Today, Madrona is thrilled to announce the 2023 Intelligent Applications 40 — the top private companies leveraging artificial intelligence and machine learning. This year, we received over 350 nominations from over 60 venture investors across 50 top-tier venture and corporate investment firms.
It is hard to believe how much innovation has occurred since we announced the 2022 IA40 list. From the release of ChatGPT to the proliferation of new foundation models like GPT-4 and Llama-2, we have witnessed a new era of applied AI. A year ago, Madrona hosted the inaugural Intelligent Application Summit. There, we explored the nascent landscape of emerging generative-native AI applications, which were just beginning to harness the potential of foundation models. We also predicted and witnessed the emergence of application frameworks with the launch of LangChain the week of the summit. Since its launch, LangChain has captured the minds of millions of developers, standing as one of the fastest-growing open-source projects in history – it's no surprise they were one of the IA40 2023 winners this year!
We will celebrate the 2023 IA40 winners and debate where applied AI is headed at the second annual Intelligent Applications Summit on October 10th and October 11th in Seattle hosted in partnership with AWS, Microsoft, NYSE, McKinsey & Company, and Pitchbook. You can find more on our event site or email firstname.lastname@example.org.
Highlighting how dynamic the field of applied AI/ML is, there are 29 first-time winners on the 2023 IA40 list! Only seven companies have been selected as IA40 winners all three years, and four others have won twice. Two foundation model players made the list this year — OpenAI and Anthropic. OpenAI not only made the list for the first time — but the company has also raised the most capital of any company to date — $11.3B. While unsurprising, the absence of any foundation model companies in previous years' lists indicates the massive shift we have seen over the last 12 months as these models have become generally available and used by builders.
Companies on the IA40 list have collectively raised $24.5B from inception to July 31, 2023 – the cut-off for tabulation. Looking at just this year through September 20, 2023, they have raised over $13.7B, which includes OpenAI's $10.3B round. Funding activity in the Generative AI space has shown no signs of slowing down this year. According to Pitchbook, in the last 8 months, more than $14.5B of capital has been invested in over 150 Generative AI companies. Recent weeks have continued to demonstrate the heightened enthusiasm in the AI/ML space, exemplified by Amazon's $4B investment in Anthropic, Databricks' announcement of their Series I Round, raising over $500M at a $43B valuation, and early-stage IA40 winner Poolside reportedly raising over $100M.
Funding data categorized as of 7/31/2023
This year, we also introduced 5 Rising Star Enabler Winners. These five companies represent Enablers that have raised less than $50M of capital and were top runners-up to the list.
This year's list is filled with Gen-native AI applications, which we define as those built on top of one or multiple foundation models and can generate new content without explicit instructions (e.g., MidJourney, Numbers Station, Harvey). These companies have emerged in the last few years thanks to major advancements in FMs, new application frameworks, and other tools enabling developers to build. They generally leverage vector databases and embeddings and boast multi-modal generation and other copilot-like capabilities. These Gen-native companies are taking a fresh perspective on a customer problem and are looking to find new ways to innovate and speed up workflows. We expect to continue seeing more Gen-native AI applications being built in the years to come that will displace previous incumbents.
We define Gen-enhanced companies as high-growth, rapidly scaling players that did not begin life as Generative AI companies but are quickly enhancing their core products with GenAI and FMs. Many IA40 late-stage winners (e.g., Canva, Notion, Airtable) are examples of Gen-enhanced companies. These companies are proving they can integrate generative capabilities into their increasingly intelligent applications and have advantages with engaged customers, existing workflows, and relevant data. They have the distribution, brand recognition, and network effect over the Gen-native companies. An interesting battle is brewing between the Gen-native and Gen-enhanced that will continue to play out in the months ahead.
This year, IA40 winners from the early stage through the late stage are harnessing foundation models like GPT-4, Stable Diffusion, Anthropic, and Llama-2. In contrast, last year, only a few early-stage and mid-stage companies had started adopting foundation models, mainly relying on closed-source models such as OpenAI's. Late-stage winners primarily leaned on internally developed models tailored for specific tasks. However, this year marks a significant shift, as nearly every company now opts for off-the-shelf foundation models due to their improved performance. Additionally, there's a notable uptick in companies starting to tinker with open-source models. Open-source models offer transparency, customizability, and cost savings compared to closed-source counterparts. Often, companies do not require the extensive features provided by closed-source models, making open-source models equally favorable — especially given increased model performance. We expect to see more companies leveraging open-source models and tooling to build commercial products in the year ahead.
Navigating the world of applied AI presents numerous challenges today as the entire ecosystem continues to evolve. Developers often grapple with integrating, testing, and deploying models and knowledge sources in production. Novel application frameworks are emerging to streamline tasks such as model integration, fine-tuning, retrieval augmentation, prompting, and monitoring. Notably, companies like LangChain (winner) and LlamaIndex (rising star) were non-existent a year ago but have swiftly become essential to the developer community. Given the inherent complexities in turning foundation models into intelligent applications in production, we believe that enabling technologies like LangChain, Pinecone, and Unstructured.io will continue to empower developers to explore models, knowledge sources, and agents and test what works best for their customer use cases.
Companies on the IA40 list have collectively raised $24.5B from inception to July 31, 2023 – the cut-off for tabulation. Since then, they have raised an additional $961B, making a combined total of $25.4B raised, which does not include the Anthropic announcement this week. Only one early-stage winner, MidJourney, has raised no outside capital. Meanwhile, OpenAI has raised close to $12B of capital. It is also interesting to note the median last known valuation of the winners, with early-stage winners raising at a $83M median, mid-stage at $575M, late-stage at $5.7B, and enabler-stage at $4B.
IA40 companies such as ElevenLabs, Synthesia, Descript, Wellsaid Labs, and Cresta all leverage speech-to-text or text-to-speech technologies. Earlier this year, we also saw IA40 winner OpenAI announce WhisperAI, an automatic speech recognition system trained on 680,000 hours of multilingual and multitask supervised data. While some of these technologies have been around for a while, they haven't been able to reach their full potential until recently. Many of these companies are inherently multilingual and bring the power of voice to any program or platform, revolutionizing creativity across all industries, from media to gaming to podcasting to customer success and more.
At the app layer, an increasing number of the IA40 winners are leveraging open-source models (e.g., Midjourney, Runway, StabilityAI leveraging Stable Diffusion). Open-source models are smaller and run on a company's dedicated instance, resulting in lower end-to-end latencies and allowing for more fine-tuning and adjusting. Companies like Numbers Station and Typeface are also adopting more of a "model-cocktail" approach by mixing open-source and closed-source models, calling upon specific models for different use cases. Historically, for many companies, every use case of new output would have required separate models, but now companies can use zero-shot diverse generation via FMs.
Foundation models have shown emergent capabilities of complex reasoning, reasoning with knowledge, and out-of-distribution robustness, none of which are present in smaller, more specialized models. As a result, the most significant shift this year is the number of companies leveraging these FMs. OpenAI and Anthropic each started as research companies and are perhaps two of the most prolific and cutting-edge model providers. We expect more foundation models will be winners in future years, including a few primarily open-source model companies.
In 2021 and 2022, the IA40 enabler winners were primarily focused on the traditional modern data stack – solving problems around storing, cleaning, and ETLing data to perform analytics. In 2023, the main themes center around having a data-driven approach to AI. While FMs are powerful, they are limited to reasoning about the facts they were trained on. In 2023, many IA40 enablers focus on mechanisms for operationalizing and connecting data. Companies such as Pinecone, LangChain, Databricks, and LlamaIndex (a rising star) are integral in allowing developers to connect their proprietary data and achieve more personalization.
Open source plays a few key roles across the IA40 2023 list. Open-source models are being deployed and implemented across AI applications for the first time — RunwayML and MidJourney leverage Stable Diffusion. Companies like Hugging Face also continue to play a crucial role in the open-source landscape by providing the latest and greatest open-source model architectures, making these models more accessible and understandable to developers. Developers can tweak the weights, tune the algorithm, and even train the model to run on their data so the model can be more vertical-specific for their use cases. Beyond open-source model deployment, companies like LangChain and LlamaIndex have become the developers' open-source haven for building and deploying intelligent and generative apps. Several other IA 40 enablers like Neon, dbt Labs, and WhyLabs (rising star) also leverage open-source components to build community and broaden distribution.
Last year, we anticipated many of the late-stage companies would engage in M&A, and this year, the trend played out with Databrick's $1.3B acquisition of MosaicML. We expect this trend to continue, but we wouldn't be surprised if a handful of the growth and enabler companies were to go public in 2024!
At Madrona, we have been partnering with and investing in intelligent application companies for the last 10 years and watching their impact across many verticals. These applications will define the future of software and the next generation of computing, and we believe they deserve to be known and recognized!
You can read more about the process and methodology here.