May 16, 2024 · 55m · mad
AI, Data and Blockchain: a VC perspective | Tomasz Tunguz, Founder of Theory Ventures
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 episode of The MAD Podcast, host Matt Turck interviews venture capitalist Tomasz Tunguz, founder of Theory Ventures, to discuss macro funding trends in AI, the evolution of the Modern Data Stack, Web3 unit economics, and strategies for building early-stage venture firms.
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 21.4% of the talking time here. How this is scored →
Matt as informed peer 3.5 Guest teaching 3.8 Guest disagreement 1.1 Matt pushing back 1.4
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 4:55 Tomasz rejects standard consensus on Ethereum
Tomasz explicitly explicitly dismisses mainstream views on Web3 by arguing that Ethereum should be evaluated primarily as a highly profitable database software business rather than a speculative asset.
Hardest push from Matt ▶ 25:26 Matt refutes Tomasz's AI revenue decay theoryMatt refuses Tomasz's premise that changing architecture drives fast AI revenue drops, directly asserting that revenue atrophy stems from initial trial and experimental budgets running out.
Biggest teaching moment ▶ 41:07 Tomasz's historical taxonomy of BI toolsTomasz educates Matt on why BI is fundamentally about governance rather than charts, mapping out two decades of shifts between centralization (MicroStrategy, Looker) and decentralization (Tableau, Excel).
Matt holds his own ▶ 32:35 Matt cites insider insights from dbt founderMatt demonstrates deep expertise in the data infrastructure landscape by citing his direct discussions with dbt founder Tristan Handy regarding the death of the Modern Data Stack.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome and Background on VC Perspectives in Tech | 2 | 4 | 1 | 1 | Matt introduces Tomasz and frames the interview as two VCs chatting about tech trends. Tomasz shares data regarding $80B being invested into AI in 2024 and crypto developer trends. | |
| Web3 Databases and Ethereum as a Financial Asset | 3 | 5 | 2 | 1 | Tomasz presents a novel perspective comparing Ethereum to a high-margin database company producing $400M in quarterly profit. Matt listens and synthesizes the core data privacy and compliance use case. | |
| Decentralized Compute, Security, and AI-Blockchain Intersections | 2 | 4 | 1 | 1 | Tomasz illustrates decentralized Web3 use cases such as customer wallets replacing vendor data duplication. Matt prompts for Tomasz's perspective on the intersection of blockchain and AI compute. | |
| The AI Stack: Hardware, GPUs, and Cloud Infrastructure | 5 | 3 | 1 | 2 | Matt demonstrates strong domain context by noting CoreWeave's pivot from crypto mining to AI infrastructure in New Jersey and questioning whether massive cloud provider CapEx is positive for startups. | |
| Foundation Models, Small Language Models, and Constellation Architectures | 3 | 6 | 1 | 1 | Tomasz outlines technical architectures, contrasting open source SLMs like Llama 3 with massive LLMs and introducing 'constellation models' and error-quashing steps in LLM chaining. | |
| Generative vs. Classical AI, RAG, and Information Retrieval | 4 | 4 | 1 | 1 | Matt highlights social media hype around GenAI replacing all software, while Tomasz explains why classical AI models remain essential for tabular data and classifier guardrails alongside RAG. | |
| AI Developer Tooling and Real-Time Vector Computers | 2 | 4 | 1 | 1 | Tomasz breaks down Theory's investments in developer tooling, specifically explaining Superlinked's 'vector computer' concept using TikTok recommendation engines as an analogy. | |
| The AI Application Layer, Revenue Quality, and Enterprise Accuracy | 5 | 3 | 2 | 5 | Matt directly challenges Tomasz's suggestion that AI application revenue atrophy stems from changing architectures, arguing instead that initial revenue spikes were driven by experimental trial budgets. | |
| Copilots vs. Autonomous Agents and Vertical AI Models | 3 | 4 | 1 | 1 | Matt inquires about copilot versus full agent execution preferences. Tomasz responds by comparing software agents to industrial assembly line robots in terms of productivity multipliers. | |
| AI Unit Economics, Gross Margins, and Workforce Shifts | 4 | 4 | 1 | 1 | Matt references Tomasz's writing on gross margins and unit economics. Tomasz presents a thesis that enterprise AI adoption will trigger a one-time step-function increase in S&P bottom-line margins. | |
| State of the Modern Data Stack, Tool Fatigue, and Apache Iceberg | 6 | 3 | 1 | 2 | Matt demonstrates high authority by citing his interview with dbt founder Tristan Handy regarding whether the Modern Data Stack is dead, prompting Tomasz on tool fatigue and Apache Iceberg. | |
| Industry Consolidation, Snowflake vs. Databricks, and Microsoft Fabric | 5 | 3 | 1 | 2 | Matt actively steers the conversation on vendor dynamics, asking whether Fabric is crashing the party and tracing Databricks' progression from ML infrastructure into relational data warehouses. | |
| High-Performance Query Engines: Databricks SQL, DuckDB, and MotherDuck | 3 | 5 | 1 | 1 | Matt specifically highlights DuckDB and MotherDuck. Tomasz explains the technical architecture of WASM-compiled client-side query execution powering analytics in browser applications. | |
| Modern Business Intelligence & Theory's Investment in Omni | 4 | 5 | 1 | 2 | Matt notes that BI felt like an unloved area lacking innovation. Tomasz explains Theory's Omni investment by mapping the 20-year historical pendulum between centralized data modeling and decentralized empowerment. | |
| Data Governance vs. Data Democratization Philosophies | 5 | 4 | 1 | 2 | Matt challenges why data democratization remains unfulfilled in practice, describing typical enterprise bottlenecks. Tomasz contrasts two competing executive philosophies on data governance. | |
| M&A Strategy and Lessons from the $2.6B Looker Acquisition | 3 | 5 | 1 | 1 | Matt asks how startup acquisitions can be engineered. Tomasz draws on the $2.6B Looker sale to explain that major acquisitions require an internal champion betting their career over 12-18 months. | |
| Go-to-Market Strategies and Ecosystems in Data Infrastructure | 3 | 3 | 1 | 1 | Tomasz shares lessons on ecosystem co-selling between AEs and discusses founding Theory Ventures, detailing internal firm culture where every team member executes an annual experiment with an expected 70% failure rate. | |
| Interview Conclusion and Final Remarks | 1 | 0 | 0 | 0 | Standard podcast outro where Matt thanks Tomasz and prompts listeners to subscribe and leave reviews. |
Statements from this episode (43)
Tunguz: ~$80B will be invested in AI venture capital in 2024
“I think there'll be something like eighty billion dollars invested in 2024 in AI.”
Tunguz: VC funding peaked at $300B in 2021, fell to $175B
“Venture capital, like it hits peak, I think in 21 at 300 and then it fell to one 75.”
AI focus creates pricing arbitrage in non-AI startup categories
“There's probably some pricing arbitrage in other categories where people aren't paying attention.”
Crypto token equity rounds will hit hundreds of millions to billions
“And then many of the token launches will raise equity rounds before they go public. And those will be in the several hundred million to billions again.”
Only three crypto publications matter for fundraising or product announcements
“So now there are three crypto publications in totality that matter for a fundraising announcement or a new product announcement.”
Ethereum's market cap was 7.6 times Snowflake's in Q1 2024
“The total market cap of Ethereum is 7.6 times snowflakes market cap.”
Ethereum's $400M Q1 net income makes it the most profitable software business
“In Q one, if you were to look at Ethereum as a business and we can have this debate about whether it is or not, but if you were to look at it as a business, it produced four hundred million roughly in free cashflow or net income, which makes it the most profit…”
Ethereum should be evaluated as a database company
“And so the way that I look at it, and this is not a broadly held view, but the way I look at it is that's a database company.”
Ethereum write costs fell from 1M to 1,000 times AWS RDS
“Three years ago, if I were to write the same data to Ethereum, it would cost me a million times more to write that row. Today, it costs about a thousand times more.”
Most major software companies will eventually integrate a Web3 database
“And so there's this asymptotic cost curve that will come down, and at some point, our belief is that most major software companies will actually have a Web three database as part of their stack.”
~35 US states will enact data privacy regulations by late 2025
“I think by the end of 20, 25, something like 35 states in the US will have their own privacy regulation.”
GPU cloud providers resemble REITs, requiring heavy debt and Nvidia ties
“I think you really need to be pretty sophisticated when it comes to financial engineering, what kind of debt products that you use. Kind of like a REIT, a real estate investment trust, and that you'll, you'll produce really good profits, but you need a lot of …”
Top three cloud providers spend $60B to $70B quarterly on CapEx
“Yeah, 60 to seventy billion per quarter now. From the top three clouds, which it's and it just keeps going up”
Amazon spiked cloud CapEx 18 months before Google and Microsoft
“Amazon Spiked almost 18 months before Google and Microsoft did, so they must have seen some of these inference workloads coming.”
Hyperscaler GPU CapEx heavily benefits early-stage AI startups and VCs
“And ideally somebody else is, it's not venture capital dollars that are being used to buy and manage those GPUs. So it's a huge benefit.”
OpenAI inference pricing drops 160x between model generations
“If you look at the pricing page for, like, OpenAI, and you look at the 4.5 turbo, and the cost per inference, and compare it to the next most recent model, there's a 160 X difference in pricing.”
Next-gen AI will route queries via hybrid constellation models
“We think that the next generation architectures look like that, at least for handling very simple tasks.”
Chained LLMs suffer from compounding errors across multi-step pipelines
“One of the challenges that will happen with some of these large language models, particularly when they're chained or we let them operate for hours at a time is the intern effect where they're off at the beginning. And that's just because they're chaotic and n…”
Generative AI is fundamentally a multi-gigabyte knowledge compression engine
“The way we think, I think about generative is, it's a really great knowledge compression engine, right? Like I can compress the internet into a model that's like three gigs, right?”
$1B to $10B foundation model training costs price out startups
“As the cost to train all these large language models becomes bigger and bigger, Amazon's talking about a billion dollars for a single run or ten billion dollars for a single run. It's beyond the realm of startups to really spend any time there.”
Core AI IP advances may shift from foundation models to retrieval
“So maybe the core intellectual property advance actually happens in the information retrieval step and the embeddings and the vectors as opposed to the underlying models.”
Companies will need custom formulas combining qualitative and quantitative vector data
“Each company will need to embed a different formula for defining vectors, For the content or the recommendations that you need to make that will have many different components that will be both qualitative and quantitative.”
Spun out, the top three cloud providers would be worth $2.5T
“So, if you look at web two infrastructure you take the top three clouds by size, and you were, hypothetically, spin them out as separate mark, as separate public companies. It'd be worth about 2.5 trillion.”
40% of deployed Robotic Process Automation bots are currently broken
“RPA, our analysis shows that a lot of the existing robots that are, about 40% of the existing robots deployed in RPA are broken, and that's either because the human process changed, Or the software underneath it changed.”
Average enterprises review under 1% of 5,000 to 7,000 daily security alerts
“We just invested in a company that's automating security analysis for the average enterprise of 75 security products. Each of those security products produces Alerts about 5000 to 7000 alerts per company per day. Less than one percent are reviewed.”
Machine learning's economic value lies in getting from 80% to 98% accuracy
“Machine learning systems it's easy to get to an 80% solution, but all the values in that marginal 15 or 18 to get to 95 to 98% accuracy.”
Microsoft and ServiceNow report 50-75% developer productivity gains from AI
“So Microsoft and ServiceNow have said there's 50 to 75% increase in developer productivity.”
Assembly line mechanical robots replace about 2.5 human jobs
“The crudest, the crude analogy that we have, which is Or the best that we found is for a mechanical robot on an assembly line, how many humans jobs does that robot replace? And the answers are about 2.5.”
Autonomous AI agents will likely boost productivity 4-5x more than copilots
“So theoretically, if it's anywhere close can a copilot or sorry, an agent is probably four to five times more productivity boosting job producing than an agent, than a copilot.”
Google's generative search queries initially cost 10x more than standard queries
“Well, Google, when they initially released search, their generative search experience, the cost to serve a query was 10 X a standard one.”
Average company gross profit will grow 2-5% within 24 months via AI
“We, I mean, and these are just guesses, but we think that the average company in 18 to 24 months should be two to three to five percentage points more gross, have more gross profit.”
Cost cutting among enterprise data leaders has run its course
“Fed raises rates and cost cutting becomes the number one priority for all of these data leaders. I think that's now run its course.”
Marking up storage accounts for roughly 11% of Snowflake's revenue
“It's about 11% of Snowflake's revenue is marking up storage.”
Expect major data stack vendor consolidation in the coming quarters
“So I think you'll see quite a lot of consolidation in the next couple of quarters.”
Antitrust locks Big Tech out of M&A, empowering Snowflake and Databricks
“But you look at like Amazon, Google, Facebook, Microsoft, all are under an attack trust and basically impossible for them to acquire. So there's, you don't really have the big chip stacks at the table anymore, which means if you're an acquirer like Snowflake o…”
Amazon Redshift has fallen behind in next-generation cloud data warehousing
“Because the clouds themselves, you look at Redshift in the buyer's mind has really fallen behind in the next generation cloud data warehouse.”
Databricks SQL product line hit $200M in revenue and is doubling
“Databricks, the SQL products, two hundred million in revenue doubling.”
80% of cloud database query workloads are under 100 megabytes
“80% of the, those workloads are less than a hundred megs in size.”
Business intelligence software is fundamentally about governance, not charts
“Is it's really a governance tool. Like, the charts themselves are important and interesting, but the surface area of the products are enormous and take years to build, because it's all about how do we make sure that the data is going to the right person and it…”
Looker's embedded BI drove 33% of revenue with 2 of 750 employees
“So at Looker, when we sold the business, a third of the revenue was embedded. And out of a team of about seven to 750, there were two people working on that product.”
Most data leaders favor restricted access to ensure data accuracy
“The one school of thought, what would you call classical, is that it's better for the organization to have less data if the data is all correct. And it's very easy to misinterpret data. There's there needs to be a lot of education on statistical significance o…”
$50M+ startup acquisitions take 12-18 months of internal championing
“So when someone buys a business, not an acquihire, but let's say like 50 to fifty million plus, someone in a company, in the acquiring company is saying, I am betting my career, effectively, in this company, that we should buy this company at this price. And t…”
Looker originally brought Snowflake into sales deals before momentum inverted
“When Looker went to market, we initially partnered with Snowflake, and we were bringing Snowflake into deals, and that worked really well, and then Snowflake started to grow really fast, and Snowflake was bringing Looker into deals, and so so why is this?”