Vendors are deploying latest generative AI tools daily in a market that has been in comparison with the Wild West. However, since the technology is so latest and continually evolving, it might be extremely confusing, with platform providers sometimes making speculative guarantees.
IT company GAI Insights hopes to bring some clarity to enterprise decision makers with the discharge of the primary known buyer's guide to large language models (LLMs) and Gen AI. More than two dozen vendors were reviewed and 7 emerging leaders were identified (OpenAI is way ahead). Proprietary, open source, and small models may even be in high demand in 2025 as leadership prioritizes AI spending.
“We’re seeing an actual migration from awareness to early experimentation to really putting systems into production,” Paul Baier, CEO and co-founder of GAI Insights, told VentureBeat. “This is exploding, AI is changing your entire IT stack of the corporate.”
7 emerging leaders
GAI Insights – which goals to be the “Gardener of Gen AI” – reviewed 29 vendors in common enterprise Gen AI use cases reminiscent of customer support, sales enablement, marketing and provide chain. They found that OpenAI stays the clear leader with a 65% market share.
The company notes that the startup has partnerships with a wide range of content and chip providers (including Broadcom, with which it develops chips). “Of course they’re the primary, they defined the category,” said Baier. However, the industry is “fragmented into subcategories,” he noted.
The six other vendors GAI Insights identified as emerging market leaders (in alphabetical order):
- Amazon (Titan, Bedrock): Takes a vendor-neutral approach and is a “one-stop shop” for delivery. It also offers customized AI infrastructure in the shape of specialised AI chips reminiscent of Trainium and Inferentia.
- Anthropic (Sonnet, Haiku, Opus): Is an “impressive” competitor to OpenAI, with models which have long context windows and perform well on coding tasks. The company also has a powerful deal with AI security and has released several tools for enterprise use this yr, along with artifacts, computing and contextual retrieval.
- Cohere (Command R): Offers enterprise-focused models and multilingual capabilities, in addition to private cloud and on-premises deployments. Its Embed and Rerank models can improve search and retrieval with Retrieval Augmented Generation (RAG), which is vital for corporations that wish to work with internal data.
- CustomGPT: Offers a no-code offering and its models feature high accuracy and low hallucination rates. It also has enterprise features like sign-on and OAuth and provides analytics and insights into how employees and customers use tools.
- Meta (Llama): Offers “best-in-class” models starting from small and specialized to frontier models. Meta's Llama 3 series with 405 billion parameters can rival GPT-4o and Claude 3.5 Sonnet in complex tasks reminiscent of reasoning, mathematics, multilingual processing and long context understanding.
- Microsoft (Azure, Phi-3): Takes a dual approach by leveraging existing tools from OpenAI while investing in proprietary platforms. The company can also be reducing reliance on chips by developing its own chips, including Maia 100 and Cobalt 100.
Other providers evaluated by GAI Insights include SambaNova, IBM, Deepset, Glean, LangChain, LlamaIndex and Mistral AI.
Vendors were evaluated based on several aspects including product and repair innovation; clarity of the product and repair, advantages and features; Track record of product launches and partnerships; defined goal buyers; Quality of technical teams and experience of management team; strategic relationships and quality of investors; money collected; and evaluation.
Meanwhile, Nvidia continues to dominate with an 85% market share. The company will proceed to supply products across all hardware and software stacks and can innovate and grow at a “rapid” pace in 2025.
More top trends for 2025
While the genetic AI market continues to be in its early stages – only 5% of corporations have applications in production – there might be massive growth in 2025, with 33% of corporations bringing models into production, GAI predicts Insights. Gen AI is the highest budget priority for CIOs and CTOs as the fee of AI computing has dropped 240x within the last 18 months.
Interestingly, 90% of current deployments use proprietary LLMs (versus open source), a trend the corporate calls “Own Your Own Intelligence.” This is as a consequence of the necessity for greater data protection, control and regulatory compliance. Key use cases for Gen AI include customer support, coding, summarizing, text generation, and contract management.
But ultimately, says Baier, “there may be an explosion in almost every use case immediately.”
Noting that an estimated 90% of knowledge is unstructured and contained in emails, PDFs, videos and other platforms, he marveled that “generational AI allows us to speak with machines and the worth of unstructured “We have never been in a position to do that cost-effectively before. Now we will. A wide ranging IT revolution is currently happening.”
In 2025, there may even be an increasing variety of vertical-specific small language models (SLMs), and open source models may even be in demand (although their definition is controversial). There may even be higher performance on even smaller models reminiscent of Gemma (parameters 2B to 7B), Phi-3 (parameters 3.8 B to 7B) and Llama 3.2 (parameters 1B and 3B). GAI Insights notes that small models are cost-effective and secure, and that there have been necessary developments in byte-level tokenization, weight pruning, and knowledge distillation that end in minimizing size and increasing performance.
Additionally, voice assistance is anticipated to be the “killer interface” in 2025 because it offers more personalized experiences and on-device AI is anticipated to see a major boost. “We see an actual boom next yr when smartphones with integrated AI chips come onto the market,” said Baier.
Will we actually see AI agents in 2025?
While AI agents are currently the talk of the town, it stays to be seen how profitable they might be in the approaching yr. There are many hurdles to beat, says Baier, reminiscent of the unregulated spread, the indisputable fact that artificial intelligence makes “unreliable or questionable” decisions and works on the idea of poor quality data.
AI agents have yet to be fully defined, he said, and people currently in use are mostly limited to internal applications and small deployments. “We see all of the hype about AI agents, but it’ll take years for them to be widely utilized in corporations,” Baier said. “They are very promising, but not promising next yr.”
Factors to think about when using genetic AI
Because the market is so crowded and the tools are so varied, Baier offered some key advice for corporations getting began. First, watch out for vendor lock-in and accept the fact that the corporate's IT stack will proceed to vary dramatically over the subsequent 15 years.
Because AI initiatives should come from the highest, Baier suggests that the C-suite conduct an in-depth review with the board to explore opportunities, risks and priorities. The CEO and Vice Presidents must also have practical experience (not less than three hours initially). Before deployment, consider running a risk-free chatbot pilot using public data to support hands-on learning and experiment with on-device AI for field deployments.
Companies must also appoint a manager to oversee the combination, arrange a competence center and coordinate projects, advises Baier. It is equally necessary to implement guidelines and training for using genetic AI. To support adoption, publish a usage policy, conduct basic training, and discover which tools are allowed and what information shouldn’t be entered.
Ultimately, “Don’t ban ChatGPT; Your employees are already using it,” assures GAI.
