Key takeaways

• The European Parliament approved the much-awaited AI Act in March, which aims to prevent the misuse of high-risk AI while promoting innovation. It is expected to be enacted in May/June 2024, and member states are responsible for establishing their own national oversight agencies.

• The GenAI ecosystem raised USD 6.5 billion (grew 59.0% QoQ) across 81 funding rounds, representing the highest funding amount within a quarter since 2021.
• Three major rounds—Amazon’s USD 2.75 billion investment in Anthropic, Chinese chatbot developer Moonshot AI’s USD 1 billion raise, and Infrastructure provider Kyndryl’s USD 500 million post-IPO debt—collectively accounted for 65% of the total funds raised during the quarter.

• Foundation Models: OpenAI launched its new model Sora in text-to-video generation, while Stability AI introduced Stable Video 3D. Similarly, Google expanded its portfolio with two video models. Most other updates were new model introductions, with startups like Anthropic (Claude 3), Mistral AI (Mistral Large), and Inflection AI (Inflection-2.5) launching models to compete with OpenAI’s GPT-4.

• GenAI Infrastructure: Incumbents exhibited significant activity in the development of GPUs and AI chips, with NVIDIA launching Blackwell GPU and Meta unveiling a new GPU cluster infrastructure, while Qualcomm released the Snapdragon 8S Gen 3 chipset, which would help to handle large models like those launched by Mistral AI and provide better performance. The emphasis on on-device processing too saw significant growth, with both incumbents and disruptors introducing new solutions, including RTX 500 and 1000 GPUs by NVIDIA and those by startups like DEEPX and Expedera.

• GenAI Applications: Incumbents and disruptors continued to add new features to enhance existing products, including voice capabilities to chatbots (e.g., Character.ai, OpenAI) and the ability to add sound effects to videos (e.g., Pika) as well as updating underlying models and other services to offer more precise and personalized replies for chatbots (e.g., Google Gemini,GPT Store).

• Multiple partnerships focused on integrating FMs into mobile phones were announced, including Google’s partnerships with MediaTek and Samsung. Developers continued broadening the accessibility of their models on cloud platforms, with Mistral AI leading the way by collaborating with major cloud providers such as IBM, Amazon, and Microsoft. 

• NVIDIA partnered with Lenovo, Cohesity, Amazon, Microsoft, Cisco, and Oracle to offer integrated data security, management, and computing infrastructure solutions. Disruptors Cerebras Systems, VAST Data, and Lambda reported similar partnerships. 

• Collaborations were also reported toward incorporating AI capabilities directly into hardware devices, including OpenAI releasing a ChatGPT app for Apple's Vision Pro headset and Adobe making its text-to-image tool available in the Apple Vision Pro headset.

• More than 10 M&A deals were reported within the quarter. Infrastructure provider Databricks dominated, with two new acquisitions: Einblick and Lilac. We tracked two acquisitions in the GenAI Applications space, with marketing content generator Jasper acquiring Clipdrop and Typeface acquiring TensorTour. Notably, Apple acquired DarwinAI and Brighter AI ahead of a major expansion into GenAI across its products and services planned for this year.

• Rising prevalence of small language models (SLMs) - SLMs present a potential solution to the drawbacks associated with large language models (LLMs), as they are designed to be more streamlined, requiring fewer parameters and less training data. This makes them faster and more cost-effective to train, as well as easier to deploy, particularly on smaller devices or in settings with limited computing power. Additionally, the capability of SLMs to undergo fine-tuning for particular applications offers increased adaptability and customization. Microsoft has reportedly been developing smaller and more economical GenAI models, supplementing its Phi-2 model launched in Q4 2023. Google also introduced Gemma, a model compact enough to run on a personal computer. As small and medium-scale enterprises increasingly embrace GenAI technology, demand for SLMs is poised to rise, driven by their cost-effectiveness and minimal resource requirements.

• GenAI’s expansion by integrating with hardware devices - As capabilities evolve, GenAI application developers are exploring ways to broaden GenAI applications by integrating GenAI directly into hardware. Noteworthy developments included OpenAI launching the ChatGPT app and Adobe making its text-to-image tool compatible with Apple's Vision Pro headset. Additionally, Brilliant Labs introduced lightweight AR glasses, with AI assistant Noa, while reports suggest that Google may incorporate Gemini into its earbuds. These integrations hold significant implications for the future of digital content creation and user experience, enhancing creativity and interaction in virtual environments.
• Advancement in open-source models - While open-source models have been in development for some time, initial studies indicated that their performance lagged behind proprietary models. However, recent advancements have shown a significant improvement in the quality and performance of open-source models, particularly with leading companies like Mistral AI and Stability AI introducing models with performance closer to their proprietary counterparts. Meta's open-source artificial general intelligence (AGI) initiative and xAI’s decision to open-source the base code of the Grok AI model on GitHub demonstrate a shift toward collaboration and transparency in the development of AI technologies. This should enable new features, unique applications, and design interfaces, as developers can leverage open-source models without having to build from scratch.