Fireworks AI, recognized on Forbes' list of next billion-dollar startups, is breaking new ground under CEO Lynn Chao's leadership. Their generative AI platform acts as an AI infrastructure company, empowering application developers and product engineers to create disruptive, next-generation ideas without the worry of setting up infrastructure, significantly shortening the typically long setup process for AI capabilities from five years to five days.

generative AI is changing the landscape of technology by generating human-equivalent or superior content. Fireworks AI aids in creating innovative solutions, whether for startups or digital natives, from generating texts and codes to forming complex business workflows. Distinct from companies like OpenAI, Fireworks AI emphasizes practical, enterprise-focused solutions like high customization, cost-efficiency, and low latency, tailored to meet the specific needs of businesses.

Main takeaways from the video:

Please remember to turn on the CC button to view the subtitles.

Key Vocabularies and Common Phrases:

1. disruptive [dɪsˈrʌptɪv] - (adjective) - Causing or tending to cause disruption; unsettling or innovative. - Synonyms: (disturbing, revolutionary, innovative)

Let's start super high level. What is fireworks AI and what does it do? Right? So fireworks AI provide this generate AI platform. We are AI infrastructure company to help the application developers and product engineers build next generation disruptive ideas.

2. infrastructure [ˈɪnfrəˌstrʌkʧər] - (noun) - The fundamental facilities and systems serving a country, city, or area, as transportation and communication systems, power plants, and schools. - Synonyms: (framework, base, foundation)

So fireworks AI provide this generate AI platform. We are AI infrastructure company to help the application developers and product engineers build next generation disruptive ideas.

3. generative [ˈdʒɛnərətɪv] - (adjective) - Capable of producing or creating something. - Synonyms: (creative, productive, formative)

What is generative AI and how would you explain its purpose to people who haven't been following the industry?

4. hyperscalers [ˈhaɪpərˌskeɪlərz] - (noun) - Large-scale companies that focus on cloud computing and data storage. Their services support scalability of global internet, social media, and applications. - Synonyms: (cloud providers, tech giants, mega-companies)

So me and my co founders, we have been working at hyperscalers for a long time, specifically at Meta and Google.

5. innovative [ˈɪnəˌveɪtɪv] - (adjective) - Featuring new methods; advanced and original. - Synonyms: (new, novel, fresh)

We are AI infrastructure company to help the application developers and product engineers build next generation disruptive ideas, innovative ideas.

6. viable [ˈvaɪəbl] - (adjective) - Capable of working successfully; feasible. - Synonyms: (feasible, achievable, possible)

If you lose money at small scale and then scale like 1000 times, that means you're going to bankrupt soon. You're going to make sure you have a viable business.

7. latency [ˈleɪtənsi] - (noun) - The delay before a transfer of data begins following an instruction for its transfer. - Synonyms: (delay, lag, wait)

So latency is hyper important.

8. proprietary [prəˈpraɪəˌtɛri] - (adjective) - Owned by a private individual or corporation under a trademark or patent. - Synonyms: (exclusive, patented, restricted)

I have my proprietary data, I want to leverage that to customize.

9. venture capitalists [ˈvɛntʃər ˈkæpɪtəlɪsts] - (noun) - Investors who provide funds to startups or small businesses with the potential for substantial growth. - Synonyms: (investors, financiers, funders)

So your strategy effectively is to raise money from venture capitalists in Silicon Valley, but also some of the biggest, boldest tech companies in Silicon Valley

10. scalability [ˌskeɪləˈbɪlɪti] - (noun) - The capability of a system, network, or process to handle a growing amount of work, or its potential to be enlarged to accommodate that growth. - Synonyms: (elasticity, expansibility, flexibility)

The interactive part is critical, part of product experience. So latency is hyper important. So we as a company have a deep expertise in reflected in our product offering that will provide extremely low latency, high cost efficiency for business viability and high quality through deep customization.

Disruptive Innovations By Fireworks AI Building the Future

Hi everyone, I'm Maggie McGrath, senior editor at Forbes. AI startups dominate this year's list of the Forbes next billion dollar startups. And one of those startups is Fireworks AI. Here to explain what her company does is co founder and CEO Lynn Chao. Lynn, thank you so much for being here.

Hey Maggie, I'm so happy to be here. Thanks for having me. So fireworks AI is on our next billion dollar startups list. Congratulations. Let's start super high level. What is fireworks AI and what does it do? Right? So fireworks AI provide this generate AI platform. We are AI infrastructure company to help the application developers and product engineers build next generation disruptive ideas, innovative ideas, and they can run miles ahead while not worrying about how to set up infrastructure to get there faster.

Let's start with another definition, or continue with another definition. What is generative AI and how would you explain its purpose to people who haven't been following the industry? Generate AI is a completely new technology wave that doesn't exist before. It basically can generate the content at the quality, or sometimes even better quality than human being can generate across various different areas. Of course they can generate. We are very familiar with chat GBT. It can generate text. And that capability has been used for many applications to build all kinds of assistance, medical assistance, to address shortage of nurses and doctors, legal assistance, to address shortage of lawyers and, and so on and so forth. It can also generate code where that is the high end of the work, to kind of write software, to build all kinds of apps and products. It can also generate business workflows to make the business operation more efficient. It can generate other interesting application languages like SQL for data analysts across the board. That's why it's a disruptive technology. Whenever there's emerging disruptive technology, it pushes a huge tidal wave of the whole industry transformation. For example, we have been through cloud first transition from on prem to running on cloud, extremely flexible infrastructure. That's a massive transformation. A lot of business opportunities to later on mobile first where we do not have to tie with our desktop to do our work. We can communicate, interact anywhere, wherever we go through our mobile phones. Now to this general AI technology where the content generation or any generation of cogeneration, of image generation or video generation or content understanding can all be powered in an unprecedented way in our everyday life.

So I expect a much bigger tidal wave that's going to come, that we're already in the process of going through a super exciting time, a much bigger tidal wave. It feels like we're already in the middle of a tsunami I can't imagine it getting much bigger. But you launched fireworks AI in 2022 and just take us back. What was the business opportunity that you saw at the time? And what convinced you that a company like this needed to be built? Right? I can start a little bit from our origin story.

So me and my co founders, we have been working at hyperscalers for a long time, specifically at Meta and Google, where we have been building AI infrastructure to power the entire company doing their AI first transition. As the industry is going through right now. Through that journey, we enabled a lot of interesting applications and specific use cases, high impact, all the way from ranking recommendation, that's every day in our digital life, to content integrity. Hey, we don't want bad content to appear. And the bias of our view to all kinds of fun, interactive AI effects. As if this is a family conversation. I can wear interesting hats, or there could be bubble or the coping translation if we were in a game, working with players from multiple nationalities and so on, so forth. So AI is literally everywhere in our day to day life, and we have been through that journey, how to scale the infrastructure to provide the best performance, best cost efficient, best quality, and enable those product features.

When we are doing that, we also observe across the industry, as the industry is catching up and also going through that wave, it is, there's excruciating pain because they don't have a team like us to help them move forward faster. They don't have the right hardware, they don't have the right software. And we just see we are very impact driven and we see huge amount, even bigger impact if you can start this company and help the whole entire industry go into this process, we kind of enabled the hyperscalers to today's point through a journey of five years with thousands of engineers.

Our vision is to enable app developers, product engineers in the industry, to get to launch their newest ideas on this infrastructure, not five years, but five days, and without an army of machine learning engineers and infrastructure engineers supporting them.

So that's how I missed it. So I just want to emphasize that. So for a company that realizes they need to build out their AI capacity, it could take up to five years to accumulate the hardware, the software, the manpower and fireworks. AiH basically gets that five years down to five days.

Yes. So who are the companies that are calling you for help? Who do you count as clients? Right. So that's another interesting story, because in my mind, I think as I start this journey from the go to market point of view, I would say, hey, this is a new technology wave. So the most tech forward me will be my first wave of customers, and then the more tech advanced but less forward, and then the kind of more conservative ones. So that's kind of in my mind, that sequence and that maps to the startups, most tech forward, most bold, most aggressive to the digital native enterprises, where they have a very strong engineering team. They are also kind of embracing new technology to traditional enterprise where they want to wait a little bit, see how things play out and so on.

Right? So that's kind of. But right now we're working across both with every one of them, with all these segments for DNAI and the startups are using us to create brand new product experience that never existed before. Digital native companies are using us to revive and bring new interesting spin to their product experience. And traditional enterprises are using us to improve their productivity across the board, in their own task force, or with their other engagement where they need a lot of human support. So it's super exciting time that we see the power of this technology get adopted and leveraged in all different corners of businesses. Interesting.

So when you say digital first, my mind goes to someone like a doordash. Is that a potential customer? That's right, company like e commerce or like marketplace. Many of those examples are ideal customers leverage technology. Now obviously there's a lot of talk about OpenAI, but why is fireworks AI the better, in your view, enterprise solution than something like an open AI for getting startups and mature companies to where they need to be in terms of generative AI and building out that technology? Right. So OpenAI is a great company I really admire and they push this wave of awareness of, hey, this new technology is really, really powerful. Why? Their focus is AGI, which is very admirable goal, but at the same time our focus is we want to bring practical value to the industry. We want to solve practical problems where the enterprise cares most about.

So I think from the beginning, our anchor point of focus are very different and to enterprise they really care about not kind of the AGI side, but more like, hey, I have a specific business task where fit into my specific proprietary data distribution. I have my proprietary data, I want to leverage that to customize. So customization is a key point of enterprise adoption of JNAI at the same time, because most of those JNI technology will power b two c or b two developer facing application. So it can scale quickly. When it scale quickly. A critical part of business metrics is people don't want to lose money or bankrupt quickly. If you lose money at small scale and then scale like 1000 times. That means you're going to bankrupt soon. You're going to make sure you have a viable business. So cost efficiency is very important. And last but not least because it's consumer facing and developer facing, it needs to be highly interactive.

The interactive part is critical, part of product experience. So latency is hyper important. So we as a company have a deep expertise in reflected in our product offering that will provide extremely low latency, high cost efficiency for business viability and high quality through deep customization is your proprietary data. So that's our anchor where our Yden strategy, product strategy and the product offering is drastically different from OpenX product offering.

Interesting. So do you find that the inbound interest from startups and other companies, is it word of mouth or are you having to go out and recruit in marketing and tell your story so that startup founders and company CEO's know that you exist? Right. So we have our self serve platform. So the purpose of that is we want to have people just onboard without even needing to talk with our sales team. They can get first hand experience and taste of the product and they can scale their business at the first our self serve platform and then talk to us when they need a lot more from our platform. So I think that is doing really well for us as in this help us attract more developers and more startups and even some enterprise developers. They can get a sense and build prototypes and explore first and then go into production with us.

It feels like all technology can grow and evolve at a fast pace, but nothing more so than AI. So what is it like to found and grow a company within this space? How do you navigate the changing technology on a day to day basis? Right. So this is a really fun time. I would say the market is very different. The market is moving at extremely fast speed compared with any other market. I think that's the biggest challenge we have seen so far is as I mentioned before, I would imagine in my mind just a sequence of adoption, market adoption, and that's not the case. My assumption was wrong, it's happening everywhere. So that, that's why we are making sure that from product offering point of view we take care of a broad spectrum of customers from the developers through self serve platform and enterprise customers through our enterprise facing platform.

There's a lot of overlap because at the end they're all developers, individual developers or developers within enterprise using us. But there are also many differences and we as a company we are moving really fast, like taking out both side of the business and in the customer needs and we are also extremely customer obsessed. The critical part of the company culture is we believe only if our customer is successful and will be successful. So we'll do all it takes to make sure our product design, our product features is best geared towards the productivity business enablement for our customers.

Now, Fireworks AI has raised about $77 million from investors. What have those pitch meetings been like? How receptive is Silicon Valley to your specific solution? Right, so we are, we have the best investors to support us, benchmark and sequoia. So I'm extremely happy with the fundraising part. I think our team carries pretty strong credibility and has shown our capability of building very strong technology product and driving go to market. So that's why the fundraising part is a reflection of that. And the partnership with both leading investors has been phenomenal and we are really, really great partners, especially since the early journey of the company. They have been providing a tremendous amount of support and I'm extremely thrilled to work with them.

On top of that, we also have industrial investors, including Nvidia, AMD, MongoDB, Databricks, Snowflake and so on. So that is a reflection of our partnership strategy beyond. And we want to drive co development of a bigger product or ecosystem building with those other leaders in the industry to help our customer get the best advantage out of various different product offerings, from hardware to data systems platforms to vector search research and so on.

I see. So your strategy effectively is to raise money from venture capitalists in Silicon Valley, but also some of the biggest, boldest tech companies in Silicon Valley. And that in turn will help fireworks AI and you build out your hardware, your software and your overall capabilities.

That's right. I think raising money is, is one angle, but more actually even before raising money, we have already been working together very closely.

Interesting. Now you mentioned that you and your team have strong backgrounds. You were at Meta before you helped start this company. Talk to me a little bit about how your time at meta helped fuel what you are building at fireworks AI. Are there lessons or skills that you took from that job that you're applying to this one? Definitely a lot. So first of all, in terms of product design, I have the fortune to be the head of Pytorch at Meta and Pytorch is now the dominating AI platform across the industry. It get to that point because it has very opinionated product view.

As in AI is a highly dynamic field, a lot of innovation in this space. The way to do that is create extremely simple, easy to use interface to the users of the Pytorch framework, but hide all the complexity in the backend to enable productionization. So when we start this journey, it almost feels like mission impossible, because the design of extremely easy to use interface and highly sophisticated enablement of production feels like the intersection is zero and we're just thinking about like something that cannot be achieved at all, while the goals looks nice. So it took us a while to figure out how to do it. We made many mistakes, and now we're getting there because of this design philosophy.

So when we think about fireworks, we use the same design philosophy, but in a completely different domain. So first of all, in the JNI space, I mean, the challenge here is, as I mentioned, the first level challenge is latency, cost efficiency, because the JNI models are the biggest, most complex, deep learning models in the whole entire world, and we have to solve that problem. Otherwise, the business is really hard to build on top of this technology. So that's the first tier. And second is we also realized the individual model is not enough to solve a complex business task in a lot of time.

It requires multiple models. Multiple models. Each is an expert. It's like human being. Each is an expert in its own domain to come together. Solve analogy is when we have a complex health situation, we sometimes require multiple doctors in its own specialty to give opinions and come together to discuss the truth. Similarly, in our real life, we have been talking a lot about large language models, texting, text out. But just as we are talking now, we are not communicating through texting each other. We are communicating through audio and also videos, actual frames of images. Right? That's our real life day to day.

So the content generation space cannot just be dominated by logic language model, and also need to cover multiple modalities, and also, even within one modality, multiple models. Each is an expert in its own area to best solve an industry problem on top of that, even with that is not enough, because individual models are limited in its own knowledge. It's like individual human is limited its own knowledge too. For model, it's limited because it's training data.

The data we use to train the model fuse the knowledge into the model. It's finite, not infinite. There are a lot of knowledge in the world that's not living inside the model, but in other places, for example, behind a lot of APIs, we can get real time information about weather, about stock price, about news from Forbes, for example, and so on and so forth. So we are also building a compound AI system on top of the fast inference engine as the foundation, the compound AI system. The goal is to blend all this different knowledge from many different models, different modalities and APIs to coordinate together solve a complex business task.

Now it sounds really complicated, how do we even do that? It's a very hard problem. So now carry on the Pytorch design philosophy. We're going to build a very simple API for user to use our system and hide all the complexity I just mentioned in the backend. Now what is that API? The API is prompt, it's prompt. So the same as you talk with chat GPT, the same as our conversation. We use that as our API to drive the execution of our compound edge system.

So that's our vision. And we are going to share more with our audience, with our developers and product engineers very soon. About the new releases. I normally ask founders, if I talk to you a year from now, what do you hope to be able to tell me your company has accomplished? But it sounds like a lot of that mission impossible, complex back end, simple front end products that you're building might be a little bit under NDA right now. But is there anything you can tell me that if we were to speak one year from now that you and or fireworks AI will have accomplished? Right.

So I think it comes back to we are customer obsessed and I really within a year I think we're already making a lot of great progress already. So probably it will happen sooner. Much sooner than a year is to onboard across the board, from startups to digital native enterprise, to traditional enterprise companies, onto our new compound air system, powered by our fast and most cost efficient inference engine to solve their business problem in the fastest way and the most cost efficient way. So that's my dream. And I think today there are already many customers using us and in a year I would like to ten x. That customer is in a year ten x. Well in the meantime, you're on the next billion dollar startups list. Congratulations. Lynn Chow, thank you so much for joining us and explaining to us what your company does. We so appreciate your time. Thank you.

Technology, Innovation, Entrepreneurship, Generative AI, Startups, Infrastructure