Are LLMs plotting to drain your startup bankroll?

In partnership with

If you are an A.I. startup you likely have a burning hole in your bank account.

That crackling sound?

It’s your LLM. Every token processed by a large language model (LLM) sends dollar signs flying. NVIDIA Gods rejoice.

Startups are facing a common villain: their own AI models. Fine-tuning and deploying behemoths like GPT-4 can feel like watching your startup’s bankroll evaporate in real time. Sure, they’re powerful—but boy, are they expensive.

Enter Proxis, a Y Combinator Summer 2024 startup with a mission to stop the cash burn madness and give startups a break from the GPU inferno. 

The Growing Problem: AI Models Eating Your Budget Alive

Imagine this: Your AI startup is soaring, investors are excited, and your models are delivering some serious magic. But lurking behind the curtain is the real monster—the enormous cost of keeping these models running.

Fine-tuning frontier models like GPT-4 on proprietary data is 100x more expensive per token than smaller, more efficient models. And it’s not just about money. These giants are 10x slower than leaner, meaner alternatives, slowing down your operations and leaving you with a GPU bill that could finance a small country.

The worst part? You’re often stuck with closed-source model providers, which means you can’t even tinker with or tweak your own model. You’re locked in, and as costs skyrocket, your margins vanish. Not exactly the scaling dream you pitched to your investors, right?

 Proxis to the Rescue: Distilling LLMs, Slashing Costs

Enter Proxis, a startup that’s basically handing you the cheat code for this AI game. With Proxis, you don’t have to throw money at oversized models or break the bank with costly GPU infrastructure. Proxis leverages a little magic called model distillation—a fancy way of saying they take massive, slow models and boil them down to smaller, super-efficient versions that get the job done for a fraction of the price.

“Like Morning Brew (but for venture capital)”

Join 15,000+ VCs and investors getting smarter on venture capital by reading this free daily newsletter.

Start getting the headlines that matter, lists of companies breaking out, and other resources and playbooks you won’t find anywhere else.

Become a smarter investor in < 5 min a day.

Subscribe for free

Think about it like this: Instead of relying on GPT-4 to answer your AI needs, Proxis helps you distill Llama 405b into a smaller, turbocharged version that works 5x faster and costs 1/10th of your current setup. And guess what? It’s open-source, meaning you have full control. No more being tied to an expensive, closed-source provider who holds the keys to your AI kingdom.

How did Proxis figure out how to disrupt the AI matrix?

Credit goes to the brains behind the operation:

- Liam Collins (CEO) brings serious expertise in building zero-to-one systems from scratch. With a background that includes investment banking and software engineering, Liam knows how to build systems that can scale without burning through cash like a GPU on overdrive. He left his MBA program at Wharton to go all-in on Proxis.

- Jackson Stokes (CTO) is the tech wizard who optimized the Gemini model at Google for efficient deployment at a massive scale. His experience in making AI models run like a Ferrari instead of a tank is the backbone of Proxis’s model distillation success.

 The Market is Ready: LLMs Are Booming, But So Are Costs

Let’s face it: Llama models are exploding in popularity, with usage growing 10x in the last year.

While these open-source models offer a breath of fresh air, using them effectively without wasting money on bloated setups has remained a challenge—until now. Proxis is stepping in at just the right moment to help startups manage LLM costs while staying competitive.

Gone are the days when you had to throw money at the biggest, most powerful models just to stay relevant. Now, with Proxis, you can use distilled models that are just as good for your needs but at a fraction of the cost. And did we mention they’re 5x faster?

 The Takeaway: Don’t Let Your AI Burn a Hole in Your Wallet

As an AI startup founder, you’re already juggling a million things—burning cash on expensive LLMs shouldn’t be one of them. Proxis gives you the power to distill the best open-source models into efficient, cost-effective solutions, helping you stop the GPU cash burn and start scaling smarter.

Ready to stop watching your bankroll go up in GPU smoke? You can sign up for the waitlist today and take the first step toward affordable, efficient AI that won’t break your budget.

Shameless Pitch Opps.ai

Get data on your ideal customer or investor.

Tell Valet to “Create a Campaign Targeting by Ideal Customer [Geography, Industry, Title, Company Size, Title of Contact]

Valet our A.I. Agent researches your data and builds your campaign turning days of work into minutes.