JAKARTA &ndash; Global investment for the development of data centres is projected to reach USD 580 billion, or around IDR 9.69 quadrillion, by the end of 2025, according to the latest report from the International Energy Agency (IEA).
The IEA states that this surge in investment is driven by the rapid advancement of AI technology, which has also increased demand for both fossil-based and renewable energy.
Citing TechCrunch on Monday (17/11), the investment figure even exceeds global oil exploration spending, which stands at USD 40 billion, or around IDR 668 trillion.
Meanwhile, major technology companies are entering a phase of data centre construction on an unprecedented scale.
<ul>
<li>OpenAI has committed USD 1.4 trillion (around IDR 23.38 quadrillion) to build new data centre infrastructure.</li>
<li>Meta is preparing USD 600 billion (around IDR 10.02 quadrillion).</li>
<li>Anthropic is planning an investment of USD 50 billion (around IDR 835 trillion).</li>
</ul>
However, this massive growth is inseparable from major challenges related to energy availability.
Quoting FindArticle on Monday (17/11), electricity demand for AI data centres has surged rapidly, while renewable energy capacity in many countries remains behind.
Companies such as Google, Microsoft, and Meta claim to have balanced almost all of their annual electricity consumption by purchasing renewable energy through long-term contracts or green energy certificates.
However, this does not mean their data centres operate with emission-free electricity at all times. Company data shows real-time clean energy supply currently reaches only 30&ndash;70% of total consumption, depending on region and season.
Outside hyperscale data centre operators, the situation is more varied. Many co-location data centres and other companies do not yet have long-term renewable energy strategies, especially in new regions emerging as AI growth hubs.
With the global energy mix still relying on roughly one-third low-carbon sources, the share of renewable energy actually used by AI data centres is estimated to be far smaller than the official claims of technology companies.
Meanwhile, corporate demand for clean energy continues to hit record highs. Power purchase agreements are increasingly shifting toward 24/7 carbon-free models that combine wind, solar, energy storage, geothermal, and grid services.
Yet energy infrastructure remains a key stumbling block. Lengthy interconnection queues and limited transmission networks mean that connecting new energy projects can take years.
Data centre operators often choose locations with large electricity supply that still relies on fossil energy, unless they invest in clean power sources on-site from the start.
IEA projections show that hyperscale data centres in mature markets may be able to maintain near-100% annual energy matching by 2030. In regions with strong wind, solar, or hydro potential, real-time clean energy supply could rise to 50&ndash;80%.
Globally, including co-location operators and new AI players, annual matching levels are estimated to be in the range of 50&ndash;70%. Without accelerated development of transmission and energy storage, real-time performance could be far lower.
Efforts to secure clean energy sources continue to develop. On-site solar power installed at data centre campuses can only cover a fraction of demand, so operators rely on large-scale wind and solar projects with long-duration storage. (DK/KR/ZH)

Global data centre investment soars due to rise of AI, yet clean energy supply continues to lag.

Surge in AI data centre investment reaches USD 580 billion

JAKARTA &ndash; Meta dalam beberapa bulan terakhir melakukan serangkaian perubahan besar pada divisi kecerdasan buatannya (AI).
Setelah membagi unit AI, membentuk laboratorium superintelligence, hingga merekrut peneliti dengan tawaran fantastis, kini perusahaan menunjuk Alexandr Wang, pendiri Scale AI berusia 28 tahun, sebagai pemimpin utama seluruh operasi.
Dikutip puck.news, Meta sebelumnya sempat membagi divisi AI menjadi dua:
<ul>
<li>
Tim produk yang menangani integrasi AI pada Facebook, Instagram, dan WhatsApp
</li>
<li>
AGI Foundations yang berfokus pada pengembangan model bahasa besar Llama
</li>
</ul>
Namun, menurut laporan terbaru, struktur itu kembali dirombak Wang menjadi empat kelompok:
<ol>
<li>
TBD Lab untuk superintelligence
</li>
<li>
Unit produk
</li>
<li>
Tim pusat data dan infrastruktur
</li>
<li>
FAIR (Fundamental AI Research) yang dipimpin oleh Yann LeCun
</li>
</ol>
Dalam memo internal yang diperoleh Business Insider, Wang mengumumkan pembubaran AGI Foundations dan menegaskan bahwa riset FAIR akan menopang TBD Lab.
Ia juga memastikan bahwa setiap kepala divisi, termasuk Yann LeCun, kini melapor langsung kepadanya, kecuali kepala ilmuwan baru Meta, Shengjia Zhao, salah satu pencipta ChatGPT.
Meski belum ada PHK, The New York Times melaporkan bahwa opsi pengurangan staf tengah dipertimbangkan. Meta menolak mengomentari isu tersebut, tetapi juru bicara Meta Andy Stone menegaskan melalui X bahwa perusahaan hanya melakukan perencanaan organisasi dasar, menciptakan struktur yang solid untuk upaya superintelligence, serta menjalani latihan penganggaran dan perencanaan tahunan.
Wang pun membantah kabar negatif seputar restrukturisasi. Dalam unggahan di X, ia menulis: &ldquo;Meta benar-benar semakin berinvestasi di Meta Superintelligence Labs sebagai sebuah perusahaan. Laporan apa pun yang menyatakan sebaliknya jelas keliru.&rdquo;
Namun langkah Meta dipandang sebagian analis sebagai upaya tergesa-gesa Zuckerberg untuk mengejar persaingan. Roadmap yang disampaikan Wang menekankan penskalaan model besar untuk mencapai kecerdasan super di seluruh pelatihan pra, penalaran, dan pasca, pendekatan yang sama seperti mayoritas perusahaan lain. Ia juga menyebut rencana eksplorasi model &ldquo;omni&rdquo; atau multimodal yang mampu menghasilkan teks, gambar, dan video.
Pendekatan ini menuai kritik dari Yann LeCun, yang berulang kali menolak istilah AGI dan menyebut LLM sekadar &ldquo;token generators.&rdquo;
Maksudnya sebagai genarator tokken adalah&nbsp;LLM hanya memprediksi kata berikutnya (token demi token) berdasarkan pola statistik dalam data teks. Mesin itu tidak benar-benar memahami dunia nyata atau punya representasi konseptual yang stabil tentang realitas. LeCun menekankan bahwa kecerdasan umum (AGI) mestinya jauh lebih dari sekadar prediksi teks:
Menurutnya, model generasi baru seharusnya memahami dunia fisik, memiliki ingatan yang kuat, dan lebih mampu merencanakan serta bernalar. Ironisnya, Yann LeCun kini justru masuk lebih jauh dalam proyek superintelligence Meta yang bertumpu pada LLM berskala besar.
Meski begitu, ambisi Zuckerberg pada AI tetap berkelindan dengan kepentingan bisnis. Meta masih mengandalkan pendapatan iklan dari platform sosial dan perpesanan, sementara investasi pada realitas virtual dan kacamata pintar diproyeksikan menembus US$100 miliar tahun ini.
Di tengah perlombaan AI, membangun citra sebagai pemimpin riset tampaknya sama pentingnya dengan menghadirkan produk komersial nyata. (DH)

Roadmap yang disampaikan Alexandr Wang tidak beda jauh dengan LLM besar yang sudah berjalan.