Artificial Intelligence and Machine Learning. What do they mean? All you need to know - this is the ‘Dot-Com Bubble’ of the decade. Within the next few years - every piece of software and hardware will utilize some form of artificial intelligence. As investor money continues to be poured into the space, entrepreneurs around the globe seek to create the next ‘Google’ of today’s age. I’ve recently been diving into the heads of innovators around Silicon Valley as they try to build the next generation of infrastructure, tooling, or application-use companies involving AI.
I classify building in “AI Infrastructure” as the act of working directly on foundational models for broad sector application. Naturally, industry incumbents are positioned to win core development due to the immense amount of funding, time, and data required to train new generalizable models. Here, research oriented companies like Anthropic and DeepMind tend to excel. When seeking to build tooling in parallel to incumbent innovation (tooling), the space becomes more accessible to startups trying to tackle gaps in the market. Companies like Scale.AI have proven to be successful. However, for earlier stage companies looking to disrupt the space it can be a tough landscape to navigate. There is a fine line between not having a technical moat and going so niche as to build a GPT-4 ‘wrapper’. Similarly, incrementally improving compute speeds or memory problems does not constitute for a generational company. Personally, I’m excited about a few of these early-stage startups trying to disrupt the ecosystem. This article outlines how three ambitious teams look to differentiate themselves in terms of company thesis, approach, and moat. Note that all data was personally aggregated and held no association to any venture capital firms or external work.
Despite being one of the youngest and most hyped teams in San Francisco, Govind and the Automorphic team seem to be exceeding expectations. The 19 year-old dropout and his crew appear like the furthest traditional founders away from a research-based company. Yet, their work in improving the basis surrounding language models isn’t going unnoticed. They’ve already had some very prominent investors commit capital (not-disclosed) and have had early traction in building autonomous agents. Similarly, they are attracting quality inbound with their Firewall and TREX tooling, suggesting a transition to revenue generation in the coming months. I believe their success in building an unorthodox research-based company stems from two core points.
A) Firstly, Automorphic is able to attract the Bay’s top talent due to exposure and the challenging circumstances it holds. “Smart people want to work on hard problems”, Govind told me as we discussed his possibilities in adding team members. If they’re able to recruit several older experts with domain expertise, they begin transitioning into a big player in the space.
B) The second differentiation is held in Automorphic’s ability to ship products whilst simultaneously advancing research. If they are able to become revenue-positive they extend their runway substantially (if not infinitely), and are able to grow internally. By having the ability to synthesize their research into tangible products, they are able to differentiate themselves from sole-research based incumbents.
Recruitment through challenging problems and strong Silicon Valley presence
Blending approach between research-orientation and revenue-generation
Opening doors towards partnerships with incumbents: research —> product
Baseten’s motto is simply “Machine Learning Infrastructure that just works” - making it a very interesting case to look at. The ~30 person post Series-A startup isn’t even trying to hide their direct compete to giants Nvidia, Microsoft, and anyone else building base AI infrastructure… By focusing their work on deploying the best open-source models, they truly DON’T have a technical moat. What makes them confident in future success? The answer lies in team composition and intrinsic belief. On one hand, companies like OpenAI and Anthropic have been proactively hiring ‘top-talent’ across the nation. However, once gathered, it appears as a jumbled group of technical individuals. Emphasis on compartmentalization and management becomes a massive focus to ensure unity and unidirectional growth. However, the core team at Baseten is composed purely of ex-employees at Gumroad, all working together ~8+ years ago. Having the ability to build a company after working together for several years allows them to draw upon unanimous growth-curves of prior experiences. Combining their tight, domain-centered team with a few young, bright hires has created a well-oiled machine. I’ve also heard extensive feedback surrounding the strong culture they’ve built internally. It will be fascinating to see how large of a role this plays into the future of BaseTen. I can foresee the impact of quicker movement speeds and purposeful shipping being appealing to the enterprise-market. The opposite scenario is also plausible, with a better funded player throwing greater firepower towards core issues, and simply driving them out of the market. Questions have to be asked of long-term defensibility.
Closely knit team ensure one-directional movement and unanimous goals
Balance of domain experts and young hustlers ensures growth on both sides
Questionable long-term market capture
Two brilliant recent-grads from Stanford have been building in Stealth as of January 2023. Maintaining complete discretion- they are creating an evaluation platform for LLM’s in specialized use-case scenarios. The team has been taking a mature long-term approach that is rarely seen in the entrepreneurial ecosystem. Originally basing their company off a research paper in the healthcare sector, they first began building for a tiny customer segment. By narrowing down their niche target-customers, they appear to be building in a non venture-backable market. However, over the course of the next few months, they hope to ship expert evaluation models to their design partners and grow to capture 80% market share (of the niche space). By training specialized models for such specific consumers, they are preparing themselves as the market expands in the future. Instead of building/competing in an existing space they have chosen to CAPTURE a niche field, BET on the need expanding to different sectors, and REMAIN as industry leaders, as the market scales from sub ~1B to ~10B+ (estimates). By holding a macro-vision for the space and building in fluid markets, they truly are taking a unique approach. Although highly risk prone, I align with their unorthodox vision in market expansion and creating a generational company. Hope to share more when they launch to the public.
Building directly for niche design-partners and perfecting product
Capturing large percentage of niche market
Expanding outwards as market-cap increases (+ remaining #1 in niche spaces)
It’s been a thrilling journey seeing these differentiators come to life. Only time will tell. What do you think of their long-term validity?