Exa Labs: Who we are

We’re a founder-led venture studio dedicated to accelerating the adoption of frontier technology in emerging markets by building ventures that work for the realities of these regions. We start each day asking the same question:

How can we accelerate the adoption and impact of frontier technology in emerging markets?

What we’re building

We operate across three layers: infrastructure, intelligence, and applications.

Our thesis: building across these three domains creates the a stack that drives adoption. Local capacity makes advanced systems practical and governable; intelligence tuned to realities makes them accurate and reliable; applications make them usable and valuable. When these layers are designed together, adoption and value compound.

Infrastructure

We prioritise modular, sovereign-ready capacity placed close to demand with predictable economics. Proximity matters: distance is a primary driver of latency, and edge/near-edge placement materially improves performance for real-time and high-volume workloads, one measurement study showed >80% latency reduction versus centralised clouds when capacity sits near users. Local capacity also simplifies data residency/compliance and reduces exposure to over-reliance on tech infrastructure internationally going forward. Siting alongside reliable renewables lowers delivered cost and improves resilience over time, an important lever as we scale compute. (IEEE Computer Society, xumengwei.github.io, pifsinternational.org, RMI)

Intelligence

We develop small and large language models (SLMs and LLMs) native to emerging markets. These models are trained on the cultural, commercial, and operational realities of the Global South that represent more than 85% of the world’s population, yet contribute less than ~15% of global AI training data. Grounding training in local data, business practices, languages, and systems makes the models useful in real conditions, not just on abstract benchmarks. Our focus is contextual relevance and a native data advantage, so emerging markets receive AI that understands their environment and reflects their realities. Exa is building toward equitable data infrastructure and access enabling EMDEs to develop, share, and govern the training resources that have historically advantaged high-income economies.

Applications

On top of the intelligence layer, we build applications that make frontier AI directly useful. Our first focus is AI Operators: autonomous agents that take end-to-end responsibility for business functions. They’re trained on how you work, fluent in the systems you already use, and accountable for outcomes with clear audit. Operators are the starting point. The same intelligence can power ultra-low-cost, AI-native devices at the edge. Built for constrained environments and diverse languages, they prioritise privacy, resilience, and day-one utility, bringing intelligence to the point of work.

The Problem we exist to solve

The world is moving into an AI native economy, but adoption is uneven. Emerging market and developing economies hold about 86% of the global population, yet sit far from the frontier on capacity, data, and deployment. The gap shows up clearly in the AI Readiness Index: North America averages around 82, while EMDEs average around 41, tied to basic foundations like networks, data availability, and compute. Without a different approach, many countries will remain technologically dependent on the current leaders. We exist to close that gap with ventures designed for these realities. (Oxford Insights, 2024)

At the foundation is compute. Today, just a handful of hubs host most of the world’s data-center capacity. Africa, home to nearly a fifth of humanity, accounts for only ~1% of that. That concentration drives higher latency, higher costs, limits data residency options, and centralises control. The path forward is clear: place capacity closer to demand, align it with local governance, and power it with abundant renewable energy. The Global South holds ~70% of the world’s renewable potential, yet much of it remains untapped. Pairing that resource with sovereign infrastructure is one of the great opportunities of the coming decade.(xalamanalytics.com, Reuters, RMI, IEA)

Intelligence has inherited its bias from the web. Frontier models lean heavily on web-scale corpora that skew toward English and Global North sources. Studies repeatedly find geographic bias in model behavior, factual gaps, and weaker performance on local norms. When majority of people live outside developed economies, a system trained mostly on other contexts will underperform on local languages, entities, and norms. (arXivACL, Anthology, Chatham House)

And even when the models exist, applications often fail on contact with reality. In many low- and middle-income countries, the bottleneck is not coverage but usage and affordability: four in ten people live within mobile broadband range but do not use it (GSMA, Reuters). Device mix skews older; and in parts of Sub-Saharan Africa, most connected smartphones are still 3G-only, which changes what fast and usable mean in practice (GSMA). Software that assumes always-on, high-end devices, homogeneous SaaS stacks, or card-on-file payments will stall. And quality degrades when systems aren’t built for local context: model rankings shift when evaluated on culturally sensitive knowledge rather than Western-centric sets; models show geographic disparities in factual recall across regions and income levels; and native, culturally aligned training data outperforms translation-based recipes on multilingual benchmarks. (arXiv, facctconference.org, aclanthology.org)

Finally, enterprises themselves are not uniformly SaaS-native. Surveys show many firms remain at the level of basic tools, with finance, skills, and infrastructure holding back adoption. In practice, adoption means fitting technology to existing processes and controls, not asking institutions to become different companies overnight. (DCED)

Put simply, the capacity is concentrated, the energy is underused, the data is skewed, and the applications often assume the wrong defaults. If nothing changes, today’s readiness gap becomes tomorrow’s outcome gap. Our answer is to build across infrastructure, intelligence, and applications so advanced systems are practical to use, accountable to the people they serve, and fit for the realities of emerging markets.

What motivates us

Technology diffusion rarely follows neat geographic lines anymore. Innovations can emerge anywhere and spread in complex, multidirectional waves, sometimes simultaneously across continents, sometimes in reverse flows that challenge old assumptions. Yet too often, technologies still converge around the needs and infrastructure of developed economies, creating adoption barriers because they weren't designed for the diverse realities where most people live. The result is a pattern where the global south remains on the periphery of technological progress.

We have built our careers across emerging markets and advanced economies, and that vantage point shapes how we work. We have seen technology falter when it arrives shaped by outside assumptions, and we have seen progress when systems are grounded in the realities of the people who use them. Our team carries both the lived experience of operating in emerging markets and building at the frontier of technology in advanced economies. That combination is rare, and it is what enables us to design ventures that bridge global capability with local context. It is why we believe we are positioned to close the adoption gap and ensure the benefits of frontier technology reach the places where the world is growing fastest.

We design for resilience and capability that compound over decades. That’s why our studio spans infrastructure, intelligence, and applications: the stack that lets innovation travel, endure, and serve the places where the world is growing fastest.

Research, insights & collaboration

We are assembling a network of advisors, operators, and partners across emerging markets. The purpose is clear: to guide an equitable vision of technology adoption and to make sure that the systems we build take root with as little friction as possible.

Our horizon stretches far beyond the next product cycle. We think in terms of decades, not quarters - how frontier technology can be shaped to serve the next hundred years of development in emerging markets, not just the next wave of global demand.

We are looking to work with those who share this vision: partners who bring depth, operators who know the ground, and brokers who can connect ideas to opportunity. If this speaks to you, we would welcome a conversation at partnerships@exalabs.ai