Generative AI Digest: A shift to the application layer

November saw a wave of announcements related to generative AI at the application layer as providers such as Google LLC, Microsoft Corp., Snowflake Inc. and others move full throttle toward the opportunity with AI agents. Additionally, model providers Mistral AI SAS and OpenAI LLC are enhancing capabilities around their enterprise chat products, building more compelling user experiences and encouraging more robust application development on their platforms.

The recent US election results have significant implications for the AI industry, with many predicting a more lenient approach to agency regulation and antitrust enforcement. Meanwhile, earnings reports from Alphabet Inc., Microsoft and Amazon.com Inc. over the past few weeks demonstrated major AI infrastructure investments but also seemingly sizeable revenue boosts from AI demand. Alphabet suggested that its cloud business had expanded 35% year over year, Microsoft reported its AI business (products and cloud) had achieved a $10 billion run rate, and Amazon suggested its AI offering was seeing "triple-digit" revenue growth, boosting overall AWS sales 19% year over year. Both Meta Platforms Inc.'s and Microsoft's reported earnings exceeded analyst expectations. These developments present a positive vision for AI accelerationists, although some commentators are noting a slowdown in advancements with frontier-level models, which may temper their excitement slightly.

Product releases and updates

ChatGPT search was formally introduced Oct. 31, following July's announcement of SearchGPT — a prototype that indicated OpenAI was seeking to compete more aggressively with Google and GenAI search specialists such as Perplexity AI Inc. The search capability was immediately rolled out to ChatGPT Plus and Team users and will be rolled out to other users in several phases. Partnerships with publishing companies such as Axel Springer, Condé Nast, Dotdash Meredith and the Financial Times, as well as sports, weather, stocks and map data providers, are being used to help populate and link through to results.

Reportedly, only hours before OpenAI made its ChatGPT search announcement, Google announced that "Grounding with Google Search" would be rolled out in Google AI Studio and the Gemini API, for all Gemini 1.5 models. This grounding capability can be turned on at the cost of $35 per 1,000 grounded queries, with Google claiming this grounding reduces hallucinations and provides access to more up-to-date information. The company also announced that Google Maps would integrate Gemini to address queries about a site or to suggest a destination, for example. AI capabilities are also being rolled out for Google Earth and vehicle navigation app Waze.

Microsoft made various announcements about AI agents at its Microsoft Ignite conference. It released new specialized out-of-the-box agents in Microsoft Copilot for tasks like project management, interpretation and employee self-service. It also unveiled Copilot Actions, a new feature that automates repetitive tasks within Microsoft 365, which can streamline workflows like summarizing meetings, generating reports and preparing for meetings. Copilot Studio, a tool for building AI agents, has gained autonomous capabilities, an agent library and deeper integration with Azure AI. Additionally, Microsoft is enhancing Copilot's capabilities across various Office apps, including PowerPoint's translation feature and improved integration with SharePoint. Finally, Microsoft has released updates to Azure Foundry portal (formerly Azure AI studio) that orient around the governance challenges of managing and evaluating multiple agents and applications, plus an Azure AI Foundry SDK for developers to build, test and manage AI applications.

At its developer BUILD conference, Snowflake unveiled its new Snowflake Intelligence platform, through which users can build dedicated agents to generate insight from their data. Built on the Cortex AI framework, these agents help simplify data preparation and governance. The feature is currently set for private preview.

Eleven Labs Inc. has launched a new feature enabling users to create customizable conversational AI bots on its developer platform. This capability allows for adjusting variables such as tone, response length and agent persona, integrating various large language models such as Gemini, GPT or Claude. Users can also incorporate their own knowledge bases and define data collection criteria for interactions. While ElevenLabs has primarily focused on text-to-speech services, it aims to enhance its offering with speech-to-text in the future, to compete with major players like Google and OpenAI. The startup is currently seeking fresh funding.

French startup Mistral AI has enhanced its Le Chat chatbot platform to better compete with ChatGPT, with new features including the ability to search the web with citations, process large PDF documents and images, and a "canvas" tool for creating and modifying content like documents and presentations. These advancements are powered by Mistral's new models, including the 124-billion-parameter Pixtral Large, which excels in multimodal tasks (both text and image), and the updated Mistral Large 24.11, designed for improved long context understanding. Both models are available under research and enterprise licenses.

Weka.IO Ltd. has introduced a high-performance storage technology for NVIDIA Grace CPU Superchips, designed to enable fast data access and reduce the energy consumption of AI infrastructure. The capability was previewed at Supercomputing 2024 and represents a partnership between NVIDIA, Supermicro and WEKA. WEKA also announced WARRP, the WEKA AI RAG Reference Platform, which it presents as a blueprint to simplify AI inferencing for RAG-based systems at scale.

DeepSeek, a Chinese AI research lab and subdivision of High-Flyer Capital Management, has launched the R1-Lite-Preview, a reasoning-focused LLM accessible through its web-based chatbot, DeepSeek Chat. The model aims to deliver high-level reasoning capabilities. It also features a transparent chain-of-thought reasoning process, allowing users to follow its logical steps in real time, enhancing accountability. While the model is available for public testing, DeepSeek plans to release open-source versions and APIs in the future.

Israeli audio AI startup aiOla Ltd. has launched Whisper-NER, a new open-source model designed to enhance privacy in audio transcription by integrating automatic speech recognition with named entity recognition (NER). Built on OpenAI's Whisper framework, Whisper-NER automatically identifies and masks sensitive information during transcription, addressing some common concerns related to data security. The model is available on Hugging Face, Inc. and GitHub.

Cisco Systems Inc.'s Webex Customer Experience portfolio now includes the Webex AI Agent, which helps organizations provide prompt and effective support. Its AI Agent Studio is a design tool for business users and IT administrators to train and deploy AI agents efficiently. Additionally, Cisco added new AI Assistant features for Webex Contact Center.

Dell Technologies Inc. announced new products and services — many aimed at the challenges companies are facing in adopting GenAI. Relevant announcements include new server products designed for AI workloads and Dell Design Services for AI Networking — a packaging of services to help customers better optimize the IT architecture, and data use, associated with AI workloads.

Beijing Byte Dance Telecommunications Co. Ltd. has enhanced Doubao, its ChatGPT alternative, with a new feature that transforms text and images into videos. Additionally, ByteDance's Jimeng AI app, launched in mid-2024, has updated its Seaweed video model, claiming quicker and more detailed video creation.

Funding and M&A

Amazon has increased its investment in Anthropic PBC to $8 billion, adding another $4 billion to the foundation model startup. Alongside news of the round, Anthropic announced that AWS would be its primary cloud and training partner going forward and that it was collaborating with the hyperscaler on its AWS Trainium hardware. AWS will reportedly remain a minority shareholder.

Red Hat Inc. has acquired Neuralmagic Inc., a six-year-old startup that specializes in optimizing and accelerating AI inferencing workloads. Like Red Hat, Neural Magic has a legacy in open source. Of particular interest to its acquirer were its contributions to the vLLM project, an open-source library and distributed LLM engine serving to reduce inferencing speed and memory footprints, which Red Hat already uses in its Red Hat Enterprise Linux AI and Red Hat OpenShift AI products. Prior to the acquisition, Neural Magic had raised $50 million over three rounds of funding, including investors Andreessen Horowitz, Comcast Ventures and Verizon Ventures.

SAS Institute has purchased IP from UK-based startup Hazy Limited, focusing on synthetic data creation. By integrating Hazy's technology with the SAS Data Maker platform, users can create richer synthetic datasets to enhance AI projects. The updated SAS Data Maker is set to preview in early 2025 and will work with SAS Viya, the company's AI integration platform. This is SAS Institute's first deal in over two years.

Sierra Technologies Inc., a conversational AI customer service startup, announced a $175 million funding round from Greenoaks Capital Partners, ICONIQ Capital and Thrive Capital Management. The company has a post-money valuation of $4.5 billion. This round brings the total funding for the startup, which was incorporated in 2023, to $285 million.

Read AI Inc., creator of an eponymous cross-platform workplace co-pilot, announced a $50 million series B round led by new investor Smash Ventures Management Company. With this, the company's total funding reaches $81 million, and its post-money valuation stands at $450 million. Returning investors Madrona Venture Group and Goodwater Capital also participated.

Writer Inc. raised $200 million in a series C round at a $1.9 billion valuation. The company, which initially emerged as an application provider delivering GenAI-based marketing and writing support, has been repositioning as a generative AI application development platform with a broader focus on agents and continues its investment into proprietary foundation models. The latest funding round brings its total funding to $348 million. New investors include Adobe Ventures, Salesforce Ventures and Citi Venture Capital International.

Politics and regulations

In late October, the Biden administration released the National Security Memorandum on Artificial Intelligence. The NSM is designed to drive greater use of AI by the federal government and directs actions to improve the security and diversity of AI chip supply chains, keep AI developers abreast of cybersecurity and counterintelligence information, and engage in a relative competitive-advantage analysis of the US private sector AI ecosystem. President Joe Biden is also working to allocate funding from his 2022 CHIPS Act, which offers manufacturing incentives for semiconductor production in the US, before President-elect Donald Trump comes to office.

With the change in administration, there have been bills in both the Senate and House that aim to preserve the US AI Safety Institute. One bill is the Senate's Future of AI Innovation Act, sponsored by Senators Maria Cantwell and Todd Young. Another is the House's AI Advancement and Reliability Act, from Representatives Jay Obernolte and Ted Lieu, which would maintain the institute's functions under a new name.

Trump has appointed North Dakota Gov. Doug Burgum to lead the US Interior Department and a new National Energy Council. The council will focus on expanding energy supplies and reducing costs to support AI development. Trump considers this effort important for national security and economic prosperity.

Hyperscaler efforts to link up datacenter expansions with sources of nuclear power are hitting hurdles. Earlier this month, the Federal Energy Regulatory Commission rejected a proposed interconnection service agreement for the Susquehanna nuclear power station in Pennsylvania, which would have increased power capacity for an AWS datacenter nearby. The decision raised concerns over grid reliability and potential impacts on consumer costs. This follows news that Meta also had to fold plans for a datacenter expansion near a nuclear power source in an unspecified US location due to environmental concerns.

Saudi Arabia announced a $100 billion program, known as Project Transcendence, to strengthen its domestic AI sector. The strategy for this investment appears to be broad — reportedly to build new datacenters, enhance education programs, attempt to build up the local startup sector and encourage global technology companies to establish a Saudi Arabian footprint.

The European Union's AI Office released the first draft of the General-Purpose AI Code of Practice. This Code aims to help developers of general-purpose AI models comply with the EU AI Act. There are three more drafting rounds to come in the next five months, with the 36-page code expected to expand significantly. Some attention is being paid to the outline suggesting that AI developers will need to forecast risks, potentially identifying at what stage different risk thresholds would be exceeded as models improve..

..

Gain access to our full news & research coverage and the industry-specific data that informs our insights

This article was published by S&P Global Market Intelligence and not by S&P Global Ratings, which is a separately managed division of S&P Global.
451 Research is a technology research group within S&P Global Market Intelligence. For more about the group, please refer to the 451 Research overview and contact page.