In the early decades of the IT industry, vendors brought to market proprietary networking, communications, and file systems that worked well if used within the same company or with companies using the same vendor’s architecture, but were quite difficult to use otherwise.
The Internet changed all that. The adoption of open internet and email protocols in the 1970s and 1980s by research communities, followed by the rise of web standards in the 1990s, transformed the IT industry, forcing companies to embrace a culture of open standards and collaborative innovation. Similarly, Linux emerged as an open-source operating system in the 90s. Initially adopted by research communities, it eventually gained widespread corporate support. Over time, an increasing number of companies have embraced Linux, contributed to its development, and now support hundreds of open source projects in just about all industries.
Could open source now play a similar transformative role in the evolution of Generative AI (GenAI) technologies and business strategies?
To explore this important question, Linux Foundation Research and its partners conducted a web survey in August and September of 2024. The report, “Shaping the Future of Generative AI: The Impact of Open Source Innovation,” highlights how open source technologies are accelerating GenAI adoption while setting a foundation for the future of AI.
316 respondents industries and companies of all sizes around the world completed the survey. 29% were from organization with more than 10,000 employees, 24% from 250 to 10,000, 27% from 10 to 250, and 10% from organization of less that 10 employees; 55% of respondents were from The Americas, 18% from Asia-Pacific, 17% from Europe, and 9% from the rest of the world.
The survey comprised 46 questions that addressed various aspects of GenAI and open source adoption. Let me summarize some of the key findings from the survey.
GenAI adoption and use in organizations
“Organizations are adopting GenAI because of its ability to address a broad array of strategic and tactical needs, including content creation, personalized customer experiences, decision support, process automation, employee training, and research and planning.”
To what extent has your organization adopted generative AI?
Very high — 16%, high — 27%, moderate — 41%, slight — 10%, no adoption — 6%.
What activities does your organization undertake with GenAI models? Use of GenAI models for inference — 65%; build or train GenAI models — 38%; support the use of GenAI internally — 43%.
How much have the following GenAI techniques improved the performance of your organization?
Prompt engineering: exceptional — 15%, considerable — 42%, moderate 22%;
Retrieval Augmented Generation (RAG): exceptional — 21%, considerable — 36%, moderate — 16%;
Fine-tuning pre-trained models: exceptional — 12%, considerable — 32%, moderate — 23%.
“Text, code, and structured data are the most common modalities for GenAI because they are widely available, interpretable, and foundational to a broad range of applications.”
What generative AI modalities are you using or planning to use in your organization?
Text — 81%, code — 74%, structured data — 48%, multimodal — 47%, DevOps — 41%, speech — 35%, vision — 34%, audio — 27%.
What are your top five use cases for GenAI? Process optimization and automation — 25%, content generation — 17%, code generation — 14%, customer service and support — 11%, research — 6%.
How well are these GenAI use cases integrated into the business?
Process automation and optimization — 88%, content generation — 93%, code generation — 86%, customer service and support — 82%, research — 57%.
How open source is expanding the role of GenAI
“The adoption of open source infrastructure is emerging as a strategic differentiator for organizations highly committed to GenAI.”
What percentage of your organization’s code infrastructure that supports GenAI initiatives is currently from open source? More than 75% — 61%, 50 to 75 — 52%, 25 to 50 —75%, 1 to 25 — 81%, 0 — 32%. The average open source adoption across all responding organizations is 41%.
How much does your organization contribute to GenAI open source projects?
Frequently (more than 4 committees per month) — 17%, occasionally (1 to 4 commits per month) — 18%, rarely ( less than 1 commit per month) — 20%, not at all — 45%.
What are the most popular frameworks that your organization uses for building and training GenAI models? PyTorch — 63%, TensorFlow — 50%, CUDA — 42%, vLLM — 19%.
Which application frameworks do you use or plan to use for model inference? LangChain — 44%, LlamaIndex — 30%, Auto-GPT — 17%, AutoGen — 13%, Haystack — 9%, Semantic Kernel — 9%.
How does the open source nature of a tool or model influence its adoption within your organization? Strongly positive — 29%, positive — 42%; neutral — 17%, negative — 5%, strongly negative — 1%.
More than half of respondents (52%) indicated that transparency and trust in the source code were the primary reasons for choosing open source GenAI solutions, followed by cost efficiency (46%) and security (32%).
How important is it to use open source AI tools that are hosted by a neutral party like the Linux Foundation, instead of a corporate entity? Extremely important — 16%, very important — 17%, important — 21%, somewhat important — 24%, not important — 15%, not at all important — 7%.
GenAI and the cloud native approach
Cloud native and hybrid cloud strategies provide the elasticity and efficiency required to manage large-scale AI workloads and are thus foundational to how organizations deploy and host their GenAI models.
Where does your organization host the generative AI models for inference applications?
We self-host in the cloud — 49%, we use managed GenAI services — 47%, we self-host on premises — 38%; we deploy models locally on device — 16%, we do not use GenAI inference applications — 6%.
“Organizations with greater open source usage are significantly more likely to self-host on premises (50%) compared to those with limited open source adoption (24%). This could imply that heavy open source users likely possess the technical expertise and infrastructure to manage complex deployments independently.”
How frequently does your organization deploy generative AI models into production for inference?
Daily — 8%, weekly — 10%, monthly — 23%, occasionally — 30%, rarely — 10%, never — 5%.
Where does your organization host generative AI model building and training?
Cloud-based infrastructure — 65%; on premises — 46%; in a hybrid environment — 29%;
on infrastructure managed by third parties — 19%.
What are your organization’s primary motivations for building out generative AI infrastructure?
Security and complete control of data —43%, costs — 36%, data sovereignty and residency —36% , long-term cost savings — 35%, privacy — 33%, competitive advantage — 32%, performance — 32%,
IP protection — 31%.
Challenges in GenAI adoption
What are the most important factors when choosing a GenAI model or tool?
Accuracy or performance — 60%, security — 54%, cost — 49%, privacy — 48%, compliance with regulations — 40%, ease of integration — 39%, performance — 35%, user experience — 34%, being open source — 32%, scalability — 29%.
How much of your organization’s investment in generative AI has been converted into revenue gain? Substantial — 5%, significant — 14%, moderate —19%, some — 24%, little — 38%.
How has the adoption of generative AI impacted employment in your organization?
Hiring more employees 19%; no impact 67%; reducing number of employees — 14%.
The future of GenAI is open
How do you expect the use of open source generative AI in your organization to change in the next two years? Substantially increase — 26%, increase — 47%, stay the same — 14%, decrease — 2%, substantially decrease — 2%.
To what extent do you agree or strongly agree with the following statements?
Open source is critical for a positive AI future — 82%, open source will have an overall lower cost than proprietary AI — 66%, the benefits of open source AI outweigh the risks — 61%, open source AI will become the industry standard — 55%, open source will have less risk than proprietary AI — 50%; open source will have more capabilities — 42%; open source will provide better performance — 34%.
To what extent do you agree or strongly agree with the following statements regarding the future of AI?
AI needs to be increasingly open — 83%, AI will be pushed toward edge devices — 74%, AI training will be necessary for companies to remain competitive — 72%, AI needs regulation from government — 67%, small, specialized AI models will become more prevalent than large, centralized ones — 62%.
Conclusions and recommendations
“Open source frameworks, tools, and communities are making advanced GenAI models accessible to organizations, enabling a foundation for tailored solutions and fostering a collaborative ecosystem. High adoption rates among organizations with open source foundations emphasize the adaptability, cost-effectiveness, and transparency that open source provides.”
“Moving forward, it is recommended that organizations prioritize open source solutions as a core strategy for GenAI adoption. Moreover, active contribution to open source projects should be encouraged, as this engagement not only strengthens the broader AI community but also allows organizations to shape tools and frameworks to better meet their specific needs.”
“In summary, open source will remain indispensable for the future of GenAI, driving accessibility, innovation, and ethical considerations,” said the report in conclusion. “Organizations that embrace open source strategies and contribute to these communities are well-positioned to remain at the forefront of AI advancements. By fostering an ecosystem that prioritizes shared knowledge, responsible governance, and continuous innovation, the industry can ensure that GenAI technology develops in ways that benefit organizations, individuals, and society as a whole.”
Comments