How regional tech champions can succeed in GenAI

Focus on industry verticals

Established players—including hyperscalers, closed and open-source data and AI platform providers, and large language model (LLM) incumbents—dominate infrastructure, foundation platforms, and foundation models. Still, regional champions can capture market share by creating specialized solutions tailored to verticals and enterprises. Tech champions could concentrate on specific use cases to improve performance, quality, experience, and efficiency. In the Middle East and Africa, Polaris forecasts that by 2032 the media and entertainment and the banking, financial services, and insurance FSI sectors are likely to be the top verticals, accounting for 70% of the $7.8 billion Gen AI market.²

Apps currently capture 30 to 35% of the value across the GenAI tech stack. Their strength lies in understanding the “control point” (the most important system in the business and the last to be turned off to save money) in each industry and building solutions around it. This approach entails fine-tuning models for industries or enterprises and developing front-end applications with unique and industry-specific workflows. Oracle, for instance, integrated a clinical digital assistant tool to its electronic health record platform, Cerner, reducing administrative tasks for clinicians and improving overall value proposition of their platform.³

Follow an integrated stack approach

Tech champions could develop solutions combining computational infrastructure (the software used to manage computing resources) and LLMs. This strategy enhances commercial returns, given its higher customer stickiness and capital efficiency. Dell, for instance, partnered with Hugging Face to offer solutions integrating infrastructure and GenAI model to enable customers to build and deploy their own. GenAI applications.⁴

Own, localize, and control the infrastructure and computing stack

To ensure adaptability, scalability, and resilience in an ever-evolving tech landscape, regional tech players need control the infrastructure layer (the data centers and equipment to support solutions). Companies adopting this strategy would need to focus on providing GenAI solutions in a localized environment, which allows for greater data security and supports customization for regional large enterprises and government customers. Watad, a Riyadh-based AI and cybersecurity company, implemented Mulhem, the first Saudi Arabian domain-specific LLM model trained on local datasets and developed by leveraging NVIDIA’s on-prem high-performance computing system.⁵

Differentiate with access to unique datasets and proprietary IP

Solutions relying solely on publicly available data will not create incremental value or deliver a competitive advantage. That makes it critical for regional tech champions to gain and manage access to proprietary datasets. To develop intellectual property (IP) across key verticals, they should find co-investment opportunities with customers that could offer market traction. This strategy would provide tech champions with access to unique data sets while still being able to own the IP. Tech champions could also expand their access to unique data sets by striking strategic partnerships and licensing agreements with other organizations such as data providers, industry leaders, and research institutes.

Drive cost efficiencies without hindering performance

Trends in GenAI indicate a continuous increase in the size and cost for LLMs and applications built on top of them. For example, training cost for GPT-4 and Google’s Gemini Ultra are estimated to be around $78 million and $191 million respectively.⁶ Given LLMs are becoming larger—and thus pricier across the board—regional tech champions could seek to contain costs while maintaining performance (that is, limiting hallucinations, answers that are plausible but wrong). For instance, they could use an LLM controller (a tool for ranking and synthesizing output from multiple LLMs) to manage a cascade of smaller and cheaper models before relying on larger models, an approach that can significantly reduce costs.