State of Foundation Models - 2025
Substack · June 2025 · 100+ slide comprehensive analysis of the foundation model market
Profile
The Research-First Investor
There is a specific kind of venture investor who shows up in your inbox after reading every paper on your problem domain, mapping every adjacent startup, and publishing an essay on the topic before you even sent a cold email. Davis Treybig is that investor. He does not enter a new space until he has spent months inside it. This is not patience - it is precision. By the time he writes a check, he has already formed the thesis in public.
In May 2025, Innovation Endeavors announced his promotion to General Partner after six years at the firm. The announcement described him as someone who spots "systems that will define the next decade" and is "the first person many founders call when they want to spar a new idea." Both things track.
"I typically don't make an investment in an area until I have spent months researching that space."- Davis Treybig
His focus is the software layer where AI and physical infrastructure collide: computing infrastructure, machine intelligence tooling, developer platforms, cybersecurity, and the databases and data pipelines underneath. These are not the consumer bets. These are picks-and-shovels plays from someone who can read a systems architecture diagram the way a doctor reads an X-ray - and sees what everyone else skipped over.
* * *
Origins
Austin to Palo Alto, the Long Way
Davis Treybig grew up in Austin, Texas, absorbing his father's stories about Silicon Valley startups over the dinner table. That early curiosity - not just about technology, but about the companies being built around it - planted something that took a decade to fully surface.
At St. Stephens Episcopal High School in Austin, he graduated as valedictorian. At Duke University, he doubled down: a double major in Computer Science and Electrical Engineering, a perfect 4.0 GPA, the George Sherrerd III Memorial Award, and the Walter J. Seeley Scholastic Award. Valedictorian again. The pattern was consistent before the career began.
Before he finished at Duke, he was already interning at Google as an APM on the Chrome team in 2015, digging into performance monitoring and crash tooling. He came back full-time in 2016, cycling through Google Assistant developer platform, then Pixel Software - the kind of rotations that teach you what builders need before you go try to fund them.
"He combines extraordinary technical insight with deep empathy for founders."- Innovation Endeavors, on Davis Treybig's GP promotion (2025)
By 2019, he had left Google, co-authored a book on breaking into product management with fellow Google APM Alan Ni, and started his climb at Innovation Endeavors. The book - "The Product Diploma: Breaking Into Product Management Out of College" - is still available on Amazon and still circulates among aspiring PMs. It is a minor but telling detail: he documented the path he had just walked before pivoting away from it.
* * *
Craft
Thinking in Public, Investing in Private
Davis runs a Substack called "Thinking in public about topics I find interesting in computing infrastructure." The description is understated. The essays are not short takes. In June 2025, he published a 100+ slide "State of Foundation Models" deck. Earlier pieces mapped out LLMs as compilers, the UX challenges of agentic systems, and why the biggest bottleneck for LLM startups is interface design - not model quality. He called that last one in November 2022, before most people were arguing about it.
The writing is not marketing. It is his research process made public. Each essay represents a space he is stress-testing. Some become investments. All become sharper frameworks he can bring to a founder conversation. His Medium archive runs back to 2021 and traces the intellectual arc from MLOps infrastructure to foundation models, data stacks, and developer tools with unusual consistency.
His GitHub, under the handle DavisTrey, still has a 2D physics engine from a Duke software design class. It also has "ExperimentationResources" - a curated collection of writing on building A/B testing platforms. He built that repository years before leading the Eppo Series B in 2024. Eppo is an experimentation platform.
He co-hosted a podcast episode on investing in MLOps in February 2022, two years before the Eppo check. He paneled at TEDAI San Francisco in 2024 on computing infrastructure and ML applications. The public work and the private work are the same work, just running at different speeds.
* * *
Portfolio
Where the Checks Go
His investment sweet spot is $10M, with checks ranging from $1M to $20M at pre-seed through Series A. He gravitates toward "highly technical teams pushing the frontier of what is possible across computer science systems" - teams that combine deep systems expertise with design intuition. Not just engineers who can ship; engineers who understand why the interface is the product.
Current board seats span Earthly Technologies (fast, repeatable build systems), Bauplan (data versioning and pipelines), Dosu (AI for developer workflows), and Mesa - each a bet on the infrastructure that developers will depend on as AI workloads become routine. He has also been deeply involved with Eppo (experimentation platforms), Augment, Panther, Continual, and Modyfi.