Ten markets, one question
The ToD Pricing tab sets out why this console exists—AI compute is starting to get priced against the clock, the way electricity already is in most of these markets. This page turns that argument into a working comparison: seven of the ten electricity markets covered here already run time-variant tariffs, from three fixed regulatory blocks to continuous half-hourly settlement, and only Indonesia and Cambodia remain flat. The country model lets you plug a hypothetical data centre into any of the ten grids and see what a given mix of workload flexibility and PPA structure would cost, using each market's real tariff data rather than a generic assumption.
The console is organised around that one question at every level. This overview compares all ten markets side by side on pricing structure, tariff level, and AI-linked demand; the ToD Pricing tab lays out the full analytical argument, ranks every market by how exposed it already is, and carries the assistant-adoption picture alongside it; Country opens the live calculator for any single market; and Assumptions and References document exactly which numbers are sourced and which are modelled. Figures in standard type are sourced to a named tariff order or market operator; figures in amber, bold, are estimates built from the stated methodology—that distinction holds everywhere in the console, not just here.
Click any market below to open its full profile →
X: data centre share of grid load. Y: modelled AI share of that data centre load. Bubble size: adoption index (1-10)—a directional proxy for AI user-base intensity, not a user count. Only India has a directly sourced user figure (see ToD Pricing).
The ToD pricing thesis
Jian Zhang's original observation was about DeepSeek: in mid-2026 it moved its API to time-of-use pricing—higher rates in named peak windows, cheaper off-peak—and the read was that this could spread from China to markets where compute providers compete openly.[16] That framing is right about the business practice, but worth being precise about what actually "ripples" and what doesn't. What spreads globally is a pricing practice—AI providers choosing to bill by time of day as energy becomes too large a share of inference cost to average away.
What does not spread from China is the underlying electricity market structure in the other nine markets this console tracks. India's ToD mandate, Thailand's TOU tariff, Australia's five-minute NEM settlement—none of these exist because of China; most predate this AI pricing conversation entirely, several by decades, and were built for domestic grid-management reasons that have nothing to do with AI. The real, defensible correlation is narrower than "China's tariffs ripple outward": a globally-diffusing AI pricing practice is now running into ten independently-built electricity tariff structures, and the financial consequence of that collision varies enormously depending on which structure a given market already has. Because that consequence depends partly on where AI demand itself is concentrated, this page now also carries the assistant-adoption picture that used to sit on its own tab—who is using what, and where, alongside how each grid prices the power behind it.
Global assistant share, market ranking, and composite exposure
| Mkt | Str | Spr | DC |
|---|
- Market share of major AI assistants
- Sensor Tower's True Audience metric
- Unique users, app+web, end of May 2026
- All 10 markets, sorted by spread
- Columns: structure, spread %, DC share
- Row order anchors the 3 charts to the right
- 1-10 directional adoption index
- Population scale + internet penetration
- Not a measured user count
- Modelled % of DC load that is AI-specific
- No market publishes this directly
- Estimate, not a sourced figure
- 0-10 synthesised score
- 40% spread + 30% DC share + 30% adoption
- Answers: most compounded AI-timing exposure
Mkt/Str/Spr/DC: market / structure / peak-off-peak spread / DC share of grid, using the same three-letter codes as the country deep-dive.
India, specifically
ChatGPT counted roughly 330 million monthly users in India as of May 2026[14][15]—its largest market outside the US, close to a third of its global base. Gemini follows at roughly 229 million[14][15] in the same market. India's grid is already ToD-mandated for large commercial and industrial load; its AI user base is also the largest in this console by a wide margin. Those two facts have not yet been priced against each other.
Volatility doesn't track regulation type
Australia (80%) and New Zealand (70%) post the widest spreads in the console despite being liberalized, continuously-settled markets—volatility here comes from the market itself, not a regulator's design. Vietnam (68%) is the widest spread among the regulated-block markets, the product of a single concentrated five-hour evening peak rather than a gradual curve. Thailand (37%), Taiwan (35%), and India (40%) form a moderate, predictable middle band. Indonesia (0%) and Cambodia (5%) remain effectively untouched by time-of-day economics. None of this dispersion traces back to China—it reflects each market's own generation mix, regulatory history, and grid maturity, built independently and mostly well before AI entered the picture.
The contracting gap sits where spread meets sophistication
A wide spread only becomes a data centre problem if the contracting hasn't caught up. Australia pairs its 80% spread with the region's most mature indexed-PPA culture—operators there have already absorbed this lesson. Vietnam pairs a 68% spread with a market where third-party facilities were only just reclassified onto a materially higher tariff category in December 2025—a live illustration that category risk can dwarf timing risk, and that the contracting gap is widest precisely where the volatility is second-highest among regulated markets.
Where AI demand, grid volatility, and assistant market dynamics overlap
India's sheer volume, noted above, sits on a moderate 40% spread grid rather than an extreme one—the aggregate cost story in this region looks likelier to be driven by that volume than by any single market's extreme spread. Singapore and Australia combine above-average spread with the highest modeled AI-share-of-data-centre-load in the console, making them the clearest cases of double exposure. Zooming out to the assistant market itself: it has moved from near-monopoly to a three-way split between ChatGPT, Gemini, and Claude, driven by distribution and enterprise trust rather than a capability gap. DeepSeek's own share stays concentrated in China and price-sensitive, open-weight-friendly segments—the same economics behind the time-of-use pricing this whole page is about. Vietnam and Australia, the widest peak/off-peak spreads tracked here, are where that pricing logic is likeliest to surface next outside China.
Synthesis
The open question for this region isn't whether AI compute gets priced against the clock—most of it already sits downstream of grids that price that way today, for reasons that predate AI entirely. The open question is which operators structure PPAs and workload scheduling for that reality ahead of time, and which absorb a mismatch on both sides at once, the way Vietnam's data centres did in December 2025. A weaker, secondary channel is worth naming rather than overclaiming: China's own AI-driven demand could eventually influence global LNG and coal prices that several of these markets import against—a real but indirect and currently unquantified link, distinct from and much smaller than the pricing-practice diffusion above.
A second finding, underneath the first
Building this console surfaced something not originally being looked for: the markets that were hardest to source cleanly here are not a random subset. Cambodia carries no citation anywhere in this console. Taiwan's tariff clock windows are shown as illustrative pending verification. Indonesia's and Cambodia's own tariff structures are flat in practice, so there is little granular pricing data to publish in the first place. These are also, not coincidentally, the markets earliest in smart-meter rollout and wholesale-market modernisation. Singapore can publish a wholesale price every thirty minutes because its metering and market infrastructure make that possible. Cambodia cannot yet, at scale. That is not a gap in this research so much as a second, correlated finding sitting underneath the first: data transparency and time-of-day pricing sophistication move together, market by market. As smart-meter and market-reform programmes reach the markets currently hardest to see into, the expectation isn't just better data for a console like this one—it's that those markets become mechanically capable of the same kind of pricing this whole page has been about. Worth watching as its own signal, not just a footnote on this one.
Tariff structure
Data centre relevance
PPA market
Model a data centre against this grid
Modelled outcome
Assumptions and formulae
Every number here is either sourced to a named tariff order or market operator, or modelled by one of the formulae below.
Block tariffs
India, Thailand, Vietnam, and Taiwan set a small number of fixed time windows, each with a published rate, unchanged until the next tariff order. The dial for these markets shows the weekday pattern only.
Continuous settlement
Singapore, Australia, New Zealand, and the Philippines settle wholesale power every 30-60 minutes against a market-clearing price. The 48-segment ring represents this mechanism, not a plotted price series.
Flat tariffs
Indonesia and Cambodia apply the same rate regardless of time of day for most categories.
Peak/off-peak spread formula
Where a full published schedule was not used directly, peak and off-peak rates are derived from the C&I average and the stated spread:
Load-shift and PPA blend formula
Inference cost per subscriber
5 kWh of grid-side energy per active subscriber per month is a placeholder used to show the mechanism, not a measured or industry-reported figure.
AI share of data centre load
No market publishes this breakdown. The figure shown per country is a modelled index reflecting known AI infrastructure investment intensity.
Adoption index
A 1-10 directional index combining population scale, internet penetration, and observed AI infrastructure investment. The only measured figures on this page are the two India user counts in the spotlight above.
Composite exposure score
A 0-10 index built entirely from figures already in this console, weighted 40% spread, 30% DC share of grid, 30% adoption index. It is a synthesised read, not a sourced or industry-standard metric—treat it as directional shorthand for "which markets carry the most compounded exposure," not a benchmark.
Generation mix sourcing
India and Taiwan reflect IEA/Ember/EIA reporting for the latest full year. Singapore, Thailand, Vietnam, Indonesia, the Philippines, Australia, and New Zealand reflect IEA/Ember/national-operator reporting rounded to the nearest whole point. Cambodia's mix is the least well-sourced entry and should be treated as indicative pending verification against EDC's own reporting.
Representative tariff examples used per market
India: Maharashtra's C&I tariff, not a national average. Thailand: MEA/PEA Type 4 large industrial (69kV and above). Vietnam: the 22kV business tariff band, the classification third-party data centres were moved onto in December 2025. Taiwan: Taipower's general industrial structure; current time-of-use clock windows are shown as illustrative pending direct verification against Taipower's published schedule.
Where each spread percentage comes from, market by market
Directly sourced from a published rate schedule: India, Thailand, Vietnam—the spread is calculated from the actual peak and off-peak rates cited in References. Estimated from typical settlement volatility: Singapore, Australia, New Zealand, Philippines—these markets have no fixed peak/off-peak rate to cite; the spread is an informed approximation of how far the continuously-settled price typically swings, not a single official number. Structural, not estimated: Indonesia and Cambodia's low spreads follow directly from a flat or near-flat tariff—there's nothing to estimate. Illustrative pending verification: Taiwan's spread is built from a representative peak/off-peak assumption, flagged elsewhere as not independently verified against Taipower's current schedule. Every spread figure not in the first category should be treated as directional, not quoted as an official rate.
References
Primary sources behind every sourced figure in this console, numbered to match the [n] citation markers used throughout. Modelled figures are documented in Assumptions instead of cited here. Links marked with an asterisk go to the source organisation's homepage rather than the exact document, where a stable direct link wasn't available.
Tariff orders and market operators
- Mercom India – Maharashtra MYT Order coverage, reporting MERC's Modified Multi-Year Tariff Order (FY26)
- Provincial Electricity Authority (PEA) – Time-of-Use tariff schedule, set under the Energy Regulatory Commission (ERC) framework, May-Aug 2026
- Vietnam Ministry of Industry and Trade / EVN – Decision 1279/QD-BCT (9 May 2025), Decision 963/QD-BCT (22 April 2026)
- Circular 60/2025/TT-BCT – Vietnamnet coverage – data centre tariff reclassification, Vietnam (effective 2 Dec 2025)
- GlobalPetrolPrices.com – Indonesia residential/business electricity price tracking, cross-checked against PLN's published tariff schedule
- Energy Market Authority, Singapore – average monthly Uniform Singapore Energy Price (USEP) statistics
- Australian Energy Regulator – Default Market Offer Determination 2026-27 (final); Australian Energy Market Operator*, National Electricity Market data
- Electricity Authority New Zealand*; MBIE Quarterly Survey of Domestic Electricity Prices, 15 Nov 2025 edition (direct PDF)
- Manila Electric Company (Meralco) – official monthly rate advisory, June 2026
- Taiwan Power Company – Understanding Tariffs Q&A, summer/non-summer industrial time-of-use structure
Generation mix and market structure
- IEA – Global Energy Review 2025; Southeast Asia Energy Outlook 2024
- Ember – Global Electricity Review 2025; Global Electricity Mid-Year Insights 2025
- EIA – Taiwan Analysis Brief, April 2026
AI adoption and the ToD pricing trigger
- Sensor Tower – State of AI 2026 report, May 2026 figures
- Business Standard – Sensor Tower India-specific user counts, 17 June 2026
- TechNode – DeepSeek V4 peak-hour API pricing coverage, 30 June 2026
*Homepage link—the specific tariff order, decision, or dataset was sourced from this organisation but a stable deep link to the exact document wasn't available at time of writing. Cambodia carries no numbered citation anywhere in this console—its figures are flagged in Assumptions as the least well-sourced entry and are not tied to a specific document.