Reading Notes · Appendix B

Two Outsider Lenses on Hong Kong–GBA Healthcare

How Elon Musk and Jensen Huang would approach the medical-cost and payment-architecture problem.

Companion notes to the strategy briefing above (slides 42–43). The premise: a hypothetical in which each thinker examines the Hong Kong and Greater Bay Area healthcare system, using his own tools, knowledge and way of approaching problems, with one end goal — that everyone can afford private health insurance on a sustainable basis. These are thought experiments prepared for discussion; they are not statements by, or endorsed by, either individual. September 2026.

Part 1 — Elon Musk: the deletion playbook

1.1 The framing: rockets are a physics problem; healthcare is a coordination problem

The honest answer is that healthcare cost inflation is harder than rockets in one specific way Musk himself would identify: rockets are a physics problem, healthcare is a coordination problem. Physics doesn’t lobby, doesn’t have entrenched fee schedules, and doesn’t bill you after the fact. But his method would still generate a very recognisable playbook.

1.2 First-principles the cost, and ignore the industry’s categories

He would refuse to accept “medical costs are high” as a starting point, and decompose it to raw inputs: what does a hernia repair actually consume? Surgeon-hours, theatre-hours, sutures, drugs, a bed-night, administration. He would find what the briefing already shows — a procedure with perhaps HK$30,000 of physical inputs billed at HK$180,000 in a private hospital, with most of the gap being coordination waste, defensive practice, billing friction and price opacity rather than scarcity of actual care. His conclusion would be blunt: the rocket isn’t expensive — the supply chain is.

1.3 Run “the algorithm” on the delivery system

His five-step process maps almost perfectly onto value-based care:

• Question every requirement — why does a colonoscopy need a hospital? Why does the insurer pre-authorise a procedure the data says is appropriate 97% of the time? Most requirements in Hong Kong healthcare exist because someone wrote them in 1994, not because physics demands them.
• Delete the part or process — fee-for-service line items, claims adjudication for routine care, separate billing departments. Delete first; if you don’t have to add 10% back, you didn’t delete enough.
• Simplify — one bundled price per episode instead of forty line items. That is literally the warranty-backed bundle in the payment architecture.
• Accelerate cycle time — measure outcomes per episode and iterate monthly, not via five-year government reviews.
• Automate last — his famous mistake admission: automate only after the process is simplified. AI pre-authorisation on top of fee-for-service is automating waste.

He would land on capitation, tiered networks and bundled payments not by reading health policy, but by manufacturing logic.

1.4 Vertical integration, because he can’t trust the suppliers

SpaceX built its own engines because the aerospace supply chain was the cost problem. The Musk move in Hong Kong and the GBA would be the same: own the primary care layer (the router), own or contract the specialist tiers, own the insurance licence, own the data pipe (eHealth+ integration). That is essentially the architecture in the briefing — a capitated router at the top, a tiered network below, a reinsurer holding the tail. He would arrive at the Kaiser / AdventHealth logic not from healthcare policy reading but from manufacturing logic: if the insurer, the router and the provider are separate entities, each one’s margin is the next one’s cost inflation.

1.5 Obsess over the cost curve per unit of outcome

Musk thinks in dollars per kilogram to orbit. The healthcare equivalent is dollars per resolved episode — which is precisely the “single resolved episode blended cost” north star in the briefing. He would demand a dashboard: cost per cataract resolved, cost per stent with a 30-day warranty, cost per attributed life per month. Anything not on that dashboard gets deleted.

1.6 Use the GBA as Gigafactory Shanghai

His China playbook: where a regulatory environment actually lets you build fast, costs collapse. He would see exactly what the briefing sees — Shenzhen delivers the same stent at one-fifth of the Hong Kong price with nationally appraised quality. The arbitrage isn’t medical tourism; it is manufacturing location strategy. Route the volume to where cost-per-outcome is lowest; keep the frontier — complex diagnosis, brand trust, the warranty — in Hong Kong.

1.7 Iterate the insurance product like software

He wouldn’t design the perfect insurance product in a committee. He would launch a minimal capitated product on a small attributed population — say, one employer group of 5,000 lives — measure, iterate quarterly, expand. Ship the minimum viable product; let telemetry tell you what’s broken.

1.8 Where he’d hit the wall — and where the honest answer lies

• Rockets have one customer with aligned incentives; healthcare has a patient, an employer, an insurer, a hospital and a regulator whose incentives point in five directions. His algorithm assumes someone is allowed to delete parts. In Hong Kong, deleting requires negotiating with all five.
• Trust is not deletable. Patients don’t follow cost curves; they follow doctors they know. Even he can’t move fast and break things when the “thing” is someone’s cancer diagnosis.
• The deepest irony: the goal — everyone affording private insurance sustainably — is only achievable if someone accepts the boring, un-Musk-like work of capitation arithmetic, attribution contracts and stop-loss thresholds. The rocket science of healthcare is actuarial, and actuarial work doesn’t look impressive on stage.

The likely final memo, after six months examining Hong Kong and the GBA:

“The physics is easy. The parts cost is 20% of the price. The fix is a vertically integrated, capitated, tiered network with the router at the top — I would build it. But the hardest engineering problem here is not medical. It’s getting five institutions to sign the same contract.”

Which is, not coincidentally, the premise of the entire briefing: the payment architecture is the rocket.

Part 2 — Jensen Huang: the platform counterpoint

2.1 The framing: the two opposite geniuses of the same era

Musk is a deletion thinker: the answer is always “remove parts, own the supply chain, make the physics cheaper.” Huang is a platform thinker: “don’t remove the parts — accelerate them, and let a million people build on top.” Applied to Hong Kong–GBA healthcare, their disagreement would be genuine and deep. Five challenges to the deletion playbook follow.

2.2 Challenge 1 — “You measured the wrong cost curve”

Huang’s lifelong obsession is the cost curve of compute: the same computation that cost millions now costs fractions of a cent, because of acceleration. Looking at Hong Kong healthcare, he would say: Musk decomposed the cost of a hernia repair into surgeon-hours, theatre-hours and sutures, and found the physical inputs are cheap. Correct — but the most expensive input in that decomposition isn’t the sutures, it is the scarce human expertise: the specialist’s diagnosis, the routing judgment, the monitoring. And that input is about to go through the same deflation as compute.

The marginal cost of medical intelligence is heading toward near-zero — AI copilots reading imaging, drafting discharge summaries, catching sepsis early, doing the routing triage. Musk’s plan makes the supply chain cheaper; Huang’s bet is that the labour content of care itself gets substituted by accelerated compute. One attacks the denominator by deleting waste; the other attacks it by collapsing the price of expertise.

2.3 Challenge 2 — “Vertical integration is too slow and too capital-heavy”

This is the core philosophical challenge. Musk’s answer is: own the insurer, own the router, own the tiers — Kaiser-style — because every interface between entities is where margin and friction live. Huang’s entire career is the counter-argument: NVIDIA doesn’t own the datacentres, the cloud providers, the model builders or the applications — it owns the platform layer underneath all of them, and the ecosystem scales faster than any single vertically integrated company ever could.

Applied here: don’t build one vertically integrated capitated network in Hong Kong — build the CUDA of value-based care. The payment rails, the episode definitions, the outcome-verification standards, the attribution algorithms, the stop-loss pricing models — as open, cheap, irresistible infrastructure that every insurer, hospital group and GP clinic in the GBA can adopt. Musk’s plan produces one great network in ten years; Huang’s produces a market standard in three — because he knows from CUDA that when the platform is free and obviously useful, the incumbents don’t need to be negotiated with one by one. They come to you.

2.4 Challenge 3 — “You can’t crash-test patients”

Musk iterates with real hardware — blow up the rocket, learn, relaunch. Huang would point out that in healthcare the equivalent is unacceptable, and offer his signature alternative: simulation before deployment. Build a digital twin of the GBA health economy — claims data, eHealth+ records, synthetic patient populations, synthetic insurers — and run a thousand capitation contract designs, tier structures and stop-loss thresholds in simulation. Find the designs that bankrupt the router or harm patients in silico, not in vivo.

He would say to Musk: your minimum-viable-product approach with one employer group is a Falcon 1 strategy in an industry where the first explosion kills people and triggers ten years of regulation. Simulate a million lives overnight. Compute is cheaper than courage.

2.5 Challenge 4 — “Why is AI only the check?”

The briefing’s router is a human primary care group, checked by an AI copilot. Huang would ask why the intelligence layer is the check and not the substrate. In his worldview, the routing decision — which patient, which tier, which episode definition — is exactly the kind of pattern-recognition-over-longitudinal-context problem that accelerated AI does better than any human gatekeeper, at near-zero marginal cost, for every patient simultaneously.

The human GP keeps the trust and the relationship (he would grant Musk’s point that trust is not deletable) — but the routing intelligence is infrastructure, like the GPU under the game. The capitated primary care group becomes an AI-native routing platform that happens to employ doctors — not the other way around.

2.6 Challenge 5 — “Measurement is now a data-centre problem”

The briefing’s whole argument is that outcome measurement arriving is what re-tiers the market. Huang built MLPerf precisely because he understood that whoever defines and runs the benchmark shapes the industry. He would say: the reason Hong Kong has no published episode outcomes is not conspiracy — it is that measurement was too expensive. It no longer is. Automated outcome extraction from eHealth+, risk-adjustment models, patient-reported outcome collection at scale — this is a data-centre problem now. Don’t wait for a regulator to mandate benchmarks; make measuring outcomes so cheap that not publishing becomes the suspicious choice.

2.7 Where they converge

They would agree completely on: measurement first; routing as the centre of the architecture; the tiered structure — Huang sells tiers himself, frontier silicon at frontier prices and last-generation silicon as the mid-tier, and he would laugh at the “peace-of-mind buyer pays frontier prices” idea because it is literally his pricing model; and the GBA-as-Shanghai cost arbitrage.

2.8 The real disagreement

Musk believes the savings come from deleting the system’s waste; Huang believes the bigger prize is substituting compute for clinical labour, so that the content of care changes, not just its price. Musk’s endgame is the same healthcare, 60% cheaper. Huang’s endgame is a healthcare system in which the physical exam, chronic-disease monitoring, triage and half the diagnostics no longer happen in buildings staffed by scarce humans at all — so insurance affordability stops being a financing problem and becomes a solved-by-deflation problem.

2.9 The synthesis

Musk’s approach makes private insurance affordable by restructuring payment; Huang’s makes the underlying care so cheap that the insurance question shrinks. For Hong Kong and the GBA, you probably need the payment architecture to capture the deflation Huang’s technology is about to deliver — because when the cost of intelligence collapses, whoever owns the routing captures the savings.

The technology deflates regardless. The payment architecture decides who keeps the difference.

How to use these lenses

The two thought experiments are complementary reading tools for the strategy discussion:

• Musk’s lens is a cost-discipline test — for any episode, contract or process: what are the raw inputs, what can be deleted, what is the cost per resolved outcome?
• Huang’s lens is a platform test — what could become a cheap, open standard that others adopt; what can be simulated before it is tried on patients; and where does AI substitute for scarce expertise rather than merely assist it?

Held together, they sharpen the briefing’s core thesis: measurement and routing are the architecture, and the institutions that build the payment rails decide who benefits from the coming deflation of care.