Apr 26, 2024 · 49m · mad
Navigating the AI Landscape: A Survival Guide | 2024 MAD Landscape with Matt Turck and Aman Kabeer
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
FirstMark Capital's Matt Turck and Aman Kabeer analyze the 2024 Machine Learning, AI & Data (MAD) Landscape, examining market saturation, the transition from structured data to AI infrastructure, and emerging trends in open-source LLMs, AI agents, and venture capital strategy.
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 91.6% of the talking time here. How this is scored →
Matt as informed peer 7.6 Guest teaching 0.3 Guest disagreement 0.1 Matt pushing back 0.0
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 48:43 Aman jokingly refuses to work until AI agents exist
In a lighthearted exchange, Aman playfully pushes back on taking on workload for the next landscape, stating he will delay work until AI agents can do it for him.
Hardest push from Matt ▶ 31:11 Matt rejects the hype around coining a 'Modern AI Stack'Matt pushes back against popular industry terminology, explaining why he resisted using 'Modern AI Stack' due to its association with previous irrational exuberance.
Biggest teaching moment ▶ 6:47 Aman highlights private equity roll-up activityAman introduces a specific nuance that Matt's blog post omitted, pointing out that recent acquisition activity in data has been led by PE roll-ups rather than traditional M&A.
Matt holds his own ▶ 4:20 Matt demonstrates deep market memory on industry wavesMatt expertly dissects a complex market question by synthesizing nine years of VC history across the Modern Data Stack and GenAI cycles.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Viral Tweet Context: Give the Intern a Raise | 6 | 0 | 0 | 0 | Matt introduces the episode context, explaining the background of the MAD landscape map and joking about Aman's viral intern fame. Aman collaboratively plays along with self-deprecating humor about coming out of the basement. | |
| Deciphering Market Crowding Data Waves and Lack of M&A | 8 | 1 | 0 | 0 | Matt lays out a detailed historical perspective on market crowding across the Modern Data Stack and GenAI waves, citing regulatory and macro factors. Aman adds a brief observation about private equity roll-ups. | |
| Open Source AI Dynamics and Meta Strategic Play | 8 | 1 | 1 | 0 | Matt offers a deep analysis of Meta's open source strategy with Llama 3 as a power play against mobile duopolies. Aman facilitates smoothly and points out how crowded Hugging Face has become. | |
| Commercial AI Performance Value Capture and Full-Stack Layers | 7 | 0 | 0 | 0 | Matt dissects commercial value capture in AI, comparing full-stack application layers to cloud storage provider differentiation strategies. Aman acts as a supportive interviewer asking open-ended questions. | |
| OpenAI Market Dominance Executive Drama and Exponential Growth | 8 | 0 | 0 | 0 | Matt provides an insightful characterization of OpenAI and Sam Altman as defying startup laws of gravity. Aman keeps the dialogue light with a joke about making a Succession-style TV episode. | |
| Cracks in AI Hype Startup Struggles and Incumbent Advantages | 8 | 0 | 0 | 0 | Matt systematically lists signs of fatigue in the AI hype cycle, highlighting struggles at Inflection, Stability, Tome, and Jasper. Aman agrees on the reality of startup execution challenges. | |
| Venture Capital Strategies and Enterprise Budget Shifts | 7 | 0 | 0 | 0 | Matt clarifies how enterprise spend is shifting from innovation budgets to operational budgets and contrasts different VC fund strategies. Aman prompts the conversation with standard investor questions. | |
| Modern AI Stack vs Modern Data Stack and Vector DBs | 8 | 1 | 0 | 0 | Matt contrasts the technical architectures of structured vs unstructured data stacks and explains vector database market dynamics. Aman asks clarifying questions about market overlap. | |
| Emerging Frontiers AI Agents Edge Hardware and AI-Native SaaS | 8 | 0 | 0 | 0 | Matt elaborates on AI agents, edge hardware, and the emergence of AI-native vertical SaaS applications. Aman shares enthusiastic prompts regarding consumer hardware like the Vision Pro. | |
| The Evolution and Tough Realities of Modern Data Stack | 8 | 0 | 0 | 0 | Matt explains the macro economic pressures hitting Modern Data Stack startups while reaffirming the underlying architectural validity. Aman closes the interview collaboratively. |
Statements from this episode (13)
Turck says multiple $200M-$600M revenue software companies are waiting to IPO
“There's a whole range of companies that are somewhere between, call it, two hundred million NRR and five or six hundred million NRR that have been sitting on the sidelines waiting to go public.”
Hugging Face hosts over one million AI models as of April 2024
“We saw those, what, one million, over one million models on Hugging Face, and literally thousands added every day”
Turck predicts the open-source AI market will consolidate via power law
“So eventually there'll be a power law and all the things”
Turck predicts AI value will be captured at the application layer
“It feels like a lot of the value is actually going to be captured and retained through that application layer in, in a way that is maybe not entirely different from the cloud vendors.”
Turck calls Sam Altman the Steve Jobs of this generation
“It seems that Sam Altman, who's you know, the Steve Jobs of our generation, we're actually joking the, you know, Steve Jobs of the TikTok generation when you know, unlike Steve Jobs who was fired and came back years later, like he left and came back within a, …”
Turck says an incremental GPT-5 release would end the AI hype
“It sort of feels like if it's an incremental improvement, you know, that may signify that this end, this part of the hype cycle is, is ending. If it's, you know, crazy as everybody's hoping, then yes, then it feels like we're, you know, still very much in that…”
Turck says Inflection AI would not have pivoted if commercial performance was strong
“If they had something unbelievable going on, they would probably not have made that decision.”
Turck says Midjourney makes $200M to $300M with 40 employees
“Midjourney has built what seems to be an unbelievable business with, you know, people effectively playing with stuff, you know, whatever their revenue are like, two hundred million or three hundred million, like, 40 people, so Presumably insanely profitable.”
Turck says horizontal AI startups will struggle against reactive tech incumbents
“It seems that for stuff that's very horizontal startups are going to have a harder time building long lasting businesses because in this Cycle. They are facing a very different type of incumbents compared to prior platform shifts where you know, this time you'…”
Turck says early enterprise AI purchases came from corporate innovation budgets
“A lot of the money that came out that the generating AI purchases came out of were innovation budgets rather than actual operational budgets.”
Turck says AI evaluation startups currently lack significant enterprise traction
“A lot of those companies to put it bluntly don't have a lot of traction yet, because it turns out that, if you're trying to do something, and I can, I'm not picking on them, but like evaluation or monitoring and all the things, well, you need to have LLMs to m…”
Turck says vector database viability depends on incumbent database feature parity
“Do you really need a, you know, a specialized database to do that? Or in a context where some of the more general purpose players like the MongoDBs of the world have started announcing vector capabilities, you know, is that good enough? The answer to that ques…”
FirstMark favors AI application founders with direct model-level technical skills
“We certainly gravitate towards companies where founders can do both, but certainly have the technical chops for going directly into the model and doing, you know, fine tuning rag, customizing with the data, like all sorts of different things one can do at the …”