NFX’s Morgan Beller describes AI as water in a recent article. Specifically, she commentates on the increasing repetitiveness across Generative AI applications that more often than not are written on top of the same models, which use the same data, and fundamentally demand similar user experiences.

With enough demand, commoditization over time is a natural phenomenon in most products, not just tech. In this case the comparison is drawn between water and generative AI applications, where even in something as ubiquitous, cheap (if not free), and commonplace as water, bottled water companies have managed to carve out market segments and create differentiated value.

She concludes generative AI products will have to better define distribution and perceived value to customers, in addition to the other ingredients of data, model, and UX, in order to enjoy the same unfair advantages bottled water has in the market for quenching thirst.

For Southeast Asia, bottled water may not be enough. Miracles need to be performed to make generative AI rain in the region.

Bottled water is expensive in Southeast Asia

That is because the main ingredients to build generative AI in Southeast Asia are still in short or unusable supply at this point in time as we’ve written about here. The most popular models still hallucinate in the Southeast Asia context. Data for some use cases, especially in financial services, can be difficult to collect or still require digitization. Regulation also still needs to be developed around data privacy and collection in certain markets. Talent also remains in magnetized towards the West when it comes to AI, though the hope is that this will slowly change.

And even if a local generative AI company does solve for these fundamental issues, and takes the advice of the NFX article and develop distribution and find some meaningful perceived value to customers, there is still the matter of costs of scale and competition, which we’ve written about here.

The opportunity cost for adoption can also tend to be higher in markets where labor costs are lower (the perennial issue with AI commercialization in emerging markets even before gen AI applications). This makes the path to commercialization and mass adoption a lot less straightforward than launching on HuggingFace or getting an app viral on TikTok.

Southeast Asia markets and generative AI companies need their own resources

The problems around infrastructure and cost are actually closely tied to each other. How the region deals with these barriers to entry and growth for generative AI may offer clues to what companies will lead this AI transformation in the region. Singapore in particular is no stranger to navigating these problems, even when it comes to water.

Part of the Singaporean miracle was being able to create its own sustainable source of water, after years of relying entirely on imports. Singapore built out the entire infrastructure to recycle wastewater (including the Changi Water Reclamation Plant able to treat up to 900 million litres of wastewater daily or 350 Olympic size swimming pools) even with its short supply of land.

The story of NEWater presents some insights that extend into generative AI commercialization in Southeast Asian markets:

Bottled water isn’t enough for Southeast Asia

(1) Sustainable, local sources of water were necessary to develop to reduce the stress of reliance on imports. In the same way, while OpenAI’s models will remain widely adopted, there are use cases like healthcare, law, financial services, or even specific business operations, where having localized LLMs can make generative AI applications more effective, and thus more viable for commercial use. But then localizing LLMs requires having robust data infrastructure. See the work bluesheets.ai is doing to help organizations develop more streamlined data infrastructure.

(2) There is water, it’s just not in the form that can be used. In the same way, data is available — it’s just a matter of having the infrastructure to not only ensure data can be processed but also protected from legal (data privacy) and technological (cybersecurity) standpoints.

(3) Because Singapore lacks the surface area to support all the treatment infrastructure, wastewater sourcing and treatment happens below ground, as far down as 25 stories deep in an intricate network of tunnels. This parallel will be a little bit more of a stretch, but consider: because of the cost constraints that emerge when scaling generative AI solutions (think GPU costs, talent costs, etc.), the early commercial use cases will be driven by companies with enough depth in business in terms of data, unit economics (i.e., is incorporating gen AI still justified by the margins?), and market share (i.e., a market ready to use their solutions).

In our last AI Notes #17, we wrote about how Vietnamese wealth management fintech Finhay leveraged its presence as a leading smart investment platform with more than 3 million users and key partnerships with funds in the country to introduce generative AI features to their app, delivering a more convenient and customized experience to investors.

(4) Treated wastewater is not just for drinking, but also used extensively by industries like Singapore’s microchip manufacturing sector and cooling systems. For generative AI in the region, the most immediate impact from localized solutions will be seen in custom enterprise use cases, like the ones WIZ.AI is supporting with their enterprise LLM solutions and AI agent solutions built on top of their proprietary Bahasa and Thai language models, among others.

(5) Treated wastewater supplements reservoir water and other sources that would otherwise be easily depleted due to Singapore’s weather and other factors. Tying back to the NFX article, perceived value to customers is an important aspect of developing a viable product with generative AI.

Gen AI images and videos are cool and all, but what matters at the end of the day for the generative AI ecosystem to flourish in an emerging market region like Southeast Asia is whether or not solutions make a difference to the status quo that they would pay for. Water can be bottled, be plugged in with electrolytes, and rebranded in a 1000 ways, but if the customer can’t value the difference it makes, it doesn’t matter.

Generative AI is water but bottling it up for the use cases consumers and businesses in Southeast Asia markets will pay for will take more than a creative approach to distribution and value perception. Pre-existing scale, regulatory momentum, data infrastructure, and top down adoption will also need to be part of the equation. Not quite like a miracle, but better than sitting and waiting around for one.