Key takeaways

• The EU enacted the AI Act in May, marking the first comprehensive law in the world to regulate AI. In contrast, the UK's AI Regulation Bill has been temporarily dropped and is to be reintroduced at the State Opening of Parliament on July 17. Meanwhile, the US continued devising regulations, having announced the formation of an AI Safety and Security Board, while China released new GenAI draft legislature. 

• The GenAI ecosystem raised a record-breaking USD 19.6 billion in Q2 2024 over 60 rounds, compared with USD 6.0 billion in Q1 2024 and USD 4.2 billion in Q2 2023. This was heavily driven by CoreWeave’s USD 8.6 billion raise across two funding rounds (USD 7.5 billion and USD 1.1 billion), alongside xAI’s USD 6 billion round, collectively accounting for 74% of the total funds raised for the sector.
• Given the large rounds recorded, funding was heavily skewed toward the GenAI Infrastructure and GenAI Applications spaces. This contrasts with the fundraising pattern in the previous quarter, where startups in the FM space accounted for nearly 50% of total funds raised in the ecosystem. In fact, the average deal size for FMs fell by 43% YoY and 54% QoQ in Q2 2024, possibly due to an oversaturation of funds raised for the industry.

• FMs: The highlight of Q2 was the introduction of Apple Intelligence, a suite of GenAI models to assist users with daily tasks, integrated into Apple’s mobile devices, including the iPhone, iPad, and Mac. This system can complete tasks like writing text, summarizing it, creating images, and simplifying app interactions, enhancing user experience across its ecosystem. Microsoft also launched Copilot+ PCs, equipped with ~40 GenAI models that allow GenAI-powered apps to run without an internet connection. This aligns with the growing trend of embedding AI capabilities directly into consumer electronics, as seen in previous efforts by Samsung and Lenovo. 

• In addition, Q2 saw a flurry of new model launches, including large language models (LLMs) by Cohere (Aya 23), Google (Gemma 2 and PaliGemma), (Titan Text Premier), (MAI-1), (Mixtral 8x22B), and Stability AI (Stable LM 2 12B); fine-tuned models by Mistral AI (Codestral) for coding and Google Deepmind ( and ) for medicine; and multimodal models by OpenAI (GPT-4o), Reka AI (Core), and Meta (Chameleon). 

• GenAI Infrastructure: Hardware developments dominated, with AMD announcing , Intel launching the new AI chip , and Meta unveiling the next generation of Meta Training and Inference Accelerator (MTIA), hoping to challenge NVIDIA's dominance. Companies also focused on efficient processing and inference, with RaiderChip launching GenAI v1 for low-cost field programmable gate array (FPGA) devices and NVIDIA launching that can run graphics with minimal power consumption. Google also announced a number of launches in the hardware space, including a sixth generation of tensor processing unit () chips, a custom-built arm processor , and upgrades to its .
• Other notable updates include 1) platforms to support and accelerate app development (Amazon’s , Microsoft’s , Google’s , and ’ W&B Weave), 2) data platforms (Hyper-Converged Data Platform by DataStax and Data Intelligence platform by Databricks), and 3) synthetic data generation platforms (Fine-Tune Studio by TELUS International and CodecLM by Google Cloud).
• GenAI Applications: Players upgraded their chatbots and other applications like video and image generation ones with new models and features to offer emotional nuance understanding and enhanced vision capabilities. For instance, OpenAI with GPT-4o, Google enhanced Gemini chatbot with the Gemini 1.5 Pro model, xAI introduced a multimodal Grok-1.5V model with vision capabilities, and video generation solution providers and upgraded their products with new models. OpenAI and Anthropic (Tool Use) also introduced data analytics features to their chatbots. As these tools become more accessible and capable, they will likely disrupt traditional content creation processes across various industries.

• Notable trends in Q2 included 1) partnerships to obtain data for model training such as with OpenAI, Reka AI, and Upstage, 2) infrastructure providers partnering to offer integrated solutions for app development and AI deployment (e.g., Pinecone and Gathr Data, Qualcomm Technologies and TensorOpera, and NVIDIA and ), and 3) model and AI infrastructure integrations into smart devices, with NVIDIA partnering with to develop autonomous robots and LG unveiling its service robot powered by Google’s Gemini model. Google also partnered with to power AR smart glasses. OpenAI also partnered with Apple to integrate ChatGPT into Siri and other apps. 

• GenAI players also reported partnerships with consultancy firms to accelerate enterprises' adoption of GenAI. These include and partnerships with Intel; , , and partnering with Google; and ’s partnership with Microsoft. Additionally, became OpenAI's first reseller and largest enterprise customer to date. 

• We tracked nine deals in Q2; the majority were reported in the GenAI Infrastructure space (Databricks acquired , NVIDIA acquired Run:ai, and Rebellions and Sapeon Korea announced plans to merge). OpenAI reported two acquisitions during the quarter: Multi, a video collaboration platform, and Rockset, an analytics database platform. Legal tech firm Harvey acquired to expand its AI capabilities, and Brazilian neobank Nubank acquired model developer . 

• GenAI directly integrated into consumer devices: As GenAI models continue to advance and become more accessible, we can expect a significant expansion of their integration across a wide range of consumer devices—from mobile phones and PCs to robotics platforms. Mobile phone developers have been at the forefront of this trend, incorporating GenAI capabilities into their products. This momentum is now extending into the PC market, as demonstrated by the launch of Microsoft's Copilot+ PCs, which feature built-in GenAI-powered assistants. Similarly, in the robotics domain, NVIDIA's partnership with Intrinsic to integrate its platform into autonomous systems and LG's unveiling of its service robot CLOi GuideBot, powered by Google's Gemini model, signal the growing adoption of these technologies. According to ABI Research, the annual shipments of devices capable of running GenAI on-device are projected to surge from nearly 200,000 in 2024 to over 1.8 billion by 2030. This trend is further supported by the introduction of new PC chips from hardware developers, such as AMD's recently announced Ryzen Pro 8040 Series and Ryzen Pro 8000 Series, designed to bring advanced AI processing capabilities to laptops and desktops. As these trends continue to unfold, the integration of GenAI into consumer electronics and robotics may become increasingly widespread. 
• Targeted data acquisitions to improve models while adhering to data privacy regulations: Amidst growing enforcement of intellectual property (IP) regulations, AI developers are increasingly adopting strategic partnerships and licensing agreements to acquire and safeguard data. This approach allows them to enhance the competitiveness of their models while maintaining integrity. For instance, OpenAI continued its trend of partnerships with media companies, building on momentum from previous quarters. This quarter, the company announced collaborations with a range of media entities, including The Atlantic, Vox Media, News Corp, Reddit, Dotdash Meredith, Stack Overflow, Financial Times, and TIME. Similarly, Reka AI, another prominent player in the GenAI space, joined the growing list of companies forging partnerships within the media industry. Additionally, some developers are exploring the use of synthetic data as a means to augment the volume of their training data, demonstrated by the launch of TELUS International's Fine-Tune Studio and Google Cloud's CodecLM to generate data for model training.