Flybridge AI Index: July 2024 Update
We thank Aiera, an AI-powered equity research platform, for accelerating our data and insight extraction process.
Summary:
In July, the Flybridge AI Index had a return of -5%. For the prior 12 months the Index had a return of 24% to its current level of 116% since January of 2023.
Of the 31 companies in the Index, 8 gained in July and 23 declined in the month. Significant gainers include IBM (+10%), and Nice Systems (+7%). Significant decliners include Astera Labs (-24%), and Micron Technology (-16%).
Massive month for model releases with Meta's newest series of models, including their 405B model, Mistral releasing their 123B instruct model, and Google's small 2B model.
The Q2 earnings season began with 10 companies reporting during July.
Earning season
As we enter a new earnings season, the AI index experienced its first monthly decline since its launch in May of this year. Interestingly, this decline occurred despite an increase in the average YoY revenue growth from 28.9% last month to 30.5% in July as new companies reported.
The primary driver of the index's decline was a significant reduction in valuation multiples, with the NTM revenue multiple decreasing from 8.9x to 7.8x in July. Several factors contribute to this trend:
Concerns about the overall state of the economy and consumer spending.
While companies are making heavy investments in AI, there's uncertainty surrounding AI revenue generation in the short term, especially at the application layer.
We remain optimistic about AI's future impact on revenue growth. However, we anticipate that realizing its economic value, especially at the application layer, will take a couple of years, as companies are taking months or years to successfully launch generative AI products at scale. Our optimism is partly due to early signs from companies deploying generative AI products. For example, ServiceNow reported its highest-ever net new ACV growth, attributed to their AI features.
In August, 13 companies will report Q2 earnings, potentially revealing new trends. Key events to watch include Snowflake's report on August 21. Following a challenging 6 months and CEO change, observers will focus on their generative AI features' momentum and related revenue, especially as Databricks rapidly releases new features. NVIDIA's report on August 28 has become a must-watch, serving as a barometer for AI sector sentiment.
New models
The major news of July is that Meta finally released its highly awaited 405B model, along with new 8B and 70B models. Llama 3 405B was trained on approximately 15.6 trillion tokens of multilingual data, including a mix of web content, code, and mathematical/reasoning data. A very exciting aspect is how they promoted model distillation by changing the license to allow developers to use the outputs from LLaMA models to improve other models. We expect this will drive further proliferation of new models. Secondly, although benchmarks vary and are not a foolproof measure of which model is best, according to the MMLU benchmark, the gap has closed between open-source and closed-source models, as shown in the chart of Maxime Labonne.
While not yet released, OpenAI is likely in advanced GPT-5 development, which may widen the gap again. It will be interesting to see if future GPT and LLaMA model releases continue to narrow this gap, and how sustainable this constant race will be. The landscape could shift dramatically if either company achieves a major research breakthrough. The certainty is that both founders and consumers benefit greatly, as foundational model providers push each other to innovate quickly and contribute more to the community. This pace of innovation might not be possible with only one or two players in the space.
To close the month on a high note, DeepMind released a 2B model optimized for on-device use, it was trained on 2 trillion tokens using knowledge distillation What's amazing is that it scores similarly to GPT-3.5 and Mistral 8X7B Instruct at orders of magnitude smaller size.
With these two models released, Jacob Solawetz, founder of Flybridge portfolio company Arcee, recently shared an analogy to transportation infrastructure, suggesting that the future will likely be a combination of both LLMs and SLMs in parallel.
As smaller models improve, requiring less compute or even running on devices, we question how this might impact demand for players like NVIDIA. Smaller models have lower input-output prices per token, and several on-device models don't need GPUs. However, at the other end of the spectrum, Meta is already planning Llama 4, which they mentioned will require 10x the compute of Llama 3 to train. This contrast raises interesting questions about the future computer landscape. While NVIDIA's growth seems likely to continue, we also expect companies related to different components of on-device AI chips, like Qualcomm, ARM, and AMD, to see good growth in coming years.
AI Insights and News:
Tech Companies Q2 2024 Earnings Highlights:
TSMC reported their Q2-24 earnings. Some interesting insights included:
Estimated to grow 32% YoY in the next quarter, an impressive rate for a company of that scale.
Alphabet reported their Q2-24 earnings. Some interesting insights included:
Continues innovating at every layer of the AI stack, from chips to agents.
Leak news regarding a suite of new AI features for the upcoming Google Pixel 9. (Source). This is further evidence that the next wave we will see is increased adoption of on device-AI.
Google Cloud growing 29% YoY.
IBM reported their Q2-24 earnings. Some interesting insights included:
Highlighted their smaller model Granite series, ranging from 3 billion to 34 billion parameters, which are designed to be cost-effective and highly customizable for various coding tasks, potentially reducing costs by approximately 90% compared to large models.
Highlighted their recently launched InstructLab, a tool for more rapid model tuning through synthetic generation.
IBM's book of business related to generative AI exceeds $2 billion since it’s inception.
Servicenow reported their Q2-24 earnings. Some interesting insights included:
Now Assist net new ACV doubled quarter-over-quarter, becoming the fastest-growing new product in the company's history. Interesting to see they signed 11 Now Assist deals with 1M+ net annual contract value and 2 of them are over $5M. These are great early signs of willingness to pay for AI features.
Company internally deployed Now assist. Company shared their IT help desk, saving 45 minutes per active case. They are also saving 30 minutes every time a computer generates knowledge based articles. They also estimate they will save 21k hours with fast self-services and developer’s completing non complex scripts. This is impressive as you add the hours.
We have been personally impressed by ServiceNow recent speed of execution, and is a company to monitor for both a revenue and profitability perspective in the coming year
AMD reported their Q2-24 earnings. Some interesting insights included:
Delivered record Data Center GPU revenue with MI300 quarterly revenue exceeding $1 billion for the first time
Microsoft expanded the use of their MI300X accelerator to power GPT-4.
Number of AMD-powered cloud instances increased 34% from a year ago to more than 900. This increase indicates a rising adoption and a trend toward more widespread use of AMD processors in the cloud.
Shared how they have invested over $125 million across a dozen AI companies to expand the AMD AI ecosystem. Interesting opportunity for startups building the space.
Announced the acquisition of Finnish startup Silo AI for $665M in cash. The idea is to bring their extensive experiences in enterprise solutions with customers like Allianz and T-Mobile, with the aim of expanding the AMD AI ecosystem.
Microsoft reported their Q2-24 earnings. Some interesting insights included:
Over 60,000 Azure AI customers, up nearly 60% year over year.
More than 77,000 organizations have adopted GitHub Copilot, up 180% year-over-year. GitHub's annual revenue run rate is now $2 billion, with Copilot accounting for over 40% of GitHub revenue growth this year.
Qualcomm reported their Q2-24 earnings. Some interesting insights included:
Forecasting that at least 50% of PCs will be AI capable by 2027.
High focus on their on-device AI solutions, e.g., all Galaxy Z Fold and Flip 6 are powered by the Snapdragon 803.
Meta reported their Q2-24 earnings. Some interesting insights included:
A big question people frequently ask themselves is how Meta is monetizing their AI strategy if they are open-sourcing their model. In recent interviews, Mark has been sharing how they want to control their own destiny and avoid dependence on other players, as they had with the Apple App Store.
In relation to how they are leveraging their advances they shared how they are continuing to leveraging AI to enhance content recommendations on Facebook and Instagram, boosting engagement through improved recommendation models. AI is transforming advertising by predicting audience interests better than advertisers and will soon generate and personalize ad creatives.
Meta projects that their assistant will become the most used AI assistant by the end of the year. It offers functionalities ranging from information search to role-playing conversations, and creative features like "Imagine Yourself" for generating personal images in various styles. It will be interesting to see if they are able to remove the #1 spot of OpenAI, given how it is embedded inside applications billions of people use every day. There are good odds of that happening.
Note: These insights are focused on AI developments and priorities as discussed in management publications, earnings calls, and other company announcements.
Performance Overview:
Inception (January 2023) to date returns for the Flybridge AI Index are 116%. In comparison, over the same time period the Cloud Bessemer Index returned 28%, the S&P 500 returned 41%, the Nasdaq returned 64%, and the F-Prime Fintech Index returned 63%.
The median NTM revenue multiple was 7.8x.
The median quarterly YoY revenue growth rate was 16%.
The median LTM net income margin was 12%.
Additions/Deletions:
No new additions to the index in July.
