The AI Stack on IDX: Indonesia Owns Both Ends of the Chain and None of the Middle

Companion post to [The AI Stack: Why Where a Stock Sits Matters More Than Whether It Says "AI"]. September 2026.

Run the eleven layers of the AI supply chain against the Indonesia Stock Exchange and something uncomfortable shows up immediately.

There is no Indonesian chip designer. No memory manufacturer. No optical networking company. No foundation model. No enterprise AI platform. No robotics or drone manufacturer of any scale. Eight of the eleven layers have no listed representative at all.

What Indonesia does have sits at the two ends of the chain: the minerals that go into the hardware, and the buildings the hardware runs in. Nothing in between — which is, not coincidentally, where the margins are.

That is not a reason to skip the theme. It is a reason to be precise about what you are actually buying, because the gap creates a specific trap: when eight layers are missing, local money crowds into the three that exist, and valuations detach from what those companies actually operate. That has already happened.

First, a framing problem

The IDX has no official "data center" or "AI" sector. The exchange's IDX-IC classification — 12 sectors, 35 subsectors, 69 industries and 130 sub-industries, in force since 25 January 2021 — files data center operators under Technology, specifically IT Services & Consulting. That puts them in the same bucket as digital platforms like GOTO or EMTK, despite completely different economics: a data center earns long-term recurring rent, a platform earns transaction commissions off user growth.

So the theme has no official boundary. Anyone can call anything an AI stock, and many do. That is the whole reason the layer test is worth running.

Layer 1 that exists: Data Centers & Digital Infrastructure

This is the real one. Indonesian data center capacity sat around 580 MW in the first half of 2026 and is projected to reach about 3.5 GW by 2030 — roughly a sixfold expansion. The demand is not imaginary.

DCII (DCI Indonesia) — the only genuine pure-play, and the first Tier IV certified data center in Southeast Asia. Its March 2026 public expose put installed capacity at 128 MW across four locations:

The business underneath is real and profitable: FY2025 revenue of Rp2.54 trillion, up 40.1%; EBITDA of Rp1.55 trillion at a 61.0% margin; net profit of Rp1.002 trillion, up 25.7%; more than 270 customers, roughly 80% of them multinationals. The Bintan campus is positioned to catch spillover demand from Singapore, where land and power constraints have throttled expansion.

Then the valuation. DCII closed near Rp201,100 on 10 September 2026 at a PER of 327x and PBV of 104x — against an IPO price of Rp420 in January 2021.

Hold the two numbers next to each other. The 128 MW is what earns revenue today. The 2,000 MW that justifies the multiple is development potential with no customers attached to it. That is the entire question on this stock, and it is the same question the US series asks of CoreWeave and Nebius: how much of the price is contracted, and how much is a pipeline?

DSSA (Dian Swastatika Sentosa) — operates the SM+ network with SMX01 as the next project. The weakness as a data center proxy is structural: mining, energy and other group businesses can bury the data center contribution entirely. A re-rating catalyst here would be the company simply reporting data center revenue and profit with more transparency.

MGLV (NexAI Digital Infrastructure) — up more than 500% year-to-date toward roughly Rp15,000, while still posting a net loss of Rp14.10 billion through June 2026. Its NAC subsidiary operates 6 MW and independently booked Rp36.3 billion of revenue and Rp9.49 billion of net profit — but those figures have not entered MGLV's own financial statements, and a rights issue of up to 285.7 million shares still needs executing. The market is valuing a transformation story and assets that have not arrived yet. That is a legitimate thing to bet on. It is not the same thing as buying an operating business, and the price should not be read as if it were.

TLKM (Telkom Indonesia) — building through NeutraDC, targeting roughly 500 MW by 2030 per its 2026 public expose, including a Batam hub for regional demand. This is the only large-cap way to own the theme where, if the data center thesis disappoints, you still hold a real underlying business.

EDGE (Indointernet) — its largest inner-city facility handles up to 23 MW of load; EDGE1 runs up to 6 MW of IT load. Small relative to DCII, and priced accordingly.

Industrial estates — DMAS, SSIA, KIJA, BEST — the land beneath the buildings. One step removed from the capex cycle and typically cheaper than the operators. Worth noting: I have not verified individual signed data center land contracts for these four. Until you can point to one, they belong in the narrative column below, not the operating one.

Layer 2 that exists: Power — mostly narrative so far

This is where the sorting test earns its keep. Apply one question — is the AI-related demand in the revenue, or only in the press release? — and most of the "AI power" names on IDX fall over:

• PGEO (Pertamina Geothermal) — geothermal is genuinely well suited to data centers: stable, low-carbon, 24/7 baseload. But the company is at joint-study stage with IDPRO and Universitas Indonesia. No site. No customer. No PPA.

• PGAS — around 273 km of gas network in Batam, but no dedicated contract with a data center operator.

• MEDC — has an MoU with NeutraDC Nxera Batam for renewable supply. An MoU is not a power purchase agreement, and the gap between them is measured in years.

• TOWR — 36,900 towers and 182,600 km of fiber, with no named data center contract.

• MTEL — edge computing potential, no contract material to revenue.

Five names routinely listed as AI plays, and not one of them has AI-linked revenue on the books.

POWR (Cikarang Listrindo) is the closest thing to an exception, because it already sells power into the industrial estates where data centers are physically being built — a revenue link that exists rather than one that is anticipated. Verify the specifics before you act on that; I have not confirmed a named data center PPA either.

Coal generators (ADRO, PTBA, ITMG) feed the grid that ultimately powers all of this, but the connection is too diffuse to call it an AI thesis. If you own them, own them for coal.

Layer 3 that exists: Critical Materials

TINS (Timah) is the most direct semiconductor link on the entire exchange, and the least discussed. It is Indonesia's only integrated tin producer, with downstream projects covering value-added products including tin chemical and tin solder rather than simply exporting ingots. Every chip that reaches a circuit board is soldered on. That is a genuine position in the chain, not a thematic association.

Copper — AMMN and MDKA. LME three-month copper set a record on 7 September 2026, rising 0.9% to $14,510 per ton after touching an all-time high of $14,533. Data centers and grid buildout are a real part of that demand.

One correction worth making, because it is widely misunderstood: MDKA now functions largely as a holding company. Of roughly $620.3 million in revenue through March 2026, nickel via MBMA contributed about $454.2 million — over 73%. Buying MDKA for copper exposure mostly buys you nickel. AMMN is the cleaner copper position.

Nickel — INCO, NCKL, MBMA. Handle with care. Nickel is an EV battery story far more than an AI story. Filing it under AI because it is a critical mineral is precisely the error this whole series argues against.

The sorting table

Everything above, reduced to the only distinction that matters:

Column three is not worthless — early positioning sometimes pays enormously. But you should know which column you are buying, and you should not pay column-one prices for column-three assets.

The eight missing layers

Chips, memory, networking and photonics, foundation models, enterprise platforms, apps and robotics, edge AI and drones, and frontier tech — none of these have an IDX-listed representative of meaningful scale.

For an investor this means one practical thing: the parts of the AI chain with the highest margins and the widest moats are not available to you domestically. If you want exposure to them, you need US market access. If you do not have it, the honest position is that you are investing in Indonesia's role in AI — landlord and supplier — rather than in AI itself. Those are different bets with different returns, and conflating them is how people end up disappointed by a theme that worked.

For a country it means something larger, but that is a different article.

What to watch

• DCII utilisation, not capacity announcements. The number that matters is how much of the 128 MW is contracted and at what rate — not the next pipeline figure.

• MGLV's rights issue and the NAC consolidation. Until NAC's earnings appear in MGLV's own statements, the story is unverified.

• Any MoU that converts to a PPA. The moment PGEO, MEDC or PGAS signs a real offtake with a named data center operator, that name moves columns. That is the single highest-value event to watch in the local theme.

• Copper and tin prices. AMMN and TINS will move on LME regardless of how well they execute.

• DSSA segment reporting. If it starts breaking out data center revenue, the market can finally price it.

Disclosure and method

The author holds positions in some of the companies discussed in this series, and those positions change over time. Individual holdings are not disclosed. No compensation has been received from any company mentioned, and nothing in this series is sponsored.

Figures are drawn from company public expose materials, exchange filings and published research, gathered in September 2026. Where I could not verify something — noted inline — I have said so rather than filled the gap with an estimate. Prices and multiples cited are as of the dates given and move daily.

This is research and general information, not personalized investment advice. Do your own work before acting on any of it.

Companion to the 11-part AI Stack series covering US-listed names.