The Future of AI: Will We Remain Consumers or Become Creators?
Bangladesh’s $5 billion question
Bangladesh earned US$724.6 million from telecommunications, computer and information services in FY2024–25, according to Export Promotion Bureau data reported by The Daily Observer.
That was 14.5 per cent of the government’s US$5 billion target for 2025, and below the sector’s FY2021–22 peak. What is Bangladesh selling, and how much value is it keeping?
The Observer reports that much of the sector’s work remains in outsourcing, data processing, basic software development and graphic design.
Computer services accounted for about 87 per cent of recorded ICT service exports in FY2024–25, according to The Business Standard.
These services matter, but earnings can remain modest when firms mainly sell labour rather than products they own. The same reporting points to skills gaps and regional competition for higher-value work.
Artificial intelligence sharpens this choice. Countries can adopt AI, purchasing tools developed elsewhere to improve efficiency.
Alternatively, they can build on existing models, develop locally relevant data and products, and retain intellectual property rights. Adopting AI is essential, but long-term dependence is not inevitable.
Efficiency gains alone can be replicated by competitors using the same tools; ownership of AI products, however, creates opportunities to capture value through licensing, sales and exports.
Adelaide’s research advantageand its commercial test
South Australia offers a useful mirror.
The state has about 7 per cent of Australia’s population but contributed more than 20 per cent of national AI research output in 2021, according to the South Australian Government’s AI Knowledge Centre.
Adelaide is home to the Australian Institute for Machine Learning (AIML), a major university-based research group, and a A$20 million Responsible AI Research Centre focused on explainability, misinformation and safe deployment.
AIML works with industry and government on practical applications, including agriculture and regulatory compliance.
But research strength is not the same as commercial success: public evidence is clearer on papers and grants than on patents, licensing income or AI exports. South Australia must show how consistently research becomes products customers buy.
That is a lesson for Bangladesh, not a blueprint to copy. Dhaka must ask how talent and research can become locally owned products; Adelaide must keep demonstrating how research strength becomes lasting commercial value.
Bangla is an advantage, not a side project
Bangladesh’s strongest opening may be its own language. More than 170 million people speak Bengali, yet much AI is built around English.
Bangla-medium students and people with limited digital literacy can be poorly served, while technical terminology remains inconsistent. Global providers have limited incentive to solve every local problem.
The draft National AI Policy 2026–2030 puts a Bangla large language model among its priorities and proposes shared computing capacity for agencies and researchers.
The official policy website says the public consultation has concluded and the committee is reviewing responses; the policy should therefore be described as a draft, not an adopted programme.
Its ambitions will mean little without funding, quality datasets, computing access, skilled teams and a path from prototypes to products.
Local creation is possible but ownership matters
Bangladeshi teams are already trying to build.
Hishab, also known as Verbex, has developed Bangla voice technology designed to work over ordinary telephone networks, potentially reaching people with limited digital literacy or expensive internet access.
The Daily Star describes its work, while the company’s co-founder has discussed building Bangla language resources for AI. Another local firm, Intelsense AI, has announced a Bangla model.
The sector is moving quickly, and claims about patents, customers and which model is “first” should be independently verified before being treated as settled facts.
Creators rarely start from zero. Bangla models may adapt existing models; value can come from local data, tuning and products built on top.
But a product can solve a Bangladeshi problem while its company or intellectual property sits abroad. Building locally matters; retaining a stake in what is built matters more.
From adopter to owner: the steps in between
The journey is not a leap from buying tools to inventing a frontier model. Organisations can start by adopting existing AI, then adapt models to local data, build specialised products and eventually own intellectual property that earns from licensing and exports. Each stage has value. The risk is remaining an adopter indefinitely.
Turn policy into ownership
Four steps could move Bangladesh from ambition to ownership: fund a practical AI policy, Bangla datasets and accessible computing; tie public grants and procurement to clear local value creation and transparent IP terms; measure revenue per worker, products sold and licences signed, not just export totals; and strengthen intellectual-property rules for start-ups, researchers and investors.
None of this requires rejecting imported technology. Bangladesh should adopt tools that improve productivity and public services.
But it should also create conditions for local teams to adapt, build and own useful products. South Australia’s challenge is to turn research strength into commercial returns; Bangladesh’s is to turn its talent and language needs into products with lasting value.
The real measure of AI progress is not how many tools a country buys or how many strategies it announces. It is what gets built, who owns it and whether the value stays long enough to support the next generation of innovation. Renting may be cheaper today. Owning is what you can pay tomorrow.
(Writers: Prantanu Roy is a
part-timeLecturer at Adelaide University and Tasfia Tasnim
Revu is an MSc graduate of
Adelaide University)
