Welcome to Built By Berkeley, where we announce all the funding rounds by Berkeley-founded companies. This is a community effort, so please let us know if we missed a company here. 🐻
Data is the fossil fuel of AI: There has been a lot of talk over the last few weeks that the scaling laws around pre-training may be slowing down. Ilya Sutskever, one of the original co-founders of OpenAI, came out last week with a talk focused on this - full talk here and what it means. Ilya recently raised $1B for a new AI company called Safe Superintelligence.
What is the scaling law?
This is the idea that AI model performance will improve as you increase key parameters like compute, data, and model size. See a description of this by the founder of Anthropic.
What is pretraining?
Pretraining is where you take huge amounts of data to train an LLM and help it make connections between different parts of the data to predict the next word in a sequence.
Why might pretraining be slowing down and what does this mean?
Data to pretrain is effectively finite, hence the idea that data is a 'fossil fuel' - once it has all been used, there is unlikely to be more unless synthetic data proves useful, though this remains unclear currently. A lot of the easily available data, such as the open internet, has already been utilized, although large institutions such as JP Morgan have massive amounts of non-public data still.
If pre-training becomes less effective, it doesn’t mean AI slows down it just means AI companies are going to need to find other ways to improve their models, and this has wide-ranging implications:
Agentic Frameworks/Agents: Instead of purely relying on pre-trained models, companies are exploring frameworks where multiple models act as agents (interconnected systems that interact with the world or each other). This could lead to improved reasoning, problem-solving, or specialized workflows without needing to continually scale model size or data.
Inference Time Strategies: This refers to fine-tuning how models process inputs in real time (dynamic reasoning) rather than focusing purely on raw pretraining or can be called scaling test time compute. This makes outputs better by iteratively improving predictions as they process information, like running a thought experiment multiple times. This is already occurring with ChatGPT-4o.
Frontier Models Leads Quickly Erode: If pre-training becomes less effective, the edge that frontier models such as OpenAI have might get quickly eroded by small teams improving on open source models such as Meta's Llama models.
This is amazing for startups as it means that you can probably be less worried that a new model will eat your lunch through pre-training, and you can focus on building amazing applications on top of the available models.
To me, the phrase ‘Arming the Rebels’ seems to be what is happening here. The large model providers have provided amazing tools through huge investments in pre-trained models, and now the rebels, the 2 to 10-person team companies, are going to take these tools and build amazing things.
Where I think this will end up being extremely exciting:
In Search of the Small TAM: For a long time, conventional wisdom has favored either creating a new TAM (e.g., Airbnb, Uber) or targeting massive TAMs. However, innovations in AI and the reduction in costs to address a market are now unlocking smaller TAMs that can still be viable at a venture scale.
Domain Expertise to Prevail: While building software has never been easier, creating products that genuinely work for buyers remains challenging. Entrepreneurs with deep domain expertise will have an edge, leveraging the growing AI infrastructure to deliver compelling value propositions.
Small Teams & Low COGs: Small teams are poised to build massive businesses by utilizing modern solutions that streamline software development, customer acquisition, and operational efficiency.
Although a word of caution here - this space evolves so quickly, and this new potential paradigm is only a few weeks old and may not hold, with many believing the scaling laws will continue to hold and many betting multiple billions of dollars on this in areas such as inference time compute and synthetic data.
Summary by the #️⃣ & 💰:
18 Berkeley-founded companies funded
$1,403M of capital raised from the 9th December to 15th December
💡 Got any ideas or feedback on how to improve this weekly digest? Just hit reply.
Acquisitions
🧬 Nvelop Therapeutics. Merger 🇺🇸 Genetic medicine developer. 💰 Undisclosed.
🔬 Sharp Therapeutics. Merger 🇺🇸 Targeted biosensors for drug discovery. 💰 EVP Capital.
🐻 Marcel Bruchez, Co-Founder & CTO. PhD Chemistry. Article.
Closed Rounds
⚡ Intersect Power. $800M Undisclosed 🇺🇸 Utility-scale renewable energy developer. 💰 Alphabet, The Rise Fund, Trilantic North America.
🐻 Sheldon Kimber, Co-Founder & CEO. MBA Haas. Article.
🔗 Ayar Labs. $155M Series D 🇺🇸 AI chip interconnect optimization tools. 💰 Advent Global Opportunities, Atreides Management, Founders Fund.
🤖 Tractian. $120M Series C. 🇺🇸 AI Manufacturing Copilot for Industrial Organizations 💰General Catalyst, Sapphire Ventures, Next47
🐻 Igor Marinelli, Founder & CEO. Engineering. Article
📸 Luma AI. $90M Series B 🇺🇸 Multimodal AI video platform. 💰 Amplify Partners, Andreessen Horowitz, General Catalyst.
🏭 Zetwerk. $70M 🇮🇳 On-demand manufacturing for precision parts 💰 Khosla Ventures, Baillie Gifford
🐻 Amrit Acharya, Co-Founder & CEO Article
🧠 Cala Health. $50M Series C 🇺🇸 Wearable neuromodulation devices for tremor releif. 💰 GV, Lux Capital.
🐻 Kate Roenbluth, Founder & President. PhD Bioengineering. Article.
📹 Truvideo. $40M Undisclosed 🇺🇸 Video platform for car repairs. 💰 Bain Capital Ventures, TZP Group.
🐻 Douglas Chrystall, Founder. Entrepreneurship.
💪 Bsport. $31.6M Series B 🇫🇷 Fitness studio management platform. 💰 Base10 Partners, Octopus Ventures.
🐻 Marc Capelo, Co-Founder & CPO. MS Systems. Article.
💵 Avanti Finance. $14.2M Series B 🇮🇳 Digital financial inclusion platform. 💰 Bill & Melinda Gates Foundation, Rabobank Group.
🐻 Vijay Kelkar, Co-Founder & Board Member. PhD. Article.
🚀 Hyperbolic. $12M Series A 🇺🇸 Open-source AI GPU Cloud. 💰 Alumni Ventures, Blockchain Builders Fund, Polychain Capital.
🐻 Jasper Zhang, Co-Founder. PhD Mathematics. Article.
🌐 Kast. $10M Seed Round 🇸🇨 Neo-bank for stablecoin payments. 💰 DST Global, Goodwater Capital, HongShan.
🐻 Daniel Bertoli, Co-Founder. BS Business Admin. Article.
👁️ Okogen. $3.3M Series B 🇺🇸 Ophthalmic disease treatments. 💰 Brandon Capital.
🐻 Eric Daniels, Co-Founder & COO. AB Molecular Biology. Article.
🧵 Rubi. $1M Undisclosed 🇺🇸 Carbon-negative biodegradable textiles. 💰 Kayak Ventures, Necessary Ventures.
🐻 Neeka Mashouf, Co-Founder & CEO. BS Materials Science. Article.
🏫 Varthana. $0.18M Undisclosed 🇮🇳 School loan platform. 💰 Blue Earth Capital, BlueOrchard Finance, Kaizenvest.
🐻 Steve Hardgrave, Co-Founder & CEO. MBA Haas. Article.
💳 Arro. Crowdfunding 🇺🇸 Credit-building financial platform. 💰 Bling Capital, Crosslink Capital, Global Founders Capital.
🐻 Ryan Duitch, Founder & CEO. MBA Haas.
🤖 Xplorobot. Accelerator 🇺🇸 Robotics for digital twin analysis. 💰 271.vc, Elemental Impact.
🐻 Oleg Mikhailov, Co-Founder & CEO. MBA Haas.
Built By Berkeley Tracker
Date Built By Berkeley Started | Companies Funded | Total Raised ($M) |
7/8/24 | 219 | 16,606 |
Our goal is to document the startup ecosystem of Berkeley-founded companies. Please share this newsletter with any Cal Bears in your network so we can crowdsource information about all investment rounds and job opportunities.
Did we miss a company or want to announce a round? Add it here, and we will post it next week.
Do you have a job you want to post from a Berkeley Company? Add here.
Is there someone in the Berkeley ecosystem that you would want us to do a profile on? A founder, funder, or general startup person? Add here.
Built By Berkeley is not affiliated with UC Berkeley, but maybe we will be one day if we get enough subscribers….
