Hybrid infrastructure is accelerating healthcare AI adoption across APAC
Healthcare providers across Asia Pacific are moving faster on AI because hybrid infrastructure finally fits how care actually runs. Keep sensitive data on-premise. Tap cloud scale for training, collaboration and speed. Lower latency. Lower risk. Lower cost of moving data.
The momentum is clear. Lenovo's CIO Playbook 2025 reports that 65% of organisations now favour hybrid AI and high-performance computing. That shift tracks with the data surge: healthcare generates roughly 30% of global data, with volumes expected to grow nearly 40% year over year through 2026 and 2027.
Why hybrid is winning inside hospitals and health systems
Hybrid maps cleanly to clinical reality. "You can keep clinical and patient data on premise… because clinical data, patient data, personal data, needs to meet some very, very strict compliance regulations," said Sinisa Nikolic, Director and Segment Leader, HPC & AI, Asia Pacific, Infrastructure Solutions Group at Lenovo. At the same time, the cloud gives you scalable AI, easier collaboration and faster iteration.
There's a financial angle too. As Nikolic noted, "not moving data will save money." Shift compute to where it's safest and most efficient, instead of shuttling petabytes across regions and vendors.
The real bottleneck isn't the model - it's integration
"Healthcare isn't slowed by AI models. It is actually slowed by fragmentation, and that remains the biggest bottleneck," said David Irecki, Chief Technology Officer at Boomi. Hybrid adoption is becoming unavoidable as care models evolve, but multi-system sprawl adds complexity.
Across APAC, only 30% of workflows are optimised for generative AI at scale. The next leap won't come from a new algorithm. As Irecki put it: "It'll actually come from connecting the models that we already have today."
Data readiness and infrastructure gaps are holding teams back
Many providers aren't enterprise-AI ready. "Data is fragmented. It's reasonably inconsistent. It's not governed well," said Nikolic. Legacy infrastructure also struggles with the compute and storage demands of AI.
Even where the tech is present, sponsorship can lag. Without cross-functional backing - clinical, IT, security, legal and operations - pilots stall and never reach production.
Operational risk is rising with adoption
AI introduces new threats that don't require a breach to cause harm: data poisoning, model drift and unvalidated outputs. In a hybrid environment, those risks can propagate across on-premise and cloud systems fast.
Oversight has to keep up with scale. As Irecki said, "Ultimately, AI should return time to clinicians, not add risk." That means governance, monitoring and fail-safes built into the workflow - not added after the fact.
What healthcare leaders can do now
- Map workloads to the right venue: keep PHI, imaging and time-critical inference on-premise; use cloud for experimentation, non-PHI analytics and burst training.
- Stand up a unified integration layer (iPaaS/API gateway) to connect EHR, RIS/PACS, LIMS, scheduling, and AI services without point-to-point sprawl.
- Fix data basics: standardise vocabularies, implement data catalogues and lineage, enforce quality checks, and define golden records.
- Modernise the foundation: GPU-ready compute, high-throughput storage, and secure networking that supports both on-prem and cloud bursting.
- Operationalise models: MLOps for versioning, approvals, drift detection, rollback and audit trails across hybrid endpoints.
- Engineer safety: pre-deployment validation, guardrails for unvalidated outputs, human-in-the-loop steps for clinical decisions, and clear escalation paths.
- Apply zero trust to data flows: least-privilege access, encryption in transit/at rest, and continuous posture assessment across sites and vendors.
- Build a cross-functional steering group that includes clinicians; set clear success metrics tied to time-to-diagnosis, length of stay, no-show reduction, or claim accuracy.
- Start narrow: pick 2-3 high-value pathways (triage, imaging prioritisation, denials management), prove impact, then scale.
- Measure and close the loop: monitor performance, bias, safety signals and cost-to-serve; retire models that underperform.
Helpful references
For governance and safety in clinical AI, see WHO's guidance on AI for health ethics and oversight: WHO AI governance. For a practical approach to model and system risk, many teams borrow from the NIST AI Risk Management Framework.
If your team needs structured upskilling on AI workflows, integration and MLOps, explore curated programs by role here: AI courses by job.
The bottom line
Hybrid is the practical path for healthcare AI in APAC. It keeps clinical data where it belongs, gives you scale when you need it and reduces the friction of moving sensitive information. The winners won't just have strong models. They'll have clean data, tight integration and guardrails that let clinicians move faster with confidence.
Your membership also unlocks:
