AI Infrastructure Boom: What Singapore Firms Should Do

Applied Digital just posted US$126.6 million in quarterly revenue—up 139% year-on-year—beating analyst estimates of US$76.6 million. That’s not a “nice-to-have” signal. It’s proof that AI infrastructure is becoming one of the most bankable business categories on the planet, and the companies that can supply compute, power, and data-centre capacity are getting paid first.

If you’re running a business in Singapore, you might think this is only relevant to hyperscalers and data-centre operators in the US. I don’t buy that. The underlying driver—a surge in demand for generative AI workloads—shows up everywhere, including in Singaporean marketing teams trying to produce more content, ops teams trying to forecast demand, and customer service teams trying to handle more chats without expanding headcount.

This article is part of the AI Business Tools Singapore series, so we’ll treat Applied Digital’s results as a practical case study: what the infrastructure boom means, why it’s happening now, and how Singapore businesses can make smarter choices about AI tools, budgets, and timelines.

What Applied Digital’s earnings really tell us

Answer first: Applied Digital’s revenue beat is a proxy for a bigger truth: AI success is increasingly constrained by infrastructure, not ideas.

According to the report, big tech and AI companies are racing to lock in power and data-centre capacity with long-term, multi-billion-dollar deals. Applied Digital’s CEO highlighted that they’re now seeing the “earnings power” of the platform, with a full quarter of revenue from their first building recognized. Their first large-scale AI data center reached full operations at 100 megawatts in November, and they’ve started building a 300-megawatt campus (Delta Forge 1) expected to begin initial operations in mid-2027.

Here’s the key business takeaway: AI demand isn’t spiky anymore. It’s contractual. When customers sign long-term capacity deals, the demand has moved from “experiments” to “operating model.”

And when infrastructure providers outperform expectations this dramatically, it usually means downstream users—enterprises—are already spending.

The infrastructure stack is becoming the AI stack

For most companies, “AI” still sounds like software. In practice, it’s a stack:

• Data (your CRM, ERP, knowledge base, call logs)
• Models (LLMs, recommendation models, forecasting)
• Tools (chatbots, copilots, analytics)
• Infrastructure (compute, storage, networking, power, cooling)
• Governance (security, audit trails, PDPA compliance)

Applied Digital sits near the bottom of that stack—where costs are real and usage is measurable. That’s why their performance is such a clean signal.

Why AI infrastructure demand is soaring (and why it won’t cool off soon)

Answer first: Demand is rising because generative AI workloads require far more compute and power than traditional enterprise software, and companies are shifting from pilots to production.

The report notes that the rise of AI is reshaping data-centre design due to “soaring power and cooling needs.” That’s the unglamorous part most teams in marketing and ops never see—until latency spikes, usage caps hit, or cloud bills arrive.

A few forces are colliding in 2026:

1. GenAI moved from “content toy” to “workflow engine”

The first wave was: “Write a blog post.”

The second wave (what we’re in now) is: “Connect the model to our product catalogue, our SOPs, our ticket history, and let it take first action.” That means:

• Longer context windows
• More retrieval (RAG) calls
• More tool calls
• More users using AI daily

Infrastructure load scales fast.

2. Capacity is getting locked in years ahead

Applied Digital is building capacity that comes online in 2027. That’s a time lag Singapore leaders should pay attention to.

When compute becomes tight, two things happen:

• Prices rise (or discounts disappear)
• Procurement gets slower (security reviews + capacity planning)

If your AI roadmap depends on “we’ll figure out compute later,” you’ll feel pain at the worst time—when your competitors are deploying automation in customer engagement and operations.

3. The market is publicly quantifying the spend

The report cites expectations that hyperscalers could invest more than US$400 billion annually into infrastructure. Even if that number shifts, the direction is what matters: infrastructure is where the budget is going.

For Singapore businesses, that’s a cue to stop treating AI tooling as an ad-hoc SaaS line item. It’s becoming a core capability, like cybersecurity or cloud.

What this means for Singapore businesses using AI tools

Answer first: The winners won’t be the companies that “use AI.” They’ll be the companies that pick one or two high-value workflows, instrument them, and scale responsibly—with predictable cost and governance.

Singapore is full of pragmatic operators. That’s an advantage. You don’t need a moonshot. You need repeatable improvements in:

• Marketing output and conversion
• Operational efficiency and forecasting
• Customer engagement and resolution time

Let’s translate the infrastructure trend into business actions.

Marketing: you’re not buying “content,” you’re buying throughput

If AI helps your team produce 3x more assets but your approval process stays manual and slow, you’ve improved the wrong constraint.

A better approach:

1. Use AI to generate variants (ads, emails, landing copy)
2. Put brand rules into templates and guardrails
3. Measure lift with a tight experiment loop (weekly)

Practical Singapore example: A retail brand running 30+ campaigns around seasonal periods (Hari Raya promotions, mid-year sales, year-end gifting) can use AI to draft campaign variants, but the real gains come from integrating:

• product feed + pricing
• UTM rules
• channel-specific formatting

This reduces errors and shortens cycle time.

Operations: AI is best at “repeatable judgment calls”

In ops, AI shines when decisions are frequent, semi-structured, and rely on internal data:

• reorder recommendations
• staffing forecasts
• invoice exception handling
• procurement triage

If you’re in Singapore and dealing with tight labour markets and rising costs, AI in operations isn’t optional forever—it’s margin protection.

Customer engagement: aim for faster resolution, not fewer agents

Too many companies deploy a chatbot to deflect tickets. Customers notice.

The better KPI is time-to-resolution and first-contact resolution rate. AI can:

• summarize tickets instantly
• draft replies that agents approve
• pull policy snippets from your knowledge base
• route issues to the right queue

That’s the “AI tools” layer—but the infrastructure trend tells you usage will grow. So you need to plan for scale: authentication, logging, redaction, and cost controls.

A practical 90-day plan to adopt AI tools (without getting burned)

Answer first: In 90 days, you can move from experimentation to measurable ROI by focusing on one workflow, one dataset, and one set of metrics.

Here’s what works in real businesses.

Step 1 (Days 1–14): Pick a workflow with clear dollars attached

Choose one:

• Marketing: reduce campaign production time by 30%
• Customer service: cut average handling time by 20%
• Sales ops: increase qualified lead follow-up speed by 50%
• Finance ops: reduce invoice exceptions by 25%

If you can’t attach a number, it’s a hobby.

Step 2 (Days 15–30): Fix data access and governance early

For Singapore teams, governance is where projects stall. Get ahead of it:

• classify what data is allowed (PDPA-sensitive vs non-sensitive)
• set retention rules for prompts and logs
• decide where the model runs (public cloud, private, vendor)
• require audit trails for customer-facing outputs

Step 3 (Days 31–60): Build guardrails, not a “perfect model”

Most companies get this wrong: they chase model quality while ignoring failure modes.

Add guardrails:

• approved tone and phrasing
• banned claims (pricing guarantees, medical/legal advice)
• mandatory citations to internal sources for policy answers
• human approval thresholds (e.g., refunds, cancellations)

Step 4 (Days 61–90): Track unit economics and scale the right way

This is where the infrastructure story becomes relevant.

Track:

• cost per 1,000 outputs (or per ticket)
• latency and uptime
• escalation rate to humans
• conversion lift or time saved

If costs scale faster than value, adjust:

• smaller models for simpler tasks
• caching for repeated answers
• stricter retrieval (less context stuffing)

“Do we need our own AI infrastructure in Singapore?” (Most don’t.)

Answer first: Most Singapore SMEs and mid-market firms don’t need to build infrastructure; they need smart procurement and architecture choices so costs and compliance stay under control.

Applied Digital’s story is about being an infrastructure supplier. Your job is to be a smart infrastructure buyer.

A sensible stance for most businesses:

• Use managed AI services for speed
• Keep sensitive data protected via redaction, access control, and private connectors
• Avoid vendor lock-in where it matters (store prompts, evals, and knowledge base in portable formats)

If you’re regulated (finance, healthcare) or running high-volume workloads, you may consider more controlled environments. But even then, the priority is governance + observability, not owning servers.

A useful rule: if you can’t explain your AI cost per customer interaction, you’re not ready to scale AI customer engagement.

Where this fits in the “AI Business Tools Singapore” series

Answer first: This infrastructure boom is the foundation beneath every AI tool you’re evaluating—marketing copilots, ops automation, and customer engagement assistants.

In this series, we keep coming back to a practical idea: AI adoption is less about choosing a trendy tool and more about building repeatable operating habits. Applied Digital’s revenue surge is a reminder that the market is already pricing in long-term AI usage.

So here’s the stance I’d take if I were advising a Singapore leadership team this week: treat AI like a capability you’ll still be using in three years. Budget accordingly. Put governance in place. Build one workflow that pays for the next.

If you want help mapping AI business tools to your actual workflows—marketing, operations, and customer engagement—start with a single process, a clear KPI, and a cost model that won’t surprise you when adoption grows.

What would your business look like if your team could run 20% faster without adding headcount—and could prove it on a dashboard?