Thailand Data Center Investment — Market Map, Competitive Field, Deal Sources
Scope: Thailand primary, Philippines and the region as comparison. Research, not interview positioning. Revision note: rewritten 30 August 2026 after a multi-agent verification pass refuted sixteen claims in the first draft, including the wheeling charge the original economics were built on. Companion pages: th-power-sector for the regulatory detail, datacenter-foundations for the physical and financial fundamentals.
The thesis in one paragraph
The first version of this page called Thailand a growth story. That was wrong, and the correction is the thesis. Thailand has roughly 26 GW of data center grid applications sitting against about 4.9 GW the distribution utility can actually confirm, of which only around 122 MW of IT capacity was live at the end of 2025. Bangkok colocation vacancy is 23.3%, the highest of any major Southeast Asian market. The Board of Investment has taken no new data center promotion application since April 2026 and has paused consideration while it rewrites the criteria. This is not a market short of announcements. It is a market where almost nothing announced has been connected, and where the state has just stopped the queue to work out who gets power. Every attractive position in Thailand sits on the supply side of that constraint, in generation and connection rather than in colocation, because colocation is where the vacancy is and power is where the scarcity is.
The gap between announced and connected is the whole market
Start with the numbers that actually bind, in the right order.
As of June 2026, 95 data center projects had applied to the Provincial Electricity Authority for 26,045.2 MW of connection capacity. PEA could confirm 4,882.7 MW, or 18.7%, split into 3,240.7 MW firm and 1,642 MW conditional. Signed PPAs covered 2,768.2 MW across 18 projects. That leaves 21,162.5 MW with no path to power. The figures come from Thansettakij, 25 July 2026, reporting PEA's book, not from a PEA release, and they cover PEA's provincial territory only. Bangkok, Nonthaburi and Samut Prakan sit in MEA's area, and no MEA equivalent has been published. EGAT's direct-customer channel for data centers of 200 MW and above is also outside this count.
Now the other end. Knight Frank's Data Centre Atlas 2026, published 25 June 2026 on Q1 2026 data, puts Bangkok's top ten operators at 3.9 GW of total capacity against 122 MW of live IT capacity at end-2025, forecast to reach 402 MW by 2027. The same report puts Bangkok colocation vacancy at 23.3%, against Johor at 0.7%, Singapore at 4.9% and Jakarta at 20.5%.
Read those together. Thirty-two times more capacity has been applied for than the utility can confirm, more than thirty times more has been announced than is running, and nearly a quarter of what is running is empty. The commonly quoted 2.87 GW pipeline figure is a Black & Veatch estimate given to Nation Thailand on 13 April 2026, a consultant's number rather than a regulator's, and it has been overtaken by PEA's own application book.
Then the queue stopped. BOI Secretary-General Narit told The Standard on 5 August 2026 that no new data center promotion applications have been filed since April 2026, and Krungthep Turakij reported on 30 August 2026 that Thailand has paused consideration of data center projects while new criteria are written. Whether that is a deliberate freeze or a market that ran into the power wall by itself is the single most useful question to put to anyone in Bangkok right now.
What the state has actually approved, and what it has only announced
Investment approvals are the number everyone quotes and the number least worth trusting, because the counting bases differ and nobody says which one they are using.
The sourced set: 36 projects and THB 728 billion approved during 2025, per Black & Veatch via Nation Thailand; 26 projects and THB 498,716 million approved in 2025 plus 8 projects and THB 162,044 million so far in 2026, per Krungthep Turakij on 30 August 2026; and 141 digital infrastructure projects worth THB 623 billion in 2025 on a wider definition. January 2026 added seven data center projects at THB 96.88 billion, including 223 MW of True IDC capacity. March 2026 added three at THB 90.9 billion from Beijing Haoyang, Empyrion Digital and GSA Data Center 02. A May 2026 figure of THB 958 billion circulates and I could not source it to a BOI release; it is an order of magnitude above the other rounds and should not be quoted.
Do not cite a BOI total without saying which basis it is on. Approved investment value is not committed capital, and neither is delivered megawatts.
The incentive itself is stable and worth knowing precisely. A high-efficiency data center gets an eight-year corporate income tax exemption capped at 100% of project investment value. In exchange at least half of management and expert positions go to Thai nationals within three years, with human resource and local ecosystem commitments attached. Since 2026 a sponsor must also demonstrate it can be powered before it can apply for promotion, which inverts the old sequence of winning promotion and then hunting for electrons.
Readiness is now tested on four dimensions: electricity supply and clean energy transition, water management that does not compete with local communities, environmental protection, and tangible benefit to Thailand through AI talent, SME supply chain integration or research presence.
Power costs more than the market thinks, and the number everyone quotes is wrong
This section replaces the worst error in the first draft.
There is no cheap wheeling. The ERC's public consultation document, approved at its 981st meeting on 22 October 2025 and consulted from 27 October to 10 November 2025, sets Third Party Access charges for the Direct PPA pilot by voltage level with a separate peak demand charge, explicitly not on a postage-stamp basis. At 69 kV and above the wheeling charge is 1.1218 THB/kWh, made up of transmission at 0.6868 and distribution at 0.4350, plus a peak demand charge of 74.14 THB/kW/month. At medium voltage it is 1.9110 THB/kWh plus 132.93 THB/kW/month. At low voltage, 2.5235 THB/kWh plus 210.00 THB/kW/month. On top sits a system security and ancillary services charge of 0.4978 THB/kWh and a policy expense charge of 0.1447 THB/kWh, the latter reset every four months with the Ft.
So the all-in grid charge for a large data center wheeling at 69 kV and above is about 1.7643 THB/kWh before the demand charge, plus imbalance charges and a connection fee. The 1.07 THB/kWh postage-stamp figure that circulates in market commentary traces to a solar installer's marketing page and does not exist in any ERC document. Any model built on it understates grid cost by roughly two thirds.
The green tariff is the only clean product you can actually buy today, and its pricing is also widely misstated. UGT1 is priced as base tariff plus a premium, not as an absolute rate: a wholesale premium of 0.0314 THB/kWh to MEA and PEA and a retail premium of 0.0375 THB/kWh for EGAT direct customers, excluding VAT, per ERC meeting 3/2569 of 21 January 2026 and EGAT Announcement 6/2569 of 12 March 2026. It draws on seven existing EGAT hydro plants. UGT2 is the large-user product covering data centers, drawing on a portfolio of new solar, wind and biogas.
The separate data center tariff class is not a tariff yet, and the widely reported THB 5 to 6 per kWh is not the mechanism. NEPC approved the class in principle on 15 July 2026. What was approved is a marginal cost pass-through, under which data centers pay the actual incremental generation cost, largely imported LNG, rather than a fixed rate. Energy Minister Akanat Promphan said on 28 August 2026 that at spot LNG of USD 23 per MMBtu the fuel component alone exceeds THB 5/kWh. The all-in number therefore floats with gas and can sit above the reported band. The ERC has published nothing. For reference, the general September 2026 tariff is THB 3.86 per unit.
The practical consequence for an investor is that Thai grid power for a data center is not cheap, is not fixed, and is not yet quotable. That is a risk to price, and it is also the reason generation next to load is worth so much.
The Direct PPA cap is gone, but the code is still not in force
The chronology matters because so many people are working from a stale version of it.
NEPC approved a Direct PPA pilot using Third Party Access in June 2024, capped at 2,000 MW and restricted to data centers. The ERC published draft criteria in October 2025. The scheme was slated to take effect in January 2026 under the Quick Big Win banner. On 15 July 2026 NEPC approved a seven-item electricity package that extended Direct PPA beyond data centers to industry generally.
Then the cap went. Energy Minister Akanat Promphan said on 28 August 2026 that the scheme now carries no quota limit. That is two days old at the time of writing and it is not yet reflected in most commentary.
What has not happened is the part that matters. The TPA Code is not gazetted. The three electricity sub-codes covering service, connection and operation are not public. No executed Direct PPA has been confirmed. No capacity booking portal exists at any published URL, despite advisers telling clients to prepare for one. Eligibility remains demanding: BOI promotion, an IT baseload of at least 50 MW per building, a ten-year energy plan, and a 100% renewable commitment for new BOI-promoted data centers.
The same NEPC meeting on 15 July, its 176th, chaired by Prime Minister Anutin Charnvirakul, also did something the market has largely missed. It approved fixing a definite end date for every Non-Firm SPP and VSPP power purchase agreement carrying automatic renewal, and resetting purchase rates to current cost, with solar specified at THB 2.1579/kWh, directing EGAT, MEA and PEA to renegotiate. That is a retroactive squeeze on legacy renewable contracts and it affects the balance sheets of the same generators now bidding for data center load.
Capacity is rationed by bank guarantee, and by a lock-up that reaches into your cap table
Two instruments now govern who gets power, and the second one is the more consequential for a dealmaker.
The Power Commitment Guarantee was set by the ERC at THB 4.5 million per MW of reserved capacity, disclosed by ERC Secretary-General Poonpat Leesombatpiboon on 17 April 2026, approved for size by the Energy Policy Administration Commission in April, and endorsed in principle by NEPC on 15 July 2026. Half is released once the project reaches 50% of proposed consumption within one year, the balance at 70%. A separate use-your-reserved-capacity-or-lose-it condition of five to seven years sits in the ERC's screening criteria.
It is a bank guarantee, not a cash deposit, and the first draft of this page got that wrong. A 100 MW campus posts a THB 450 million instrument, roughly US$14 million of face value, but the cash cost is the annual guarantee fee, not the principal. That is a credit test rather than a funding test. It still screens out the developer with land and grid position but no bank relationship, which is exactly the counterparty worth partnering with.
The lock-up is the sharper instrument. The Royal Gazette published PEA's Notification on Criteria for the Change and Transfer of Power Users for Data Center Operators and Large Industrial Businesses with High Power Demand on 4 June 2026, effective 5 June 2026. Original shareholders must retain more than 50% for three years after a transfer. That lands directly on the acquisition thesis: buying a stalled but power-confirmed project is no longer a clean way to inherit its grid position, because the seller has to stay in.
The grid build is bigger and later than the near-term substations suggest
Two separate EGAT efforts get conflated, including in the first draft of this page.
The near-term package is the one EGAT described in its release of 16 February 2026, around a 13 February site visit by Permanent Secretary Prasert Sinsukprasert, quoting Deputy Governor Teerawut Wetatham. In EGAT's words it establishes "three new delivery points for the Provincial Electricity Authority (PEA), namely 150 MW Rayong 2 Substation, 250 MW Phanthong Substation, and 250 MW Phanthong 2 Substation (Temporary), and the transformer capacity expansion at Pluak Daeng Substation of 350 MW and at Sattahip 1 and 2 Substations of 150 MW in total," sized against urgent data center demand of 1,150 MW. Rayong 2 is complete. Phanthong and Phanthong 2 were scheduled for February and December 2026. Pluak Daeng is expected in the first quarter of 2027. No date is given for Sattahip. This release carries no monetary figure and its status is as at February 2026.
The larger effort is the Transmission System Expansion Project to Accommodate Large Power Consumer, the TSLC, a THB 31,050 million EGAT project covering Chachoengsao, Chon Buri and Rayong, running 2027 to 2033, sized to serve the first 16 large data center customers totaling 3,817.6 MW. The THB 31 billion figure that circulates attached to the near-term substations belongs here, not there.
Those sixteen customers are worth chasing by name. Deputy Prime Minister and Finance Minister Ekniti Nitithanprapas and the BOI Secretary-General have both pointed to the companies holding PEA electricity confirmation letters as the projects that are actually real. That list, not the announcement pile, is the live market.
Timing is the other thing developers underprice. Knight Frank Thailand, via Thansettakij on 14 May 2026, puts grid connection alone at 12 to 18 months for loads above 20 MW, against hyperscale sites needing 30 to 100 MW or more.
One development could reprice all of this. The Ministry of Energy is preparing an amendment to let EGAT sell directly to large data centers sited near its 230 kV and 500 kV network. If that lands, the scarcity migrates from distribution connection to transmission adjacency, and land near high-voltage corridors reprices.
Where the wires and the water are
The pipeline concentrates in the Eastern Economic Corridor. Of the 36 projects approved in 2025, Rayong took 33% and Chonburi 32%, with Chachoengsao carrying most of the rest. The EEC transmission system was designed for conventional industrial load, spread out and predictable, not for hyperscale campuses arriving concentrated and at once.
Chachoengsao is short on both electricity and water. Parts of Chonburi have the same problem while remaining where investors most want to be. The Ministry of Energy has named Samut Prakan and Bangkok as having spare transmission headroom, which is MEA territory and therefore outside the published PEA queue entirely.
Water is becoming a policy instrument rather than a constraint to manage quietly. BOI has signalled a special high water tariff for data centers, alongside mandatory cooling system inspection, water recycling and a drought plan.
Who is actually building
Capacity leadership sits with SUPERNAP Thailand, True IDC and STT GDC Thailand, differentiating on Tier III and IV certification, early liquid cooling and anchor tenancy with banks and global software vendors.
| Operator | Backing | Thai position | Location | Status |
|---|---|---|---|---|
| True IDC | CP Group via Ascend Group (100%), outside listed True Corp; GIP backing the EEC project | 223 MW across 3 BOI projects, US$1.4bn | Chonburi, Samut Prakan | Approved Jan 2026 |
| GSA Data Center | JV of Gulf, Singtel, AIS | 120 MW combined IT load | Rayong, Samut Prakan | Entities 02 and 05 approved |
| STT GDC Thailand | ST Telemedia; KKR and Singtel acquiring | STT Bangkok 2, US$200m | Bangkok | Building since Mar 2025 |
| SUPERNAP Thailand | NTT | Capacity leader; Bangkok 2, 3 and 4 | Bangkok, Amata City Chonburi | Operating and expanding |
| DayOne | Ex-GDS International; SoftBank Vision Fund, Coatue, Griffin, Baupost, INA | Chonburi Tech Park, US$1bn over 5 yrs, 180 MW grid | Amata City Chonburi | Broke ground Mar 2025 |
| Beijing Haoyang | Apollo Global Management among backers | 300 MW, THB 72.7bn | WHA ESIE 4, Rayong | BOI approved; first move outside China |
| Empyrion Digital | Singapore | TH1, 20 MW, 17,000 sqm | Bang Na, Bangkok | Q4 2026 launch |
| Bridge Data Centres | Bain Capital, merged with Chindata as WinTriX | BDC IIO Thailand, THB 24.6bn | Thailand | Bain reported selling a stake |
| DAMAC Digital (Edgnex) | DAMAC | 84 MW | Bangkok | In Bangkok top ten |
| Evolution Data Centres | with Central Pattana | 40 MW, 200 MW planned | Bang Na, Samut Prakan | Developing |
Bangkok's top ten by capacity also include Equinix, Singtel, Doma Infrastructure Group, ZDATA, Hoyinn Technologies and Galaxy Peak. PEA's June 2026 census counts 41 operating facilities.
True IDC deserves precision because it is the most Thai of the serious operators and the most likely partner. It is owned outright by CP Group through Ascend Group, its digital unit, and sits outside the listed True Corporation, so there is no public-market route in and any position is privately negotiated. It runs four Thai facilities at Bangna, Muangthong, Ratchada and Pattanakarn plus one in Yangon, launched 2003. A widely repeated report that CP is seeking around US$150 million for a minority stake is from Bloomberg in April 2022 and is four years stale; the first draft of this page presented it as live, which was wrong. GIP, now inside BlackRock, is named alongside CP as a backer of the EEC project that won IJGlobal's Digital Infrastructure of the Year.
On the hyperscalers, the figures conflict and the page should say so. AWS is widely reported as pledging US$15 billion over five years, while Knight Frank's Atlas 2026 states AWS has committed US$5 billion and Google US$1 billion across Bangkok and Chonburi. Those are probably different bases, a multi-year cloud region commitment against data center capex, and neither should be quoted without saying which. ByteDance has announced close to US$4 billion across three Thai provinces. Microsoft has named Thailand for a regional data center. Siam AI Cloud, the domestic entrant, has committed US$101 million.
The generators are the other half of the trade, and only some have signed anything
Gulf Development is modeled largest, GPSC next, B.Grimm Power third. Those are analyst modeling assumptions, not allocations: Kiatnakin Phatra's utilities note of 25 August 2026 assumes 2,000 MW for GULF, 500 MW for GPSC and 300 MW for BGRIM contributing from 2029 to 2030, with valuation uplift of THB 4.59, 3.50 and 1.45 per share and target prices of THB 75, 60 and 25, upgrading GPSC to Buy. No Direct PPA capacity has been allocated to anyone.
What has actually been signed is a shorter list and it runs through one group. Amata B.Grimm Power, a joint venture of B.Grimm Power, Amata Corporation and Sumitomo Corporation Thailand, signed a 100 MW PPA with NTT Global Data Centers for Bangkok 4 at Amata City Chonburi, announced 26 January 2026, energization in the second quarter of 2027. B.Grimm already supplies NTT's Bangkok 2 and 3. Separately, Amata B.Grimm Renewable Energy signed a Private User PPA with DayOne for solar from a 22 kV floating solar facility inside the estate, starting at 42.5 MWp.
That Private User PPA category is the most important structural fact on this page. It is a route to contracted renewable supply that does not wait for the TPA Code, because the power never leaves the estate. The exact legal basis of the tripartite arrangement between DayOne, AMBRE and PEA is not disclosed in any source I could reach, which means it cannot yet be treated as a replicable template. Establishing what it actually is would be worth more than any further market sizing.
WHAUP has captive demand through the industrial estates and has publicly pressed Bangkok to move faster on Direct PPA. RATCH, Gunkul and EGCO are named beneficiaries. GPSC has said it is evaluating data center joint ventures in Thailand and India.
What the capital is paying, and the one valuation anchor that exists
Southeast Asian data center equity raised US$3.2 billion in 2024, US$1.9 billion in 2025 and US$4.7 billion year-to-date in 2026, with roughly 85% of all sector equity raised since 2024.
The platform transactions: KKR and Singtel signed definitive agreements on 4 February 2026 to acquire the remaining 82% of ST Telemedia Global Data Centres from ST Telemedia for S$6.6 billion, about US$5.1 billion, implying an enterprise value near US$13.8 billion. The first draft called this a completed US$5.2 billion acquisition of the company, which was wrong on price, on what was bought and on status. Bain bought Chindata for US$3.16 billion in 2019 and merged it with Bridge Data Centres into WinTriX, and is reported to be selling a stake in Bridge. Stonepeak joined Ontario Teachers' and Mubadala in Princeton Digital Group, where Warburg Pincus remains largest and the platform runs over 1.1 GW across six countries. GIP is co-developing greenfield with True IDC. DayOne closed a US$4.5 billion Series C and is weighing a dual Nasdaq and SGX listing at around US$20 billion.
There is one usable entry benchmark and it has to be derived rather than quoted. STT GDC's roughly US$10.9 billion enterprise value against 2.3 GW of design capacity is about US$4.7 million per MW of design capacity. Design capacity, not live IT, so it is a forward number on an unbuilt book.
On build cost, drop the US$7 to 8 million per MW figure the first draft carried; it has no source and Knight Frank does not publish a Bangkok number at all. The defensible anchor is Cushman & Wakefield's Asia Pacific Data Centre Construction Cost Guide 2026, published 31 March 2026, putting Thailand at US$8.8 million per MW, up 14.7% year on year, eighth of fourteen APAC markets between New Zealand at US$9.3 million and Indonesia at US$8.3 million. That is construction at shell-and-core plus fit-out, not full development cost, and it reflects 2025 build conditions.
Thailand against the region
Bangkok's 23.3% vacancy against Johor's 0.7% is the comparison that should govern any entry decision. Johor is full and Bangkok is not, which tells you where tenant demand actually landed.
Malaysia froze new non-AI data centers over power and water in February 2026. Singapore remains supply-constrained, roughly 1 GW operational at 1.4% vacancy, releasing capacity through controlled calls for application with efficiency and green energy conditions attached. Indonesia's Batam anchors DayOne's SIJORI hub-and-spoke model alongside Singapore and Johor.
The Philippines is the same trade a cycle behind and worse on power: about 500 MW installed across 28 colocation facilities in early 2026 against a pipeline above 700 MW, ePLDT leading with over 60 MW and more than 30% share, ST Telemedia's US$1 billion 124 MW Fairview 1 the record single site, land above US$3,000 per square meter in preferred zones, grid congestion already pushing builds to adjacent provinces, and no Direct PPA equivalent at Thailand's stage.
Thailand's advantages are real but narrower than the headlines suggest: hyperscaler commitment, an active clean-power reform program, and grid investment underway. Its disadvantages are that the reform is not in force, the grid is oversubscribed five to one, and a quarter of live colocation sits empty.
Where deals actually are
Reordered after the corrections, least contested first.
Generation sited inside an industrial estate, contracted directly to a campus. The Private User PPA between Amata B.Grimm Renewable Energy and DayOne proves the structure works without waiting for the TPA Code, because the electrons never cross the estate boundary. Amata and B.Grimm have done it once. Repeating it at other estates with other generators is the fastest available route to powered capacity in Thailand, and it is precisely what a sponsor with renewable development experience is built to do.
Captive and dedicated generation for campuses that cannot get grid. Twenty-one gigawatts of applications have no confirmed supply and connection alone runs 12 to 18 months above 20 MW. A developer who can bring generation to a site is selling the only scarce thing in the market.
Transmission-adjacent land, ahead of the EGAT direct-sale amendment. If EGAT is permitted to sell straight to large data centers near its 230 kV and 500 kV lines, position along those corridors reprices. The option is cheap now and expensive later, and the policy is already being drafted.
Joint ventures with estate and land holders. WHA, Amata and the other estate operators control serviced land with existing interconnection and their own utility subsidiaries. True IDC with GIP and Beijing Haoyang with WHA are both this shape. These are partnership conversations, not auctions.
The sixteen projects holding PEA confirmation letters. These are the only Thai data centers with proven grid access. Identifying them is the highest-value piece of origination work available, and it is a phone call rather than a research project.
Minority positions in Thai-owned operators. Note the new obstacle: PEA's June 2026 lock-up requires original shareholders to hold more than 50% for three years after transfer, so control acquisitions of power-confirmed projects are effectively closed. Minority structures with a locked-in local sponsor are now the only clean shape.
Second-tier provinces, and MEA territory. Everyone is fighting over Chonburi and Rayong. Bangkok and Samut Prakan reportedly have transmission headroom and sit outside the published PEA queue.
What has narrowed is the acquisition of stalled BOI approvals, which the first draft ranked highly. The lock-up, the power-before-promotion sequencing and the application pause together make that a much harder trade than it looked a day ago.
Corenexus: the entity is real, the platform story is unevidenced
Resolved on 30 August 2026 from the ACRA business profile. An earlier version of this page said the company could not be verified. That was wrong, and the correction matters more than the original claim.
CORENEXUS AI INFRA TECHNOLOGY (SINGAPORE) PTE. LTD., UEN 202623453K, is a live Singapore company incorporated on 26 May 2026, three months old at the time of writing. Primary activity is Data Centres (63111), secondary is data analytics, processing and hosting (63119). Registered office is 91 Bencoolen Street, #12-03, Sunshine Plaza.
Four things in the filing cut against the JD's description of the platform.
Issued share capital is SGD 60,000 and paid-up capital is zero. There is no money in the entity.
It is not a headquarters. The sole shareholder, holding all 60,000 ordinary shares, is CoreNexus AI Infra Technology (HongKong) Co., Limited (HK-80154089), registered at Room 5042, 5/F, Yau Lee Centre, 45 Hoi Yuen Road, Kwun Tong, Kowloon. The Singapore company is a wholly-owned subsidiary of a Hong Kong holding company.
The Singapore-resident director, Lee Tok Cheng, is expressly flagged as a nominee director, and the company secretary, Mah Yee Leng, is registered at the same address as the registered office, which marks it as a corporate services provider rather than an operating office. Both arrangements are legal and unremarkable for a foreign-owned Singapore company, but together they mean there is no Singaporean principal and no Singapore office in any substantive sense.
The two substantive directors, Liu Wei and Zeng Shisheng, are both resident in Shenzhen, in Futian and Nanshan districts. That is worth weighing against the candidate sponsors considered above: Beijing Haoyang is Beijing, GDS is Shanghai, DayOne is Singapore. Shenzhen points elsewhere.
No audit firm is appointed, no charges are registered, and no AGM or annual return has been filed, all of which is consistent with a three-month-old company and is not itself a concern.
The decisive point is what the filing does not contain. No sponsor appears anywhere. There is no renewable energy developer, no data center developer, no named industrial backer. The only shareholder is the Hong Kong parent, and ACRA cannot see above it. So the JD's claim of sponsorship by leading renewable energy and data center developers is neither confirmed nor refuted by the register. It is simply absent from it.
Set the two accounts side by side. The JD describes a Singapore-registered AI infrastructure ecosystem investment platform, sponsored by leading renewable and data center developers, with a Partner at Singapore HQ and a core team that delivered over 1 GW of data centers and 5 GW of renewables in China. The register shows a three-month-old, zero-paid-up subsidiary of a Hong Kong company with a nominee director and an outsourced secretary. These are not incompatible, since groups routinely incorporate a clean vehicle and capitalize it at first close. But nothing in the public record supports the platform description, and it should not be treated as established.
The Hong Kong parent, traced 30 August 2026. CORENEXUS AI INFRA TECHNOLOGY (HONGKONG) CO., LIMITED trades in Chinese as 科聯星雲(香港)有限公司 and was incorporated on 9 April 2026, seven weeks before the Singapore subsidiary. The entire structure is therefore about five months old. The Chinese name has no public footprint in mainland or Hong Kong sources, which fits a newly formed offshore vehicle rather than the trading name of an established group.
The Singapore registered office at 91 Bencoolen Street #12-03 carries 1,259 other registered entities, 1,223 of them live. It is a registered agent's address, which confirms there is no Singapore office behind the "Singapore HQ" language.
Its shareholders could not be traced from public sources. Hong Kong's free registry layer returns name, CR number, incorporation date and status only. Shareholders sit in purchasable filings. Because the company was incorporated in April 2026, its first annual return (NAR1) is not due until roughly May 2027 under the 42-days-after-first-anniversary rule, so none exists yet. That leaves the NNC1 incorporation form, which names the founder members and initial directors, as the only document that can identify the owners. It must be bought from ICRIS, which was unavailable on 30 August 2026 owing to a scheduled maintenance outage from 9am to 9pm Hong Kong time.
One caveat on that document. NNC1 shows ownership at incorporation. Any share transfer since April 2026 would not appear until the first NAR1 is filed in 2027, so a sponsor that came in after formation would be invisible.
A same-name entity that is not linked. CORENEXUS MINERAL HOLDINGS PTE. LTD. (UEN 202540859W) is a live Singapore company incorporated 12 September 2025, at 112 Robinson Road, in wholesale of metals and metal ores with other mining and quarrying as secondary activity. It shares the name stem but sits at a different address with a different registered agent, in a different sector, and predates the AI infra entity by eight months. There is no evidence of common ownership and it should not be treated as related without shareholder data.
Questions to put to them, updated. Which entity issues the employment contract, the Singapore company or the Hong Kong parent. Whether a Thai entity exists or is planned, and how the work permit is handled. Who the shareholders of the Hong Kong company are. What the 1 GW and 5 GW refer to, projects delivered or capacity under management, and at which companies. What capital is committed to the Thailand strategy as opposed to targeted, and by whom. And what the project incentive and carry participation are written against, since carry in a vehicle with no capital and no assets is worth exactly what the sponsors behind it are worth.
Confidence and known gaps
This page was rebuilt after adversarial verification refuted sixteen claims in the first draft. What follows is what the verification could not close.
No published wheeling rate has been gazetted, only consulted. The Type 9 tariff has no primary-source rate, structure or effective date. PDP 2026's draft is not publicly obtainable and EPPO's page blocks retrieval; its data center demand scenarios run 6,799 MW low, 8,811 MW medium and 19,808 MW high to 2050, with roughly 8,000 MW of firm state supply envisaged and Direct PPA as the parallel route. The three TPA sub-codes are not public. No capacity booking portal could be found at any URL, which is absence of evidence rather than evidence of absence. There is no MEA equivalent of the PEA queue anywhere in the public record. Whether the IEAT ever issued its promised permit exemption for self-consumption solar inside estates is unresolved, and every source still says "expected mid-2025". The legal basis of the DayOne, AMBRE and PEA tripartite structure is undisclosed. Thai rules on commercial operation of backup generation and on standalone battery storage could not be established at all. The NBTC's Type 3 reclassification of data center licences is an announced intention, not a live consultation.
Treat every figure above as sourced to the date attached to it. Thai energy policy moved three times in August 2026 alone.
Sources
Primary: EGAT, 16 Feb 2026 · Bangkok Post on the B31bn EEC upgrade · Nation Thailand on the 2.87 GW estimate · Nation Thailand on tightened rules · STT GDC on the KKR consortium
Regulation: Hunton on the draft Direct PPA regulation · Formichella & Sritawat on the draft TPA Code · Dentons, what changed since early 2026 · Nishimura & Asahi on UGT1 · CASE for Southeast Asia · Maverick Consulting on the 2026 approval bar · Mahanakorn Partners on investment screening · Chandler Mori Hamada on the National Data Center Business Policy Committee
Market and grid: Eco-Business on tightened rules · Bangkok Post on the power logjam · Nation Thailand on grid, rules and workforce · DCD on the BOI round · DC Byte on Bangkok
Operators and capital: DCD on the DayOne Chonburi groundbreaking · DayOne on the Amata partnership · DCD on the NTT and B.Grimm PPA · Forbes on Beijing Haoyang at WHA · WHA Group on the THB 72.67bn campus · Empyrion TH1 groundbreaking · ARC Group on SEA data centre M&A · TNGlobal on SEA equity via Tracxn · DCD on CP Group and True IDC, April 2022 · ISI Markets on DayOne's Series C
Philippines: White & Case, Digital crossroads · Arizton Philippines portfolio · Filipino Engineer on power constraints
Topics
th-power-sector · datacenter-foundations · ai-infrastructure · country-th · green-finance · tf-th-report · tf-th-universe