LehmanSoft Japan

AI-VPP Commercial Operation of a Grid-Scale Battery in Japan: Lessons From a Real 2MW/8.1MWh Site

How an AI-driven virtual power plant operator runs a real 2MW/8.1MWh grid-scale battery in Japan across JEPX and the balancing market—honest lessons for infra investors.

Overview

Japan has become one of the most talked-about battery storage markets in the world. Grid-connection applications for standalone batteries jumped from roughly 70 GW in mid-2024 to 170.8 GW by the end of 2025, while only a small fraction of that applied capacity has actually secured grid connection and reached operation (Energy-Storage.news). That gap between ambition and connected reality is the whole story for an infrastructure investor: the spreadsheets are easy, the operations are hard, and the markets keep moving under your feet.

This article is written from the operator's side of that gap. We run a real, commercially operating 2MW/8.1MWh grid-scale battery in Saitama, dispatching it across the JEPX wholesale market and the balancing market (需給調整市場). We are not selling a forecast of what might work in Japan—we are describing what it actually takes to keep a battery earning when market rules change, prices compress, and the optimization has to happen automatically, every 30 minutes, every day. We call this approach AI-VPP: an AI-driven virtual power plant layer that decides where each block of energy and capacity is most valuable, in real time, across multiple markets.

If you are an overseas infrastructure fund, IPP, or developer evaluating a Japan BESS thesis, the lessons below are the ones the pitch decks tend to skip.

Why "AI-VPP" Is Not a Buzzword in the Japanese Market Specifically

In a market with one liquid revenue stream, a battery is almost a static asset—you buy low, you sell high, and a simple rule captures most of the value. Japan is the opposite. A grid-scale battery here can earn from at least three distinct, simultaneously available mechanisms:

  • JEPX spot (wholesale arbitrage) — the day-ahead market trades electricity in 30-minute blocks (48 slots per day). Value comes from charging in cheap hours and discharging in expensive ones (TOCOM / JPX).
  • The balancing market (需給調整市場, traded via EPRX) — payment for holding capacity that responds to frequency and supply-demand imbalance, across products from fast primary reserve down to tertiary (EPRX).
  • The capacity market and the Long-Term Decarbonization Auction (LTDA) — longer-dated availability payments, with their own duration and obligation rules.

The hard part is not participating in any one of these. It is deciding, slot by slot, how to split a finite battery between them—because every MW and every MWh you commit to one market is unavailable to another, and state of charge is a shared, depleting resource. That allocation problem is combinatorial, price-dependent, and time-coupled. It is exactly the kind of problem an AI optimization layer is built for, and exactly the kind of problem a human trader cannot solve 48 times a day across hundreds of price scenarios.

That is what we mean by AI-VPP: forecasting JEPX and balancing prices, then solving the multi-market dispatch and bidding allocation automatically, under the real physical constraints of the asset. The "we run it ourselves" part matters because the optimization is only as good as the operational feedback it learns from.

The Honest Part: The Balancing Market Premium Is Being Squeezed

Any operator who tells you Japanese battery returns are a one-way bet is not someone you should trust with your capital. The most important recent development is that the balancing market is getting less generous, on purpose.

For fiscal year 2024, batteries in the balancing market settled composite products at an average of about 15.70 JPY per ΔkW/30min—well above the all-resource average of 5.77—reflecting how valuable fast battery response was (Greenberg Traurig). But that premium has been compressing. After mid-year procurement-volume reductions, the same products averaged about 10.84 JPY per ΔkW/30min by April 2025 (Greenberg Traurig).

It goes further. The decided change, effective 13 March 2026 (delivery 14 March), cuts the price cap on premium fast-response products (primary, secondary-1, and composite) from 19.51 to 15 JPY per ΔkW/30min, with a conditional, staged further reduction toward 10 and then 7.21 JPY per ΔkW/30min only if competition does not improve (Greenberg Traurig; National Law Review). (An earlier proposal that would have cut straight to 7.21 JPY was not the version adopted.) The policy direction is clear: regulators want to remove the premium that made the balancing market the headline revenue line for early batteries.

On top of pricing, the market's structure is changing. From the trade on 13 March 2026 (delivery 14 March), the weekly products for primary through tertiary-1 reserve shift to day-ahead trading, with adjustments to procurement volumes and bidding requirements designed to lower barriers and improve price formation (EPRX; METI/ANRE).

The lesson for investors: a model that underwrites a Japanese battery primarily on FY2023–FY2024 balancing-market prices is underwriting a market that no longer exists. The revenue line that looked best last year is the one being actively reduced. This is not a reason to avoid Japan—it is a reason to demand an operator whose value capture does not depend on any single market staying rich.

What Multi-Market Optimization Actually Buys You

If the balancing premium is shrinking, where does durable value come from? From the spread between markets and within the day—and that spread is widening, not narrowing.

Japan's rapid solar build-out has produced a deep "duck curve." Midday JEPX prices now regularly collapse toward 0.01 JPY/kWh when solar floods the system, while morning and evening prices stay high—monthly averages have peaked near 15 JPY/kWh in winter (TOCOM / JPX). For a battery, volatility is the product. Cheaper, more frequent midday troughs and sharper evening peaks mean the arbitrage opportunity on JEPX is structurally growing even as ancillary payments fall.

The role of an AI-VPP layer is to capture that without leaving balancing-market and capacity revenue on the table. In practice that means:

  1. Forecasting both JEPX and balancing prices with enough accuracy to know, the day before, where each block of capacity is worth most.
  2. Co-optimizing energy and reserve so the battery isn't fully committed to arbitrage when a balancing bid would pay more, and isn't sitting idle holding reserve when a price spike makes discharge the better trade.
  3. Respecting the physics—round-trip losses, throughput limits, and state-of-charge constraints that determine how many cycles the asset can sustainably run.
  4. Adapting as rules change—re-tuning bidding strategy as the price cap drops and products move to day-ahead from 13 March 2026, rather than running last year's playbook into a different market.

A static, single-market battery cannot do this. A team of humans cannot do it at 30-minute granularity across three markets indefinitely. This is precisely where an operator's AI layer earns its keep—and why "who is running the asset and how" is a due-diligence question as important as the CAPEX number.

The Duration Shift: Why 6 Hours Is Becoming the Reference Point

There is a second structural signal investors should not miss. In the second LTDA round (results April 2025), Japan awarded about 1.4 GW of BESS across 27 projects; batteries were the most oversubscribed technology, with only ~20% of bid capacity awarded (ESS News). The third (FY2025) LTDA round (results May 2026) selected about 1.25 GW of battery storage across 19 projects (of which BESS-only cleared ≈575 MW)—but limited eligibility to projects with 6-hour duration or longer, a clear move away from the shorter-duration designs that dominated early rounds. This ≈1.25 GW is the LTDA battery total, not the capacity market (容量市場) main auction—the two are distinct instruments (Energy-Storage.news; National Law Review).

The market is signaling that it increasingly values energy shifting (longer duration) over pure fast-response capacity (shorter duration). That reinforces the same conclusion: the future of Japanese battery returns leans toward energy arbitrage and flexible multi-market dispatch, not toward a single ancillary-services premium.

A 2MW/8.1MWh asset like ours—roughly 4-hour duration, high-voltage grid connected—sits in the pragmatic middle: large enough to be commercially meaningful, small enough to iterate the operating model quickly, and configured to play JEPX arbitrage and balancing simultaneously. The point of running it is not the headline megawatts; it is that every rule change since commissioning has been absorbed in live operations, not in theory.

What a Real Operating Site Teaches That a Model Cannot

We deliberately keep the internal numbers—revenues, state-of-charge profiles, specific dispatch decisions—confidential, because they are commercially sensitive and because no responsible operator publishes them. What we can share are the lessons, which are more valuable to an investor than any single month's revenue figure:

  • Forecast error is the real cost line. The gap between a backtested optimizer and a live one is almost entirely forecast error. You only close it by operating through real days, including the bad ones.
  • Rule changes are continuous, not occasional. Between commissioning and today, the balancing market has changed procurement volumes, announced cap reductions, and restructured to day-ahead trading. An asset is only as resilient as the team re-tuning it.
  • Grid-connection reality dominates the timeline. With 170.8 GW of applications chasing limited connection capacity, having a connected, operating asset is itself the scarce thing (Energy-Storage.news). Operational credibility is hard to fake and harder to buy.
  • No single market is the answer. The honest version of the Japan battery thesis is a portfolio of revenue streams, actively optimized, with the mix shifting over time. Anyone underwriting on one rich line is underwriting the wrong asset.

The Takeaway for Overseas Capital

Japan's battery opportunity is real and large, but it is not the simple ancillary-services gold rush some 2023-vintage models assumed. The balancing-market premium is being deliberately compressed from 13 March 2026; JEPX volatility and the value of energy shifting are growing; and the auctions are pushing toward longer duration. The winners will be the assets operated by teams who can re-allocate value across markets, automatically, as the rules move—and who have the operational scar tissue to prove it.

That is the case we make from the operator's chair, with a commercially running 2MW/8.1MWh grid-scale battery in Saitama participating in JEPX and the balancing market today. We are not modeling Japan. We are running in it.

Work With an Operator, Not a Spreadsheet

If you are an infrastructure fund, IPP, or developer building or underwriting a Japanese grid-scale battery portfolio, talk to a team that already operates one. We can walk you through how AI-VPP multi-market optimization works on a live asset, what the 13 March 2026 balancing-market changes (the price-cap cut to 15 JPY and the move to day-ahead trading) mean for your underwriting, and where the durable value sits now that the balancing premium is being reduced.

Contact us to arrange an operator-to-investor briefing on AI-VPP grid-scale battery operation in Japan.

Sources

Sources:

→ Grid-scale battery operation & aggregation service in Japan