Only 5% of firms see AI returns. This company is changing that

Summary:

• There is a significant “AI divide” that’s hindering the widespread adoption of the tech. Despite massive spending on AI, only a small fraction of companies in Southeast Asia are successfully deriving value from it.
• Inf Tech aims to help bridge this gap by addressing challenges in areas such as training talent, AI trustworthiness, and value delivery.
• The key to successful AI adoption lies in strategic collaboration and targeted solutions. Companies struggling to implement AI can narrow the skills gap by partnering with specialized external firms that blend AI expertise with domain knowledge.
• Inf Tech is part of the IMDA Spark Programme, which aims to address the key challenges and support the growth of promising Singapore-based infocomm and media startups.

Without a doubt, the biggest tech story of today is AI. The segment has been fueled by more than US$32 billion in spending in 2025 alone.

Still, only a scant 5% of organizations have managed to achieve real value from the tech.

This is resulting in an “AI divide,” with just a handful of big firms establishing some AI expertise. The reality is just as stark in Southeast Asia, where AI’s transformative impact has reached only a quarter of all companies.

“It’s really a wake-up call about where the value in AI is,” says Alan Qi, founder at full-stack generative AI provider Inf Tech.

And while GPU and cloud companies are making money off AI, not much is flowing into applications.

“You need both, otherwise it’ll be like building tons of roads but no cars,” he quips.

What’s limiting AI adoption?

Deploying AI was never going to be easy, and there are a few reasons for this. Costs, Qi says, is a major one, as it can be really difficult to quantify the ROI of AI investments.

Yet, the biggest challenge is that most companies lack talent that is well-versed in both technology and business to enable adoption.

Without these skills, Qi says firms will struggle to tap into the value that’s locked within their clients’ unstructured data, scattered across files and tables. Domain knowledge is also especially critical when it comes to deploying AI in sensitive industries like finance, where issues like hallucinations can have outsized impacts.

“LLMs can sound serious but still spout nonsense, so you have to be careful about what you’re putting out,” says Qi. “Having that domain knowledge – the structures, requirements, and rules – all that helps make sure your results are accurate and compliant.”

This has become especially crucial as companies navigate the age of AI. One major financial institution, for instance, sought to build AI agents, Qi recounts. But in conversations with Inf Tech, the company discovered that it lacked foundational infrastructure.

“Experiences like these reinforce a central lesson: most firms aren’t ‘AI-ready,’ and solving infrastructure gaps is just as critical as deploying models,” he explains.

Bringing everything together requires “a full-stack, cross-disciplinary team,” which is why Inf Tech employs not just AI scientists but also system engineers and former bankers.

“It’s about how you can address problems together,” Qi says. “Not pure AI scientists, but an approach that blends AI, infrastructure, and finance.”

“Trustworthy AI”

Getting that ideal blend of business and tech talent is easier said than done, especially in the Asia Pacific, where there is a severe AI skills gap.

Qi points out that the 5% of firms with the most AI success are the ones collaborating with “external AI companies that are helping them embark on transformations that they can take to the bank.”

Qi founded Inf Tech to be that collaborative partner. An early AI leader at Alibaba and Ant Group, Qi helped build the tech giant’s infrastructure before leaving in 2021 to develop Inf Tech and a “trustworthy AI” for finance, which he says is built on two foundational principles.

First, hallucinations should be mitigated and even eliminated, which Qi explains is a fundamental requirement for financial applications, given the strict regulations in the industry.

Second, data protection is critical.

“If companies are going to use their own data in these models, they have to believe that their data is well protected,” he says.

To meet these requirements, Inf Tech designed several solutions.

The first is a “one-box” on-premises product, combining hardware and its trustworthy AI software for regulated financial institutions that cannot adopt public cloud. This created opportunities for these organizations to leverage generative AI while meeting strict compliance standards and ensuring data protection.

The second is its own neuro-symbolic computing technology and parsing model, which Qi says is the industry’s first reinforcement learning-based document parsing model. The parsing model accurately extracts structured information from various unstructured PDF files, such as companies’ quarterly earnings reports. These developments, used in combination with large language models, can effectively reduce the risk of hallucinations.

Using these two technologies, the company worked with Hang Seng Indexes, automating the highly manual process of compiling share counts and float adjustments using its Al engine while enhancing data processing quality and accuracy beyond manual capabilities.

“The solution impressed industry stakeholders, who noted it was at least six months to a year ahead of alternatives,” says Qi.

The company also helped develop a reasoning-based learning system for a financial institution by developing a unique generative AI platform. Using the system, the client was able to increase the first-time pass rate of financial certification trainees from 20% to 65% – a strong indicator of the opportunities for faster and more confident adoption of AI in areas such as education as well.

Faster and cheaper AI

Inf Tech has continued to focus on making AI deployment much cheaper and faster.

To bring costs down, the company developed inference and reasoning technologies that “share memory” across multiple GPUs and CPUs to distribute the burden of AI workflows. That makes it possible for companies to deploy several models for the same amount of money that would be needed to deploy just one.

Inf Tech’s success has now brought it to a new chapter in its journey: expanding beyond its base in Singapore to the rest of Southeast Asia.

“Enterprises in Southeast Asia are making progress very rapidly, and as they innovate and scale even more, they’ll need partners, which is exactly where we fit in,” says Qi.

Inf Tech is already making moves in markets like Indonesia, where it is developing an AI training product to address local talent gaps. It’s also working with a central credit scoring agency in the region to make credit scoring more accessible for SMEs.

“There are so many SMEs in Southeast Asia that it’d be prohibitively expensive for central banks to serve them all,” he says. “Our trustworthy AI product parses through various documents and transforms them into well-formatted credit-scoring reports.”

SEA and beyond

After Southeast Asia, the Middle East is Inf Tech’s next target, Qi shares.

“We’re positioning ourselves as enablers, so we’re always looking for local partners to work with who can help us better understand the local markets,” he says.

In the long term, Inf Tech’s focus will be on making AI services more accessible and reliable, a much more inclusive approach that builds on Qi’s belief in open-source solutions.

When it comes to how AI will reshape the world, Qi is cautiously optimistic.

“The world in 10 to 20 years is going to be so different from what we can currently imagine. Building out the infrastructure of the internet took more than 10 years, but with generative AI, all the data centers and applications are rising at the same time,” he says.

“The pace of innovation is much, much faster. It’s like the cars and roads are all being built simultaneously.”