We need to understand how data centers really work
Editor’s Note: I had the chance to talk to Max Parry, Vice President of Growth and Emerging Markets in Asia-Pacific for Equinix. This was my second interaction with him (here’s the first) and I wanted to get a sense — at least from the side of the data centers on what sustainability means especially for the Philippines where we now have the highest electricity cost in the region. When I spoke to him last, we were only second to Singapore. A lot has changed in a few months. I wanted to talk to Equinix because they were around before the whole issue of Pax Silica and data centers were being thrown around in a rather uninformed manner.
The data center underneath everything
The Philippines has a data problem. Not a shortage of it — quite the opposite. As AI adoption accelerates across banking, healthcare, BPO, and government, the volume of data being generated, processed, and stored is outpacing the infrastructure available to handle it responsibly. Into that gap steps Equinix, a company that has been running the behind-the-scenes plumbing of the internet for nearly three decades — and which now operates three facilities in the Philippines through its MN1, MN2, and MN3 sites in Manila and Makati.
“The gigawatt projects making headlines are designed to process AI for one company,” says Max Parry, Vice President for Growth and Emerging Markets in Asia-Pacific at Equinix. “Our core business is different. We are connecting thousands of companies to each other and to the clouds, networks, and partners they need.”
That distinction matters more than it might initially seem. The data centers drawing scrutiny — and in some cases, community resistance — are the hyperscale campuses being built to train massive AI models. Equinix’s Philippine facilities are built for a different layer of the stack: AI inference, the part where trained models actually respond to users in real time. That workload demands proximity to people, not necessarily raw compute at scale. It also, critically, demands lower power density than AI training — which changes the sustainability equation considerably.
Still, the energy question in the Philippines is unavoidable. The local grid runs close to capacity, power cost is now, as of this writing, the highest in Southeast Asia, and any large-scale data center operation is going to draw scrutiny. Parry’s answer involves a mix of efficiency metrics and renewable energy commitments that are genuinely substantial. Equinix claims 96% renewable energy coverage across its global portfolio — its eighth consecutive year above 90% — and states that its Philippine facilities are 100% covered through International Renewable Energy Certificates or I-REC.
The I-REC mechanism has its critics. There is a legitimate argument that purchasing certificates doesn’t necessarily mean clean electrons are flowing into servers at the moment they’re needed, and that geothermal energy — the Philippines’ comparative advantage in renewables — ought to be reserved for residential and industrial users rather than allocated to data centers. I had brought this up by saying that geothermal renewables are supposedly the cheapest form — and therefore should be reserved for the masses, not data centers. Parry acknowledges the tension. “I-RECs are one component of a broader strategy,” he says, pointing to power purchase agreements, on-site solar, and direct investment in renewable energy market development as supplementary levers. The company signed over 458 megawatts of renewable PPAs across Asia-Pacific in 2024 and 2025.
PUE numbers
On the efficiency side, the numbers are harder to argue with. Equinix’s global average Power Usage Effectiveness — the ratio of total power drawn to power actually used for computing — sits at 1.37, a 31% improvement over its 2019 baseline. To compare, other data center models in the Philippines such as Narra1 by Digital Edge have a PUE of 1.17, which is closer to the efficiency of bigger hyperscalers like Meta. (Note: a PUE rating closer to 1.0 means better power effectiveness) That being said, this is the Philippines, and to have these numbers are quite convincing. The company has piloted machine learning controls in selected facilities that adjust cooling systems in real time based on sensor data, achieving a PUE of around 1.2 at its Frankfurt site and 10% annual PUE improvements in Singapore. Whether those pilots eventually reach the Philippine facilities is a question Parry leaves open, noting that cooling strategy depends on specific workload requirements and that not every AI application demands liquid cooling.
For local enterprises planning their AI infrastructure over the next two years, Parry’s most pointed advice concerns architecture rather than hardware. The most common and costly mistake, he argues, is treating multicloud as multiple isolated cloud environments managed separately — rather than an integrated platform where workloads move dynamically based on cost, latency, and performance. “Without the right network infrastructure, this kind of flexibility isn’t possible,” he says. The implication is that the physical location of interconnection — where clouds, networks, and enterprise systems actually meet — becomes a strategic decision, not just a procurement one.
In a country still building toward that kind of resilience — the push and pull of ESG and what we want to call progress, it’s a pitch that arrives at an opportune moment — and one the market is only beginning to fully understand.
