Bangladesh has just done something it has never done before. The national budget for fiscal year 2026-27 nearly doubles the health sector allocation to Tk 69,409 crore (Tk 694.09 billion), crossing 1.01 per cent of GDP for the first time, up from 0.58 per cent of GDP in the revised FY2025-26 budget (The Daily Star, June 2026).
Alongside this, the government announced the immediate recruitment of 5,000 doctors and 100,000 additional health workers, and a plan to issue every citizen a digital health card by 2028 (BSS, June 2026). These are not small gestures. They read like the opening move of a government trying to rebuild a health system that has long run on the grit of its frontline workers rather than the strength of its institutions.
But even the reporting on the budget itself carried a caution. Health economists welcomed the increase while warning that "budget implementation remains a major challenge" (The Business Standard, June 2026).
Bangladesh's public health spending stood at just 0.40 per cent of GDP as recently as 2021, and out-of-pocket costs still make up roughly 73 per cent of total health spending, among the highest shares in South Asia (Centre for Policy Dialogue, Policy Brief 2025(2)). A budget line is a promise. Whether it becomes a functioning hospital ward or a stalled procurement file depends entirely on what gets built around it.
What the neighborhood is doing
South Asia offers a useful mirror because the constraints are similar. India has built the world''s largest publicly funded health insurance scheme, Ayushman Bharat, issuing more than 44 crore (440 million) cards and running over 186 thousand primary care centers, alongside more than 47 crore (470 million) telemedicine consultations through its eSanjeevani platform (Press Information Bureau, June 2026). India has also started layering AI into that scale: a cough-analysis tool called Cough Against TB and an automated diabetic retinopathy screening system called MadhuNetrAI now run across dozens of facilities (Press Information Bureau, 2026).
What I do not see published anywhere in that record is an independent bias audit of either tool across income, region, or gender, the exact check my own work on fairness-aware ECG models was built to run. Scale without that check just means an error, if one exists, reaches more people faster.
Pakistan''s picture is more fragile. WHO''s own framing of its 2026-2027 support describes a health system "evolving with determination" while still working to interrupt polio transmission and protect gains that remain reversible (WHO Pakistan, 2026).
Nepal has spent the past fifteen years closing a gap of its own making. Its National Health Research Policy, drafted around 2010, openly admitted its research budget sat under 1 per cent despite international commitments dating back to a 2000 Bangkok conference calling for 2 per cent of the national health budget to go toward research (Nepal Health Research Council, National Health Research Policy). Only in its FY2025-26 programs has that started to change, with a Healthy Nepal Campaign, a new National Neuroscience Centre, and an effort to fold fragmented insurance schemes into a single system (Public Health Update, 2025). Fifteen years is a long time for a promise to sit unfunded. Bangladesh has its own version of that promise waiting right now, a 2014 draft National Health Protection Act that has never been enacted.
Sri Lanka, despite tighter finances than Bangladesh's, has been unusually disciplined about writing the rules before scaling anything. In March 2026 it launched a National Policy on Infection Prevention and Control, a National Strategic Plan on Healthcare Quality and Safety, and a National Action Plan on Medication Safety, all built with WHO's Global Patient Safety Collaborative (WHO Sri Lanka, March 2026). None of these documents mention AI. I read that as a choice rather than a gap: build the accountability architecture first, so any technology adopted later has somewhere to plug in.
What Asia's wealthiest health systems are actually building
Singapore and Japan show what it looks like when a country treats "AI-enabled healthcare" as a governance problem rather than a slogan. In September 2026, Singapore's Parliament took up a bill folding the Ministry of Health, the Health Sciences Authority, and its medical professional oversight bodies into one integrated regulator, explicitly because "digital health, genomics and artificial intelligence may span various regulatory domains" (Ministry of Health, Singapore, September 2026).
Japan went further, moving between December 2025 and April 2026 on amendments to its personal data protection law that carve out narrow, specific exceptions for health data used in public research, a push toward near-universal electronic medical records by the end of 2030, and a reimbursement schedule that ties payment incentives directly to compliance with its AI Business Operator Guidelines (International Bar Association, May 2026). Neither country simply announced that AI should be part of healthcare. Both rewrote the legal plumbing underneath it, the consent rules, the device approval pathway, and the payment system, all at roughly the same time.
Bangladesh, it turns out, is closer to that path than I gave it credit for in an earlier piece. Beyond the National Health Compact signed in Tokyo in December 2025, which named AI-enabled tools as a goal under one of its six pillars (National Health Compact, Bangladesh, 2025), a draft National AI Policy 2026-2030 released in early 2026 goes considerably further. It sorts AI systems into prohibited, high-risk, limited-risk, and low-risk tiers, with healthcare explicitly named as a high-risk use requiring a formal Algorithmic Impact Assessment and human oversight before deployment, and it proposes an AI Innovation Fund of roughly Tk 200 crore to 250 crore (Tk 2 million to 2.5 million) through 2030 (The Daily Star, January 2026).
One legal analysis published shortly after the draft's release welcomed the ambition but warned the policy still needs clearer enforcement teeth, since "policies without teeth don't deter bad actors" (Ferdows Hossen, published shortly after the January 2026 draft). That is a fair worry. A risk tier on paper only matters once someone is actually funded and empowered to check whether a hospital's new triage tool was tested on Bangladeshi patients before it started making recommendations about them.
What the Nordic countries get right, and what they don't
It would be tempting to hold up Denmark, Norway, and Sweden as the finish line.
I don't think that's quite right. Denmark''s 2024 healthcare reform, now moving into full effect on 1 January 2027, restructures the entire system around 17 new local health councils, with operating funding rising to DKK 6.4 billion annually by 2030 and a further DKK 27.5 billion for infrastructure and digital solutions (AmCham Denmark, 2024; European Observatory on Health Systems, 2025). The explicit goal is to move decisions about hospitals, primary care, and psychiatric services as close to citizens as possible.cations.
Norway's 2026 budget, similarly, commits to a wave of new hospital construction alongside an increase of NOK 3.4 billion to hospital operating budgets aimed specifically at cutting waiting times.
And yet Sweden, which has universal coverage and some of the most technologically advanced hospitals in Europe, went into its September 2026 election with healthcare as the single issue voters cared about most, cited by 64 per cent of respondents, because the country has just 1.9 hospital beds per 1,000 people, the lowest rate in the EU, and most primary care doctors report carrying more patients than recommended (Novus polling via Euronews, September 2026).
The lesson is not "spend like this and the problems disappear." It's closer to the opposite. Even with far more money than Bangladesh will have for a long time, workforce shortages and access gaps do not solve themselves. Money buys buildings. It does not automatically buy the staffing, continuity, or trust that a patient actually experiences at the point of care.
My own research
None of this is abstract for me. A fairness audit for a screening tool, the kind India's AI programs do not yet appear to publish, is exactly what my adversarial debiasing work on ECG-based disease prediction was built to catch, cutting the performance gap between demographic groups nearly in half in testing. A way for fragmented public and private providers to train a shared model without moving raw patient data, the exact gap Japan is closing through new legal exceptions, is what my MedHE framework demonstrates on a smaller scale, achieving over 97 per cent less data movement between institutions while keeping accuracy intact.
A dengue triage chatbot I built runs on simple decision-tree logic in Bengali and English for clinics where the internet is slow and the nearest doctor is an hour away. And a maternal health risk tool I worked on with colleagues, tested with 14 clinicians in Bangladesh, was trusted by a majority of them specifically because it explained its reasoning in terms they already recognised, rather than handing over an unexplained score.
The common thread is not technical sophistication. It's whether a system can be checked, contested, and trusted by the people actually using it in a clinic with unreliable power and no specialist on call.
What should happen next
Bangladesh now has three things at once that rarely arrive together: a record budget, a draft AI policy with real risk categories already written down, and a health system young enough to build fairness into its foundation rather than retrofit it later. The draft AI policy's Algorithmic Impact Assessment requirement needs an actual, funded body to run it, modeled on what Singapore is building into its own merged regulator, not just a clause in a document. Any AI tool touching the new digital health card system should be tested on Bangladeshi patients, in Bangla, before it reaches a single clinic, the same discipline Japan is now writing into its reimbursement rules. And the 2014 draft Health Protection Act should be revisited alongside this new AI policy rather than left to sit for another decade the way Nepal's research policy did.
None of this is a reason to wait. It's a reason to build the plumbing now, while the money is finally moving, instead of discovering the gaps once the health card is already in a hundred million pockets.
* Farjana Yesmin is an independent researcher & machine learning scientist specialising in Trustworthy AI, Federated Learning, and Healthcare Applications