Nov 14, 2024 · 51m · mad
State of AI 2024: Frontier Models, AI Geopolitics, Robotics | Nathan Benaich, Air Street Capital
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
Host Matt Turck interviews Nathan Benaich, founder of Air Street Capital and author of the annual State of AI Report, to discuss key research, commercial, and geopolitical trends defining artificial intelligence in 2024. They explore the collapse of inference costs, China's open-source rise, European regulatory friction, consumer hardware adoption, and realistic timelines for robotics.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 18.6% of the talking time here. How this is scored →
Matt as informed peer 4.5 Guest teaching 4.0 Guest disagreement 1.6 Matt pushing back 1.9
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 27:20 Dismissing 'agentic' terminology
Nathan aggressively rejects industry buzzwords like 'agentic', calling it techy jargon that frightens non-technical users.
Hardest push from Matt ▶ 27:25 Host challenges guest on jargon rejectionMatt immediately pushes back when Nathan disdains the term 'agentic', directly asking 'Why not?' to force justification.
Biggest teaching moment ▶ 42:28 Alternative chip investment mathNathan educates the host on the poor performance of AI chip startups by calculating how $6B invested across 7 competitors vastly underperformed buying NVIDIA shares directly.
Matt holds his own ▶ 22:00 Host quotes precise report metricsMatt demonstrates deep familiarity with the report data by quoting the exact figure that top AI startups reach $30M ARR in 20 months compared to 65 months for traditional SaaS.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Methodology and Compilation Process of the Report | 3 | 2 | 1 | 1 | Nathan details his monthly workflow for compiling the State of AI report and notes the major cultural shift from doomerism to product scaling. Matt facilitates with friendly commentary on the vibe shift without deep technical probing. | |
| Foundation Model Progress, Multimodality, and China's Open-Source Surge | 3 | 4 | 0 | 1 | Nathan breaks down foundation model developments, emphasizing multimodality and the surprising surge in open-source contributions from Chinese labs like Alibaba and DeepSeek despite U.S. chip sanctions. | |
| Open Source vs. Closed Models and OpenAI's Market Position | 5 | 4 | 2 | 2 | Matt references his prior interview with Hugging Face's CEO to contextualize open source growth. Nathan counters narrative fragmentation by highlighting Ramp corporate credit card data showing OpenAI still commands 80% market share. | |
| Meta's Open-Source Strategy with Llama and Monetization Dynamics | 5 | 4 | 2 | 3 | Matt quotes Nathan's description of Zuckerberg as the messiah of open source and questions Meta's long-term business strategy. Nathan outlines Meta's market capitalization recovery and compares open versus closed ecosystems to iOS versus Android. | |
| Laboratory Revenue, Inference Cost Collapse, and Enterprise Adoption | 6 | 3 | 0 | 1 | Matt demonstrates high expertise by reciting exact revenue milestone figures from the report comparing AI startups to traditional SaaS. Nathan explains how post-training optimizations and price wars collapsed inference costs. | |
| Consumer AI Applications, Voice Cloning, and Hype Cycles | 3 | 4 | 0 | 1 | Nathan highlights voice cloning as his go-to compelling demo for non-AI experts, grounding it in the ten-year history of speech scaling laws since Baidu's Deep Speech 2 paper. | |
| Contrarian Investment Opportunities, Product Engineering, and Robotics | 6 | 5 | 4 | 4 | Nathan forcefully disdains the popular industry buzzword 'agentic' as alienating tech jargon. Matt directly pushes back by asking 'Why not?' and synthesizes Nathan's thesis on the shift from deep tech to product engineering. | |
| Skepticism Around AI Hardware and Success of Smart Glasses | 5 | 5 | 3 | 3 | Nathan expresses skepticism toward standalone hardware gadgets like AI pins while praising Meta's smart glasses. Matt steeringly probes the regulatory overhang in Europe that blocks major AI features. | |
| The European AI Talent Pool and US Magnetism | 4 | 4 | 2 | 3 | Matt examines European tech dynamics from Nathan's London VC perspective. When Nathan critiques European policy inertia, Matt counters by pointing out that the U.S. is not completely immune to bad regulations. | |
| Evaluating State of AI Report Predictions and Market Realities | 6 | 5 | 3 | 2 | Matt directs attention to specific report slides on chip market dynamics. Nathan presents a striking comparative calculation showing that $6B invested into alternative chip startups heavily underperformed simply holding NVIDIA stock. | |
| Humanoid Robotics, Autonomous Vehicles, Vision Models, and AI Resources | 4 | 4 | 0 | 0 | The episode concludes with Nathan expressing skepticism toward humanoid robotics timelines and highlighting historical leaps in computer vision commonsense, before sharing his top information sources. |
Statements from this episode (22)
Benaich: The AI existential risk debate was overblown
“I felt like the extra risk kind of debate was kind of overblown .”
Research shows the general public does not fall for AI deepfakes
“There's been enough research now that, you know, individual, like, general population doesn't really fall for deepfakes, like, whether it's calls or, like, synthetic media. Doesn't really change consumer opinion that much. People are not dumb.”
Benaich: Tech leaders underestimate human inertia in real-world AI adoption
“Tech people in SF obviously live in a bubble, and it's kind of, it's quite techno-utopist, and I think they underestimate how, how much human inertia there is to the diffusion and adoption of technology in, like, the real world.”
Chinese AI labs like DeepSeek and Alibaba perform strongly on benchmarks
“China was kind of not in this fight, like, 12 months ago, and now is very much in it. Like, their models like Alibaba's Quen and then this spin out from a quantitative hedge fund DeepSeek, which publishes code models and others, and they've been actually a ver…”
Benaich: Hugging Face data shows over 500M Llama derivative downloads
“The data as of like a week or two ago from his platform, Hugging Faces, like there's probably over half a billion Lama derivative models that have been downloaded.”
OpenAI retains 80% of corporate AI model spending, per Ramp data
“Really, like, when you sum the makers of all these models that are getting used, it's still, like, 80% OpenAI. Like, that hasn't changed, so it doesn't look like, fragmentation to me, it looks like domination.”
Benaich: Meta's AI shift is one of public market's best ROI trades
“Yeah, I mean the, it's probably one of the best like ROI trades of like a public company in like a long time.”
Benaich: Databricks will likely become the primary enabler for open-source AI
“Probably Databricks, and Databricks and Snowflake, probably more Databricks will be like this like enabler of open source.”
Benaich: Meta reported a 7% to 10% click-through gain on generative ads
“He gave some stats around how consumers click through at a higher rate, and it was material, it was like something in the range of seven to 10% click through improvement on generative ads.”
AI inference costs plummeted from $60 to $0.06 per million tokens
“The most expensive models like a year ago to now, I don't know, the most expensive one is maybe like 600 dollars, sorry, 60 dollars per million tokens, and now the cheapest one is probably like 0.06 dollars per million tokens. Not the same level of intelligenc…”
Enterprise AI app 12-month retention jumped from 43% to 65%
“Just directionally, it was like retention after one year in 20, 22 cohort was like 43%, I think. And then the cohort from 20, 23 after 12 months was like 65%. So that was pretty material.”
Nathan Benaich: AI voice generation technology is basically solved
“And for me, the demo I give them is, like, voice generation, because I think it's basically solved.”
Benaich: Consumer AI product surface is vastly larger than enterprise AI
“The surface of products that you could build in the consumer, prosumer space is vastly bigger than what you see in enterprise.”
Benaich: Mainstream consumers find AI jargon like 'agentic' intimidating
“Kids that are saying agentic, workflow, clone, like, normal people think that's scary. Like, that's not some, that's not a product somebody wants to use.”
Benaich: Vision-language models unlock general-purpose warehouse robotics
“Now it actually looks increasingly the case that you can take one of these vision language systems and train a general purpose system that can work across all your customers, and that's like a pretty big unlock, because the go to, the go live time, and therefo…”
Consumers buy Meta smart glasses for audio, not for AI features
“There's like, people like the glasses because the music is great. Like the AI sucks, basically, as far as I can tell from people, but the music is great.”
Benaich: EU AI regulations are unlikely to change despite growing pushback
“I think the nourish is too big. I mean, Michael has had this narrative for a while, but, and like France was one of the bigger kind of defenders of open source, et cetera, but I don't think it's going to change much.”
Buying Nvidia stock instead of funding AI chip rivals yielded 4x more
“Six billion turned into thirty billion roughly, of which half is CameraCon, which is a Chinese listed company. And then NVIDIA would be worth a hundred and twenty billion.”
Nvidia chips appear in 11x more AI research papers than all competitors
“If you sum all the papers that use NVIDIA chips versus all the papers that use TPUs, FPGAs, ASICs, Apple, Huawei, the sum of all NVIDIA papers versus the sum of everybody else, the delta is like 11 times.”
State of AI Report predicts a fully AI-generated app will hit top 50 App Store charts
“This year we predicted that there would be some kind of like blockbuster app, like call it, I don't know, like Apple store top 100 or top 50 that was like fully written with, in a generative fashion.”
Benaich: Humanoid robotics progress will be a long slog like autonomous vehicles
“I feel like it's going to be self-driving. Like the progress so far looks exciting. The demos look good, but we obviously don't see everything that's happening in the background, and I think it's going to be like a long slog.”
Benaich: AI visual common sense capability is basically solved
“The general capability of, like, visual common sense, and, like, I show you a scene, what's happening in it, and it's basically solved right now. Like, you know, machines can describe this in intricate detail and get the nuances pretty well.”