Key takeaways

• Regulations: EU lawmakers passed the AI Act in June, which we feel will have far-reaching implications for foundation model builders and the generative AI startup layer alike. Transparency is a critical requirement, with foundation models required to register with the EU and make public summaries of copyrighted training data. With this, the EU has taken the lead on AI governance push. China’s posture appears aligned with the EU but the US, Japan, and India remain comparatively relaxed in approach. 

  ---

• Generative AI Applications: The push for personalization was the main emerging trend (seen across Moveworks, Omneky, Writesonic and Perplexity AI). Further, there was a rising willingness to develop in-house models rather than rely solely upon foundation model APIs. Funding developments indicate a thriving space (USD 620 million in 28 rounds), with new startups like NinjaTech AI, SapientAI, and OrbyAI emerging from stealth, and yet others reaching unicorn status, like Synthesia, Typeface, and Replit.

  ---

• Foundation Models: We saw a clear push for diversification in terms of model applications, features, and modalities on offer. Also there was a stronger open-source posture from Meta with the launch of multisensory model ImageBind. Meta was especially active, as it also released Segment Anything for image segmentation and human-like AI image creation model I-JEPA. Competition at the top is also intensifying, with companies like Inflection AI and Anthropic raising funds and developing models with a view to rivaling OpenAI. Overall, six players raised ~USD 2.4 billion in six rounds led by Inflection AI (USD 1.3 billion).

  ---

• ML Infrastructure : There was increasing focus on product optimizations for generative AI workloads and an emphasis on building capabilities across the AI/ML value chain. While ML infrastructure players are critical to the broader ecosystem, the space witnessed a dip in funding during the quarter (only USD 26.8 million in 3 rounds), partly due to its resource-intensive nature and the time required to deploy capital toward meaningful advances in infrastructure. Notably, Databricks made plays across the AI/ML value chain, acquiring MosaicML, an ML model development platform for generative AI applications, and also launching Dolly 2.0, its own open-source LLM. 

Regulations: The EU leads the push for AI governance 

• On May 16, the US Senate held a hearing on AI regulation, with a focus on ChatGPT. Emerging reports on the hearing point to concerns around regulatory capture and the risks of allowing tech giants to develop the rules that govern AI. The US has made some progress in terms of domestic AI regulation, including the National Institute of Standards and Technology (NIST) AI Risk Management Framework and the Blueprint for an AI Bill of Rights. At present, however, the US remains in the early stages of this regulatory journey and there is no comprehensive federal legislation dedicated to AI.

• On June 14, EU lawmakers passed the draft law known as the AI Act (499 MEPs out of 620 voted in favor). The Act classifies AI systems based on perceived risk and requires providers of foundation models to assess and mitigate potential risks. These players are required to register their models in the EU database prior to release on the EU market. Generative AI systems based on these foundation models are also expected to comply with transparency requirements and ensure safeguards. Additionally, detailed summaries of copyrighted training data have to be made publicly available. There are, however, some exemptions for research and AI components provided under open-source licenses.

• EU lawmakers will continue to debate the specifics of the Act during the year and it is likely that compliance for companies will be enforced by 2025 or later. There are concerns that the requirement to disclose AI training data will trigger further copyright lawsuits. Meanwhile, over 150 executives from companies like Renault, Heineken, Airbus, and Siemens have signed an open letter asking the EU to rethink its regulatory push, concerned that the AI Act will “jeopardize Europe’s competitiveness and technological sovereignty.”

• China has also developed regulations applicable to generative AI services available to the general public rather than those developed for research purposes. Among other requirements, Chinese tech companies need to register generative AI models with the Cyberspace Administration of China (CAC) within 10 working days following the release of draft laws.

• Japan is comparatively lenient and expects to align more closely with the approach adopted by the US. In part, this appears to be informed by the country’s focus on leadership in the field of semiconductors. India is another country with a comparatively relaxed posture toward AI regulation.

Funding: Foundation models lead in dollar terms, with generative AI players raising the most rounds

• During Q2 2023, the Foundation Models segment attracted the most funding in dollar terms at ~USD 2.4 billion. Generative AI Applications had a record high in terms of funding rounds (29). Meanwhile, ML Infrastructure had a relatively quiet quarter with three players raising a total of ~USD 26.8 million.

Generative AI Applications

• During Q2 2023, 28 disruptors raised a total of ~USD 620 million in disclosed funding. The design, publishing, and digital assets segment attracted the bulk of funding during the quarter (~41% of total funding), with two companies raising USD 100 million and above—applied AI research company Runway (USD 141 million) and AI enterprise content company Typeface (USD 100 million).

• Marketing content (~USD 106 million) and software engineering (~USD 102 million) collectively accounted for another ~34% of total funding during the quarter.

• Q2 2023 was an especially active quarter for Generative AI Applications, with 29 funding rounds. Business process improvement led the pack with seven funding rounds in Q2 2023, with an average deal size of ~USD 11 million. Education, research, and finance followed with six rounds and an average deal size of ~USD 3 million.

• Seed funding was the most common form of funding during Q2 2023 (~66% of all funding rounds). However, early- and growth-stage funding dominated in dollar terms (~48% and ~37% of total funds raised, respectively).

• Eighteen startups raised seed funding during the quarter, with AI conversational search platform Vectara drawing the highest infusion (USD 28.5 million), followed by AI language learning app Loora (~USD 9.3 million). Six startups raised early-stage funding, with Typeface (USD 100 million) and Replit (USD 97.4 million) leading the way. Meanwhile, the growth funding raises went to Runway (USD 141 million) and Synthesia (USD 90 million). Notably, Runway’s funding infusion was the highest of any startup in the space during the quarter.

Foundation Models

• During Q2 2023, six players raised a total of ~USD 2.4 billion in disclosed funding. LLM builders predictably attracted the bulk of the funding (~97% of total funding). 

• Inflection AI, the startup behind the LLM Inflection-1 and chatbot Pi, had a significant infusion of USD 1.3 billion from notable investors like Microsoft and NVIDIA. Anthropic (USD 450 million), OpenAI (USD 300 million), and Cohere (270 million) also raised significant tranches of investment.

• The quarter also witnessed funding raises by companies like Hippocratic AI and GenHealth that are working on language models fine-tuned for healthcare applications. 

• Growth funding and unknown rounds were among the most common types of funding during the quarter. Hippocratic AI was unique for being the only startup to raise a robust round of seed funding (USD 50 million), signaling strong investor interest in healthcare-specific AI applications.

ML Infrastructure

• ML Infrastructure had a relatively quiet quarter, with three players raising a total of ~USD 26.8 million in disclosed funding. SiMa.ai, which offers Machine Learning System on Chip (MLSoC) hardware, had the single largest raise of the quarter at ~USD 13 million.

• Two of the three funding rounds in the quarter were seed-funding raises by BentoML, an open-source platform for shipping and scaling AI applications, and Openlayer, an ML validation and testing platform. 

• A total of 54 companies had at least one funding round during 1H 2023, with the majority being in the Generative AI Applications space (41 companies), followed by ML Infrastructure (seven) and Foundation Models (six).

• A further six players had two funding rounds during 1H 2023 alone: OpenAI (~USD 10.3 billion), Anthropic (~USD 750 million), Typeface (~USD 165 million), ElevenLabs (~USD 21 million), Lavender (~USD 13.2 million, both rounds in Q1 2023), and Scriptic (~USD 8.2 million).

• Generative AI Applications: Several companies received significant funding infusions during the quarter. Runway’s Series C round of USD 141 million was the highest, led by Google, Salesforce Ventures, NVIDIA, and others. Other major rounds included Typeface’s Series B (USD 100 million), Replit’s Series B infusion (USD 97.4 million), and Synthesia’s Series C (USD 90 million). Notable investors included Salesforce Ventures, SVA, Andreessen Horowitz, and Sequoia Capital.

• Foundation Models: Q2 2023 was a big quarter for large AI model builders. Inflection AI’s USD 1.3 billion venture round in June 2023 was the largest infusion, led by investors like NVIDIA and Microsoft. Other major rounds included Anthropic’s Series C raise (USD 450 million), OpenAI’s venture round (USD 300 million), and Cohere’s Series C round (USD 270 million). Notable investors included Salesforce Ventures and Andreessen Horowitz.

• ML Infrastructure: SiMa.ai’s ~USD 13 million venture round from VentureTech Alliance was the only notable funding raise during the quarter. Other notable investors included DCM Investors and Quiet Capital.

Product updates: New features abound, with a focus on personalization, model innovations, and infrastructure

Generative AI Applications 

• We tracked 18 product updates during Q2 2023 across a total of 17 disruptors.

• One disruptor, AI copywriting assistant Jasper, launched consecutive updates during the quarter. In April, the company announced its API, enabling enterprises to build its AI engine into their product. In June, it released Jasper Campaigns, an AI marketing creation platform. 

• During the quarter, several disruptors launched AI features or tools focused on generating personalized content. These included Moveworks (launched Creator Studio, a no-code platform for building custom conversational AI interfaces), Omneky (launched a new Advertising LLM capability to generate personalized content in alignment with brand identities), Writesonic (released AI Article Writer 5.0 for personalized, SEO-optimized content), and Perplexity AI (introduced Perplexity Copilot, a GPT-4 powered feature that provides more personalized responses).

• During Q2 2023, key incumbents also launched generative AI-enabled products. In May, Microsoft announced its early access program for Microsoft 365 Copilot, its AI assistant feature for Microsoft 365 applications and services. The same month, Meta released the AI Sandbox for advertisers, a testing playground for early versions of new tools and features including generative-AI-powered ad tools. In May, Google launched Product Studio, a free tool that allows Google Shopping merchants to edit and customize product images using generative AI.

Foundation Models

• We tracked 10 product updates during Q2 2023 across a total of eight players. 

• Meta had an especially busy quarter. In April, the company released Segment Anything, a new foundation model for image segmentation. The new model can remove any object in an image with a single click. In May, it announced ImageBind, a new open-source multisensory model that combines text, audio, visual, movement, thermal, and depth data. In June 2023, Meta released I-JEPA, a human-like AI image creation model that uses background knowledge about the world to complete unfinished images, instead of looking only at nearby pixels like other generative AI models.

• Google was in close pursuit in May, releasing three new foundation models on Vertex AI including Codey (text-to-code), Imagen (text-to-image), and Chirp (speech-to-text). The company also introduced PaLM2, Google’s next-gen language model.

• In April 2023, AWS launched its foundation model API service Amazon Bedrock and the new Titan language models, along with the general availability of Amazon CodeWhisperer. Meanwhile, OpenAI launched the ChatGPT app for iOS.

• Other notable foundation model builders also released a mix of new tools and updated versions of AI-powered models. In May, AI21 Labs released a new version of Wordtune for iOS including a text-generating feature and voice-activated prompts using Siri. Multimodal model builders MidJourney and Stability AI also released updates. The former launched version 5.2 of its image synthesis model (with a camera-like feature that provides zoom control over framing) and the latter released Uncrop, a new AI outpainting tool.

ML Infrastructure

• We tracked 14 updates during Q2 2023 across a total of 11 players.

• Two disruptors had consecutive updates during the quarter—ML observability platform Arize AI and integrated MLOps platform Weights & Biases. In April, Arize AI launched Phoenix, an open-source library to monitor LLMs for hallucinations. In May, the company also launched LLM observability tools to evaluate and monitor production deployments of these models. Meanwhile, Weights & Biases launched W&B Prompts in April, a suite of tools to support LLM operations, and subsequently introduced W&B Weave and W&B Production Monitoring in June. The latest additions expanded Weights and Biases’ platform support for the complete ML lifecycle.

• ML hardware incumbents like Intel, AMD, and NVIDIA were also active during the quarter, releasing products customized for generative AI workloads. In May, NVIDIA launched the GH200 Grace Hopper Superchip to run AI and high-performance computing workloads. In June, AMD launched Instinct MI300X, an AI accelerator chip for generative AI workloads, while Intel introduced new GPUs Intel Arc Pro A60 and Pro A60M for AI inference, among other applications.

Partnerships: Product collaborations dominate, as companies continue to pursue synergies

Generative AI Applications 

• Investment platform Streetbeat’s partnership with stock trading API platform Alpaca (to integrate with Alpaca’s suite of brokerage, custody, and post-trade solutions),
• AI character engine Inworld AI’s partnership with Geenee AR (to integrate Inworld’s AI character engine in the latter’s web-based augmented reality publishing platform), 
• AI music generator Endel’s collaboration with Universal Music Group (to create AI-powered functional music), 
• AI marketing content platform Persado’s partnership with digital journey continuity platform Namogoo (to leverage Persado’s Dynamic Motivation solution, alongside the latter’s intent and segmentation data to analyze visitor intent), and 
• AI marketing content platform GoCharlieAI’s collaboration with research and tech development organization SRI international (to develop its LLM Charlie into a multimodal model).

• AI voice generator WellSaid Labs and automated video producer Videate (to use WellSaid Labs’ AI voices), 
• Conversational AI provider Vectara and ultrasound education ecosystem SonoSim (to use Vectara’s Natural Language Understanding technology in its ultrasound education and training), and 
• AI voice generator ElevenLabs and audiobook and ebook streaming company Storytel (to create tailored AI voices for Storytel’s core markets).

Foundation Models

• In April, AI21 Labs partnered with Amazon to make its Jurassic-2 models available through Amazon’s Bedrock service. In June 2023, the LLM creator also collaborated with intelligence company Clarivate to integrate AI21’s LLMs into Clarivate’s solutions to enhance academic conversational search and discovery. The integration is expected to offer academics, students, and researchers access to context-specific information and personalized services.

• In April, industrial automation company Rockwell Automation partnered with Microsoft to integrate ChatGPT as a digital assistant in the latter’s manufacturing systems. The same month, LG CNS partnered with Microsoft to use Azure OpenAI to create an AI contact center and knowledge chatbots. In May, telecommunications services provider e& also partnered with Microsoft to integrate Azure OpenAI into internal operations.

ML Infrastructure

• We counted five partnerships across four players in the ML infrastructure space during Q2 2023.

• Microsoft dominated these product collaborations. In April 2023, Snorkel AI partnered with Microsoft to integrate Snorkel Flow with Azure AI services to enable data science and ML teams to develop custom AI/ML models. In May 2023, NVIDIA partnered with Microsoft to integrate its AI Enterprise software solution with Azure Machine Learning, accelerating enterprise AI initiatives. During the same month, Microsoft also partnered with AMD to collaboratively develop AI processors. In June 2023, Microsoft collaborated with specialized cloud provider CoreWeave to leverage its cloud computing infrastructure to bolster computing power for OpenAI models like ChatGPT.

• Integrated MLOps platform Weights & Biases also had a notable collaboration during the quarter, partnering with deep learning infrastructure company Lambda to offer MLOps tools and infrastructure for enterprise AI and deep learning workloads.

• During Q2 2023, multiple incumbents showed interest in specific startups. Microsoft and NVIDIA both participated in Inflection AI’s funding round. Similarly, Google and NVIDIA both participated in Runway’s funding round. Meanwhile, Google also joined Anthropic’s funding round aimed at developing AI chatbot Claude - notably, Anthropic is positioning itself to rival OpenAI’s models.

• Compared with the others, Microsoft was also especially active in terms of partnerships during the quarter.  In the foundation models space, Microsoft launched customer partnerships for its Azure OpenAI service with companies like Rockwell Automation, LG and e&. Meanwhile, in the ML infrastructure space, Microsoft embarked on collaborations with SnorkelAI and CoreWeave to release new Azure AI integrations and boost computing power for OpenAI models, respectively.

• Similar to Microsoft’s Azure OpenAI service, Amazon also forged a mutually beneficial partnership with AI21 Labs, offering its Jurassic-2 models via the Amazon Bedrock foundational API service.

M&A: Contents.com continues its acquisitive spree, while Databricks builds across the value chain

• During April–May 2023, Milan-based AI content generator Contents.com went on an acquisition spree, integrating Scribeur.com (a France-based tech platform that produces SEO and web marketing text), Traduzione (an Italian translation firm), and Scribox (an Italian content marketing specialist firm). These acquisitions broaden Contents.com’s reach in France, expand its customer base, and consolidate its position in Italy. It also follows the company’s recent expansion into the Brazilian market, strengthening its position in the global web content production space.

• In June 2023, Databricks, a data infrastructure and MLOps platform, entered an agreement to acquire MosaicML, an ML model development platform for generative applications, in a deal valued at ~USD 1.3 billion. The acquisition will integrate MosaicML’s platform within Databricks’ Lakehouse platform, offering enterprises a unified platform to create and manage a range of AI use cases and LLMs. In April 2023, Databricks also launched Dolly 2.0, its own open-source instruction-tuned LLM.

Summary of new Generative AI startups added in Q2 2023

• A total of 43 new Generative AI startups were added to our coverage during Q2 2023. Listed below are the five highest-funded startups.

Appendices

Appendix 1 

Appendix 2 

Appendix 3