Why APJ is setting the pace of the next AI era and what Western boards should know

This post is part of Global Signals—a new Cohesity series bringing on-the-ground perspectives from global technology leaders on how AI, data, and infrastructure are reshaping markets.

SINGAPORE — At CNBC Converge in Singapore recently, a pattern that has been building for the past year came into sharper focus: the center of gravity for applied, industrial-scale AI is shifting toward Asia-Pacific and Japan. Not gradually—but structurally.

This is not a story about breakthroughs in modeling. It’s a story about infrastructure, policy coordination, and—most critically—the readiness of data to support AI at scale.

The shift is happening in capital, not slogans

IDC projects APJ digital infrastructure spend at roughly US$110 billion by 2028, with gen-AI-related outlays compounding at a mid-20% annual rate. More striking is what sits behind that number. In the last 12 months, Japan cleared a path for an OpenAI–SoftBank joint venture—“Stargate Japan”—wired into an investment envelope of north of US$100B. Korea has announced a multi-year national AI push in the hundreds of billions, including a domestic GPU build-out and funding for sovereign foundation models. Malaysia’s Johor corridor has gone from an industrial footnote to hosting roughly 60% of the country’s planned data center capacity, powered by some of the region's cheapest industrial electricity.

None of that is happening in isolation. It’s being orchestrated—deliberately—by governments that view AI the way their predecessors viewed semiconductors, ports, and grids—as statecraft. Dr. Balakrishnan framed Singapore’s posture in a line I haven't stopped thinking about: the country’s strategic asset is being “boringly predictable.” In a world where trust has become a premium, predictability lowers the transaction costs of doing AI at scale. Capital is voting accordingly.

Why it’s happening now

Three structural forces are compounding and are unlikely to reverse over a 24-month horizon.

1. Power and permitting are not Western strengths right now.
   In the PJM grid—the largest in North America—the average interconnection queue time now exceeds six years. Siemens Energy’s latest results showed gas-turbine orders roughly doubling year over year, with lead times pushed out to the end of the decade. Europe’s industrial electricity is running 1.5–3x the U.S. landed cost, and that math is directly throttling the continent’s AI build-out. Meanwhile, data centers in parts of Malaysia are contracting power at 10–11 U.S. cents per kWh, with land, water, and permits bundled into single-window approvals. Electrons, turbines, and entitlements are now the rate-limiter on model economics. APJ has more of all three, more quickly.
2. APJ governments are treating AI as an industrial policy, not IT spend.
   Japan, Korea, Singapore, and increasingly Indonesia and Vietnam are pairing public capital with concessionary land, streamlined approvals, and anchor-tenant commitments from domestic champions—telcos, banks, utilities, carriers. The result is a public-private flywheel that Western markets, absent a serious re-industrialization push, are struggling to match. If you’re a Western enterprise, your APJ competitors are running on a subsidized cost curve you cannot replicate organically.
3. The price-per-token gap is widening.
   On a trailing 12-month basis, inference costs for comparable model classes have fallen roughly 70–90%—but the absolute spread between regions is widening. A token produced on cheap APJ power, against purpose-built silicon, against a sovereign or semi-sovereign data estate, is materially cheaper than the same token produced in a Western market paying three times for electricity and waiting six years for a substation upgrade. Over a ten-year horizon, that spread compounds into an entire cost-of-goods advantage.

What this means for Western C-suites, by industry

I will resist the temptation to write a generic call to action. The implications look different depending on the business you run.

Financial services

APAC banks and insurers are not waiting. DBS, MUFG, KB, and several regional regulators are already piloting agentic workflows for KYC, fraud, underwriting, and middle-office reconciliation—on regionally-hosted models trained against their own transaction estates. Western incumbents still wrestling with model-risk governance as a blocker will be outmaneuvered on cost-to-serve and time-to-decision before they finish their second committee review. 

The board question is no longer “are we AI-ready?” It is “Are our data contracts, lineage, and resilience posture good enough that we can move at APJ speed when our competitors do?”

Healthcare and life sciences

Singapore, Japan, and South Korea are building national clinical-data commons under explicit sovereignty rules—linking genomics, imaging, EHR, and claims in ways U.S. and European systems legally cannot, yet. That will translate into faster trial recruitment, better population-scale phenotyping, and earlier drug-response signals. 

If you’re a Western pharma or payer, your 2027–2028 pipeline decisions should assume an APJ partner, an APJ data enclave, and an APJ-trained specialist model in the loop. Build the data-sharing and sovereignty muscles now—not when the first sovereign-AI RFP lands.

Manufacturing and industrial

This is where APJ’s lead is most physical. Factories in Japan, Korea, Taiwan, and increasingly Vietnam are layering vision, predictive maintenance, and scheduling agents onto shop floors that are already instrumented and have clean time-series telemetry. The data is good because the operational discipline is good. 

If you run a Western plant, the honest near-term threat is not that your competitor will out-design you. It’s that they will out-yield you, quarter after quarter, because their agents have better closed-loop data to learn from. Your 2026 capex conversation should include line-level data contracts and immutable operational history as first-class line items.

Public sector and regulators

APJ governments are writing the sovereign-AI rulebook in real time—Singapore’s AI Verify, Japan’s METI framework, Korea’s AI Basic Act. For Western ministries, the risk isn’t being out-regulated—it’s being out-operated. Citizen-facing services—tax, immigration, health, benefits—are being re-platformed against sovereign models trained on national data. 

The benchmark for “good government” is quietly being reset in Asian capitals. Western agencies that don’t invest now in classified-grade, sovereign, auditable data foundations will find themselves benchmarking against a standard they can’t hit.

The bottleneck nobody is funding

At that recent roundtable discussion, I made a point that landed harder than I expected: turning on the power is only about a quarter of the battle. The next quarter is silicon. Another quarter is the model and the tooling. The final quarter—the one almost nobody in this industry is funding proportionally—is the data itself: Is it clean, is it governed, is it sovereign, is it resilient, is it recoverable, is it trustworthy enough to put an autonomous agent on top of?

Every gigawatt of new capacity, every billion-dollar training run, every sovereign model program is underwritten by an unstated assumption—that the underlying data estate is AI-ready. In most enterprises I speak with, it isn’t. And APJ has a structural advantage here, too. Because so much of the new build is greenfield, data architecture is being designed for AI, not retrofitted around legacy estates stitched together across two decades of M&A.

If I were briefing a Western board today, I would put it like this: your AI strategy is not GPU-bound, it is data-bound. Fix that, and you can compete with anyone. Don’t, and no amount of infrastructure will save you.

The race is about trust, not tokens

The open question for the next five years isn’t which region adds the most megawatts, or which train runs the largest model. It’s about which region and which enterprises within it make their data the most trustworthy, sovereign, recoverable, and AI-ready the fastest. That is a strategic choice. That is a board-level choice. And right now, many of the most consequential moves are being made in Tokyo, Seoul, Singapore, and Kuala Lumpur—not in the usual places.

If the last eighteen months have felt like watching a market you once led become harder to read, that’s not a perception problem. It’s the signal.

The question is what you do with it in the next two quarters, or spend the rest of the decade catching up.