AI Data Centres Need Power: Lessons for SG Firms

US$1 billion doesn’t get spent on “nice-to-have” infrastructure. It gets spent on constraints.

That’s the real story behind today’s news: OpenAI and SoftBank are investing US$500 million each (US$1 billion total) into SB Energy to expand data centre and power infrastructure for the Stargate initiative. SB Energy will build and operate a previously announced 1.2-gigawatt data centre site in Milam County, Texas, and it will also become an OpenAI customer, using OpenAI’s APIs and deploying ChatGPT for employees. Source: https://www.channelnewsasia.com/business/openai-softbank-invest-1-billion-in-sb-energy-stargate-buildout-expands-5849541

For the AI Business Tools Singapore series, this matters because it exposes a misconception many companies still have: AI scale is mostly a software problem. It isn’t. The bottleneck is increasingly electricity, cooling, and the ability to run reliable compute—and leaders are investing accordingly.

If you’re running a Singapore SME or a regional business team, you won’t be building gigawatt-scale campuses in Texas. But you will face the downstream effects: higher AI usage, rising compute costs, tougher governance, and customers expecting faster service. This post breaks down what the OpenAI–SoftBank–SB Energy move signals, and how to translate it into practical AI adoption decisions in Singapore.

The signal: AI’s biggest constraint is now energy

The investment is a straightforward statement: power access has become a limiting factor for AI expansion. That’s why tech companies are moving beyond buying servers and signing cloud contracts—into power infrastructure.

Stargate is described as a US$500 billion multi-year initiative to build AI data centres for training and inference, with major backers including Oracle. The CNA report also notes a broader boom where players such as Meta are committing unprecedented sums to chips, power, cooling, and servers.

Why 1.2 gigawatts is a big deal

A 1.2GW site isn’t “large.” It’s city-scale electricity.

For business leaders, you don’t need to memorise the engineering details. You need the implication:

• AI capability is increasingly tied to infrastructure scale.
• Energy price volatility can become AI price volatility.
• The winners will design AI workflows that are cost-aware, not just model-aware.

In Singapore, where land and energy are constrained, this is exactly why AI adoption has to be thoughtful. Throwing every process into an LLM is how costs creep up quietly.

A practical translation for Singapore businesses

Here’s the stance I’ll take: most companies will overspend on AI in 2026 by using it too broadly and measuring it too loosely.

When global leaders are spending billions to secure power, it’s a hint that AI isn’t getting cheaper just because models improve. Better models often increase usage—and usage drives cost.

So the right question becomes: Where does AI create measurable value per token, per minute, per workflow?

What this partnership says about “AI + sustainability” in real life

A lot of corporate AI-sustainability talk is fluffy. This partnership is not. It’s the operational version: if you want more AI, you need more clean, reliable energy and better infrastructure planning.

The CNA report notes: “Tech companies are investing directly in power infrastructure as energy access becomes a critical constraint on AI expansion.” That’s the cleanest summary you’ll read this week.

Why AI and energy are now one strategy

AI changes sustainability in two opposing ways:

1. AI increases demand (training, inference, always-on copilots, automation).
2. AI can reduce waste (optimising logistics, predicting demand, improving building efficiency, automating compliance reporting).

Companies that do only (1) get higher costs and tougher ESG questions. Companies that pair (1) with (2) can credibly claim they’re using AI to improve outcomes.

For Singapore firms—especially in real estate, logistics, retail, professional services—the opportunity is to adopt AI business tools that cut waste and lift productivity at the same time.

Example: turning AI into measurable energy savings

You don’t need a data centre to benefit from this.

• Facilities teams can use AI-assisted anomaly detection on BMS data to flag inefficient HVAC schedules.
• Finance teams can use AI to classify and reconcile invoices faster, reducing manual work and shortening close cycles.
• Customer service teams can deploy ChatGPT-style assistants that deflect repetitive tickets, reducing after-hours overtime.

The pattern is consistent: pick workflows that are frequent, measurable, and costly when done manually.

The “buy vs build” reality: most Singapore SMEs should not build models

OpenAI is funding infrastructure and SB Energy is becoming an OpenAI customer. That’s a quiet reminder that even sophisticated companies often choose “use” over “build” in many areas.

For most Singapore SMEs, the winning approach in 2026 will be:

• Buy: proven AI business tools for common functions (customer support, marketing ops, sales enablement, internal knowledge search).
• Configure: connect to your CRM/ERP/helpdesk, define roles, add guardrails.
• Measure: track cost, time saved, quality, and risk.
• Only then consider custom builds: where data moats or compliance needs justify it.

A simple decision rule

Use this rule of thumb:

• If a workflow is common across industries (drafting emails, summarising calls, creating first-pass content), start with off-the-shelf tools.
• If a workflow is unique to your operations (your product specs, your compliance logic, your proprietary datasets), consider a custom solution—but still prefer API-based approaches rather than training models from scratch.

This keeps AI adoption fast without turning your company into a research lab.

A Singapore-ready playbook: adopt AI tools without runaway cost

The infrastructure arms race is happening globally. Your job is to adopt AI locally in a way that’s economical, secure, and repeatable.

Here’s what works in practice.

1) Start with “high-frequency, low-risk” workflows

These are the easiest to standardise and measure:

• First-draft responses for customer enquiries (with human approval)
• Meeting notes and action item extraction
• Internal knowledge base Q&A for policies, SOPs, product info
• Marketing ops: repurposing long content into short formats

Aim for a first rollout you can complete in 2–4 weeks, not a 6-month “transformation.”

2) Put cost controls in from day one

If energy and compute are strategic constraints at the top, they’ll show up as line items for you too.

Do these early:

• Set usage budgets per team or per tool
• Prefer retrieval (search + summarise) over feeding entire documents into prompts
• Use smaller models for routine tasks; reserve advanced models for high-stakes work
• Standardise prompt templates for repeatable tasks (reduces retries and waste)

A good AI rollout doesn’t just increase output. It reduces rework.

3) Treat governance as part of product design

Singapore companies often wait too long to define rules. That’s backwards.

Define upfront:

• What data can be used (PII? contracts? HR info?)
• Where outputs can be stored
• Who approves customer-facing text
• What must be logged for audit

This matters more in regulated industries (finance, healthcare), but even B2B services firms will run into client requirements quickly.

4) Measure outcomes, not excitement

If you want AI adoption to survive budget season, you need numbers.

Track metrics such as:

• Ticket deflection rate and customer satisfaction
• Average handling time (AHT) and first response time
• Time-to-proposal and win rate for sales teams
• Month-end close duration and reconciliation accuracy

The goal is a business case your CFO can repeat in one sentence: “We spend X and we get Y back.”

What Singapore businesses should watch next (2026)

This deal is one datapoint in a bigger shift: AI is becoming infrastructure-first.

Here are the trends I’d keep on your 2026 radar if you’re choosing AI business tools in Singapore:

AI pricing will get more complex

Expect more tiering around:

• Speed/latency
• Model class (small vs frontier)
• Data residency and compliance controls
• Enterprise governance features

Plan procurement like you plan cloud spend: with guardrails and forecasts, not ad hoc swipes of a credit card.

“AI for employees” will become normal

SB Energy deploying ChatGPT for employees is a sign of where this is heading: internal copilots as standard tooling, like email and chat.

For Singapore companies, the operational win is usually internal first:

• Faster onboarding
• Cleaner SOP adherence
• Less tribal knowledge

Then you expand externally to customers when the workflow is stable.

Energy and sustainability questions will reach your AI roadmap

Even if you’re not asked to report AI energy usage directly, clients may ask:

• Where does the data live?
• What’s the vendor’s security posture?
• What’s your policy on sensitive data?
• How do you manage AI risk?

Sustainability is increasingly bundled with “responsible AI” in procurement conversations.

People also ask: what does this mean for SMEs using AI tools?

Will AI tools get more expensive?

They can, especially as usage scales. The better question is: will your cost per outcome go down? If AI saves 3 hours per week per employee, modest price increases don’t matter.

Should SMEs worry about data centres and power?

Not directly. But you should assume compute is not infinite and design workflows to be efficient: smaller models when possible, retrieval-based approaches, fewer retries.

Is it smarter to build an in-house model now?

For most SMEs: no. Use APIs and proven AI business tools first. Build only where you have unique data, clear ROI, and internal capability to maintain it.

Where this leaves the “AI Business Tools Singapore” roadmap

OpenAI and SoftBank investing US$1 billion into SB Energy is a reminder that AI progress isn’t only about smarter models—it’s about making AI usable at scale. Infrastructure, energy, and operations are now part of the AI story.

For Singapore businesses, the advantage is you don’t have to play the infrastructure game to benefit. You can focus on AI business tools that improve marketing, operations, and customer engagement—then run them with discipline: cost controls, governance, and measurable outcomes.

If you’re mapping your 2026 plan, start small but be serious. Pick one workflow you can improve this quarter, implement it with clear guardrails, and measure the result. Then expand.

What’s one process in your team that’s high-volume, repetitive, and measurable—something you’d be happy to never do manually again?