Anthropic
“If you look at a Rapplet or a Lovable, 80% of their pass-through is too anthropic.”
Andrew Ng: Heavy VC subsidies for AI coding tools cannot continue indefinitely
“I think we're seeing that right now with a lot of VC subsidized, you know, AI coding the laws of physics or the laws of finance says that at some point, right, this can't go on forever.”
Andrew Ng: Useful AI agentic workflows exist today, disagreeing with Andrej Karpathy
“I disagree with that. I think we're seeing useful agentic workflows right now.”
Andrew Ng: AI Fund ignores initial token costs when prototyping products
“Kind of, frankly, when we build prototype, We routinely just not worry about token costs because the first most important thing is this build a product that users love.”
Andrew Ng: AI builders should design for future margins, not today's costs
“Absolute margins are important, but when you have a view for where the technology is going, Then it lets you not build for the margins today, but what you can forecast them being in the future”
Andrew Ng: AI technology itself does not create business moats
“AI as a technology Doesn't really offer an answer to the moats for most businesses. So if you're building AI, you know, for drones or legal or for whatever, the moat is more of a function of that industry.”
Andrew Ng: Proprietary software is a significantly weaker moat today
“Previously software used to be a moat, right? If you had, you know, invested 10 years to build a software, it's really hard to replicate that. That one moat is much weaker than before”
Andrew Ng: Change management is enterprises' main AI implementation barrier
“I think the biggest barrier in most large enterprises is, is actually people and change management.”
Andrew Ng: Data is not the primary bottleneck in AI today
“I think it's definitely not data. Not that data is not important, but that's definitely not the bottleneck.”
Andrew Ng: Enterprise AI adoption and building will take over a decade
“I actually think that a decade from now, we will still be working to identify Valuable applications and enterprises and building them. Having said that, we will make a lot of progress over the next one or two years, but we're not going to be done, you know, ev…”
Andrew Ng: Advising people not to learn coding is terrible career advice
“Even earlier this year, we saw some senior business leaders advise people to not learn to code on the grounds that AI will automate it. We'll look back on that as some of the worst career advice ever given.”
Andrew Ng: Writing code by hand is becoming obsolete due to AI
“Writing code by hand is becoming obsolete, right? Oh, really don't do that. Not, but they get AI to write code for you.”
Harry Stebbings: AI investing success requires shifting human labor spend to software
“But if you look at the TAM, the secret to success in AI investing is where we see a transition from software budgets to human, from human labor budgets, sorry, to software budgets.”
Andrew Ng: AI creates value by accelerating workflows and expanding scale
“So instead of cost savings, if AI lets you do something way faster, or lets you take a task and do it a thousand times more, instead of serving a small number of people, let's serve a lot more people because it's not economic to do so. These are the two patter…”
Andrew Ng: AI will transition from vertical integration to horizontal standards
“So when an industry is immature, it turns out where to draw those boundaries to let different participants do their part and have it still interoperate. That's less clear. But then as the industry matures in this You know, more standards for if I want to publi…”
Andrew Ng: The AI application layer is already delivering clear ROI
“So what I'm seeing is for the application layer, there is very clear ROI. I think it's fantastic. So someone else trained these models who can build applications for, you know, a 100,000 dollars or a million dollars and start generating ROI.”
Andrew Ng: Moats in coding and API tools are weaker than consumer brands
“So the coding dev tools and API tools market, the mode there's weaker than having a strong consumer brand.”
Andrew Ng: AI Fund typically invests $1M at a $4M valuation cap
“Depends. We end up with some common stock for the sweat equity of building the company, and then we usually our first check-in is usually like a million dollars at a four million dollar cap so kind of 20% ownership or safe.”
Andrew Ng: AI requires existing workers to reskill, unlike past technological shifts
“The change is so fast this time round that we need people that are alive today to learn new skills, as opposed to we need their kids to learn new skills, and that's actually very challenging, and historically I don't think we've ever been good at that.”
Andrew Ng: AI companies facing existential risk drive the worst industry hype
“The companies with something to lose whose statements over time have become more moderated. So, you know, I find that as your established company, you just say more sensible things, but there are some companies that I think are at greater existential risk. And…”