AI Data Centres Are Booming—What SG Firms Should Do

A 240-megawatt AI data centre is not a vanity project. It’s a signal flare.

This week, Reuters reported that Nebius, an Amsterdam-based AI cloud services firm, plans to build a 240MW data centre near Lille, France—one of Europe’s largest when complete, with initial capacity expected to come online in phases starting late summer 2026 and around half the site operational by end-2026. The site is a redevelopment of a former Bridgestone tyre plant in Béthune. (Source story: https://www.channelnewsasia.com/business/amsterdam-based-ai-firm-nebius-build-240mw-data-centre-near-lille-france-5926481)

If you’re running a business in Singapore, the headline isn’t “France gets more compute.” The headline is: AI demand is now big enough that entire power-plant-sized facilities are being built just to feed it. And when supply is built at that scale, the companies that benefit aren’t only AI labs—they’re the everyday firms that learn how to use AI profitably.

This post is part of our AI Business Tools Singapore series, and I’m going to make a direct argument: most SMEs don’t have an “AI problem.” They have a “foundation problem.” The Nebius news is a useful mirror for what “foundation” really means—compute, cost, data governance, and operational discipline.

What the Nebius 240MW build really tells us

The core message is simple: AI is moving from experiments to production workloads, and production workloads require predictable infrastructure.

Nebius says the facility will be delivered in phases, with the first capacity expected online by late summer. That phased approach is what mature AI adoption looks like: you don’t bet the company on a single launch; you build incrementally, validate demand, then expand.

Here’s the detail that should stop any business owner mid-scroll: real estate firm CBRE estimates AI data centre development costs at US$10–14 million per megawatt. For 240MW, that implies roughly US$2.4–3.6 billion in build and construction cost.

That number matters even if you’ll never build a data centre. It tells you:

• AI capacity is expensive, and it’s being funded because buyers are real.
• Compute pricing and availability will increasingly influence your AI roadmap.
• Efficiency isn’t a “nice to have.” Wasteful AI usage becomes a P&L problem.

Why Europe is building fast—and why Singapore should pay attention

Reuters notes a wave of European companies—manufacturing, services, logistics—deploying AI tools. That’s not a “tech sector” story. It’s an “everyone” story.

Singapore businesses are on a similar curve, but with a twist: we’re a high-cost, high-productivity market. If AI doesn’t reduce cycle time, improve conversion, or cut operational friction quickly, it won’t survive budgeting season.

So when you see Europe pouring billions into AI infrastructure, treat it as a forecast for what will become normal:

“AI adoption will be limited less by ideas, and more by disciplined execution—data, cost control, and governance.”

The hidden link between data centres and your AI business tools

AI tools feel like software. Under the hood, they’re powered by GPUs, electricity, cooling, networking, and scheduling. When Nebius says it will purchase Nvidia chips shortly before use, that’s another signal: the supply chain and timing of compute resources matter.

For Singapore companies adopting AI for marketing and operations, this shows up in three very practical ways.

1) Your AI costs won’t behave like normal SaaS

Traditional SaaS is usually per seat. AI is often per usage—tokens, images, minutes, or model calls—and usage can spike unexpectedly.

What works:

• Put budgets and rate limits in place from day one.
• Design workflows where AI is used for high-leverage steps, not every step.
• Prefer “small model first” or “retrieve-then-generate” patterns for many tasks.

A simple rule I use with teams: If the AI output isn’t changing a decision, don’t pay to generate it.

2) Energy and efficiency become business concerns (even if indirectly)

A 240MW facility is massive, and it exists because AI workloads are power hungry. You may not pay the electricity bill, but you will pay for:

• premium pricing during peak demand,
• higher costs for larger models,
• compliance and reporting overhead (especially for regulated industries).

In Singapore, where sustainability reporting is increasingly part of enterprise procurement, efficient AI operations can be a sales advantage.

3) “Where your AI runs” is becoming a procurement question

Many Singapore firms are fine with public cloud and global vendors. Some can’t be—because of data residency, client contracts, or sector rules.

The direction of travel is clear: more regional capacity, more options, and more scrutiny.

If you sell to banks, healthcare, government-linked entities, or large enterprises, you should expect questions like:

• Where is data processed and stored?
• What’s the retention policy for prompts and outputs?
• Can you isolate tenants? (or choose dedicated capacity)

Those are infrastructure questions expressed in business language.

What Singapore businesses should do now (practical playbook)

The point isn’t to chase the biggest model. The point is to build an AI foundation that survives contact with real operations.

Step 1: Choose 2–3 workflows where AI can prove ROI in 30 days

If your first AI project takes 6 months, it’s going to die.

Good candidates in Singapore SMEs and mid-market teams:

• Sales: lead qualification summaries, proposal drafting, objection handling playbooks
• Marketing: ad copy variants, landing page rewrites, content briefs, SEO outlines
• Customer support: ticket triage, reply suggestions, knowledge base drafting
• Operations: SOP drafting, vendor comparison tables, internal policy Q&A

Define success with a number, not a feeling:

• Reduce response time from 6 hours to 2 hours
• Increase qualified leads per week by 20%
• Cut manual reporting time by 50%

Step 2: Put guardrails in before you scale

Most companies do this backwards. They scale, then panic.

Minimum guardrails that actually work:

1. Data rules: what can’t be pasted into an AI tool (NRIC, client pricing, medical data, etc.)
2. Review rules: which outputs require human approval (contracts, claims, regulated comms)
3. Logging: capture prompts/outputs for improvement and audits (where appropriate)
4. Cost controls: usage limits by team, and alerts for spikes

Step 3: Build “AI-ready content” and “AI-ready data”

AI tools perform like your inputs perform. If your knowledge base is outdated, your chatbot will confidently serve nonsense.

A practical approach:

• Centralise FAQs, product specs, pricing rules, policies
• Keep one source of truth (even if it’s just a well-owned Notion/Confluence space)
• Use retrieval (RAG-style) workflows so the model cites your documents, not general internet memory

This is where marketing and operations meet: the same structured knowledge that improves support also improves sales enablement and onboarding.

Step 4: Decide what you keep in-house vs. outsource to AI vendors

Nebius is betting on owning and operating AI infrastructure capacity. Most Singapore firms shouldn’t.

But you should decide intentionally:

• Use external tools for commodity tasks (drafting, summarisation, translation)
• Consider more controlled setups for sensitive workflows (client data, contracts, regulated info)
• Negotiate enterprise terms when AI becomes business-critical (SLAs, retention, security)

A useful litmus test: If downtime costs you revenue today, it’s not a “tool” anymore. It’s infrastructure.

“Neocloud” expansion is a sign: AI is becoming normal business infrastructure

Reuters describes Nebius (and CoreWeave in the US) as leading “neocloud” firms—specialists that supply AI infrastructure, sometimes through big deals with hyperscalers. Nebius has reportedly struck high-profile supply deals with firms like Microsoft and Meta, while also serving companies such as Mistral, Shopify, and ServiceNow.

You don’t need to memorise the players. You need to understand the pattern:

• Hyperscalers can’t do everything alone.
• Specialised infrastructure companies are scaling fast.
• Capacity is being built where demand is expected to persist.

That’s why the Lille facility matters. It’s not a one-off. It’s part of a global build-out that will shape pricing, availability, and performance for the AI business tools you use in Singapore.

A Singapore-first view: how to turn AI capacity into leads and efficiency

If your goal is leads, AI should reduce time-to-campaign and improve follow-up quality—not just generate generic posts.

Here’s what I’ve seen work reliably:

For marketing teams

• Generate 10–20 ad angles quickly, then let humans pick 2–3 worth testing.
• Use AI to produce first drafts, but keep brand voice editing in-house.
• Build a “message library” of proven claims, offers, and objections so AI doesn’t invent positioning.

For sales teams

• Turn call notes into next-step emails and CRM updates.
• Create industry-specific proposal blocks (logistics, F&B, professional services) to shorten turnaround.
• Use AI to identify missing info in a deal: budget, authority, timing, success criteria.

For operations teams

• Convert tribal knowledge into SOPs and checklists.
• Use AI for document comparison: vendor quotes, policy versions, contract clauses (with review).
• Build internal Q&A tools that reduce repeat questions and onboarding time.

The common thread: AI doesn’t replace judgement. It removes blank-page time and repetitive work.

Where this goes next

Nebius’s 240MW announcement is the loud part of a quieter shift: AI is now treated like electricity—something you plan for, budget for, and optimise.

If you’re a Singapore business adopting AI business tools, the winning move in 2026 is straightforward: stop treating AI as a collection of prompts and start treating it as an operating capability—with costs, controls, and measurable outcomes.

If you had to pick one workflow to make faster, cheaper, or more consistent using AI in the next 30 days, what would it be—and what number would prove it worked?