AI Infrastructure Boom: What Singapore Businesses Should Do

Applied Digital’s latest quarter is a useful signal, not because of the company itself, but because of what sits underneath it: enterprises are buying AI compute like they buy electricity—at scale, and on long contracts.

According to Reuters reporting carried by CNA, Applied Digital’s third-quarter revenue rose 139% to US$126.6 million, beating analysts’ estimate of US$76.6 million. The driver wasn’t a new app or a clever marketing campaign. It was data centre services—the unglamorous layer that makes generative AI possible. Source article: https://www.channelnewsasia.com/business/applied-digital-beats-quarterly-revenue-estimates-surging-ai-infrastructure-demand-6044896

For the AI Business Tools Singapore series, this matters because it changes how Singapore companies should plan AI adoption. The bottleneck is shifting away from “Which model should we use?” toward “Do we have reliable, affordable access to compute—and do we know how to turn it into business outcomes?”

What Applied Digital’s results really tell us about AI demand

AI infrastructure demand is no longer speculative—it’s contractual. The CNA/Reuters piece highlights that big tech and AI companies are signing long-term, multi‑billion‑dollar deals to lock in power and data centre capacity. That’s the most practical proof you’ll get that AI workloads are becoming permanent.

Two numbers from the report are especially telling:

• Applied Digital’s quarterly revenue: US$126.6M (up 139%) vs estimate US$76.6M (LSEG)
• In 2025, 17 major high‑performance computing deals worth more than US$70B were announced (B. Riley Securities analysts)

Here’s my take: when infrastructure spending outpaces software spending, it means usage has moved from pilots to production. Companies don’t reserve power and cooling for experiments.

The hidden constraint: power and cooling, not “AI ideas”

Generative AI changes data centre design because GPU-heavy workloads draw huge power and generate dense heat. That’s why the article emphasises “soaring power and cooling needs.”

For business leaders in Singapore, the translation is simple:

If your AI roadmap assumes unlimited compute at stable prices, it’s already outdated.

Even if you never buy a GPU server, you’ll feel this through:

• higher unit costs for AI features in SaaS tools
• longer wait times for dedicated capacity
• stricter usage limits (tokens, seats, throttling)
• more pressure to show ROI before expanding usage

Why this is especially relevant in Singapore

Singapore businesses adopt AI differently because space, talent, and regulatory expectations are tighter. You can’t “just hire 20 ML engineers and build a GPU cluster” as a default plan. Most mid-market firms here will win by combining:

• the right AI business tools (marketing, ops, finance, support)
• disciplined process redesign
• careful data governance
• selective use of custom models only where it pays

The infrastructure boom described in the Applied Digital story supports a practical conclusion for Singapore:

AI is being priced like a utility—so manage it like one

When hyperscalers pour massive capital into infrastructure (the Reuters piece cites Applied Digital’s expectation that hyperscalers could invest US$400B annually), they’ll still need customers to use that capacity efficiently. This leads to two outcomes:

1. AI costs become more transparent (usage-based pricing everywhere)
2. Waste becomes obvious (unused seats, duplicate tools, “prompt spam,” overbuilt chatbots)

Singapore companies that treat AI spend like cloud spend—measured, governed, optimised—will have an advantage.

What to do next: a business-first playbook for AI adoption

The best time to get serious about AI governance and ROI was last year. The second best time is this quarter. The infrastructure land-grab is your signal that AI is settling into the cost base of doing business.

1) Separate “AI capability” from “AI outcomes”

Buying AI tools doesn’t automatically create business value. Outcomes come from pairing tools with a measurable workflow change.

Use this quick mapping:

• Outcome: reduce customer response time by 30%

  • Workflow: route tickets, draft replies, enforce tone + policy
  • Tool type: AI helpdesk assistant + knowledge base search
  • Metric: median first response time, reopened tickets

• Outcome: increase inbound leads by 20%

  • Workflow: create content briefs, optimise landing pages, test ads
  • Tool type: AI content + ad creative + analytics copilots
  • Metric: cost per lead, conversion rate, sales-qualified leads

If you can’t tie an AI feature to a metric you already track, it’s usually a “nice demo,” not a deployment.

2) Budget for compute indirectly (even if you don’t run servers)

Most Singapore SMEs consume AI through SaaS, but SaaS vendors pass through compute costs. Plan for it.

A simple approach I’ve found effective:

• Create an “AI usage budget” (monthly) for teams using gen AI heavily
• Track three numbers per department:
  • spend (subscriptions + usage)
  • volume (tickets handled, pages produced, calls summarised)
  • quality (CSAT, conversion, error rate)

This turns AI from a vague innovation line item into a controllable operating expense.

3) Use a “thin layer” architecture: tools first, custom second

Most companies get this wrong by starting with custom builds. If your goal is leads (marketing) and efficiency (ops), start with proven AI business tools, then add custom pieces only where differentiation is real.

A practical stack many Singapore teams can run with:

• Marketing: AI for content briefs, SEO outlines, ad variants, call analysis
• Sales: meeting notes + CRM updates + objection libraries
• Customer support: drafting + knowledge retrieval + QA sampling
• Operations: invoice extraction, reconciliation, SOP assistance
• Management: KPI narration, variance explanations, forecasting assistants

Custom models make sense when you have:

• proprietary data that materially improves performance
• enough volume to justify engineering
• clear risk controls (privacy, security, auditability)

4) Make “accuracy” a workflow design problem

Gen AI errors are inevitable; unmanaged errors are optional. The fix isn’t shouting “be accurate” at a model—it’s designing checkpoints.

For high-risk workflows (finance, legal, regulated customer comms), use:

• retrieval from approved documents (not open-ended answers)
• required citations to internal sources
• confidence thresholds and escalation rules
• human approval on external-facing messages
• sampling-based QA (e.g., review 5% weekly)

This is where Singapore companies can be stricter than the average global rollout—and that discipline often becomes a brand advantage.

The ROI question: what infrastructure spending implies for your KPIs

When AI infrastructure demand surges, management teams get less patient with “AI for AI’s sake.” That’s good news for operators: you can win budget by being specific.

Here are ROI patterns that consistently survive scrutiny:

Marketing ROI that’s believable

• Speed-to-publish: cut content cycle time (e.g., from 10 days to 3)
• Experiment volume: run 4–8 creative tests per month instead of 1–2
• Conversion lift through iteration: improve landing page clarity, offers, FAQs

The key is to measure what changes. “We used AI to write posts” isn’t a KPI.

Operations ROI that finance will accept

• Time saved per transaction: invoices, claims, reconciliations
• Error reduction: fewer reworks, fewer missed fields
• Throughput: more volume handled with same headcount

If you’re chasing leads (the campaign goal), don’t ignore ops. A sales pipeline falls apart when onboarding, support, or billing can’t keep up.

Common questions Singapore leaders ask (and direct answers)

“Should we invest in our own GPUs?”

For most SMEs and mid-market firms in Singapore: no, not as a first move. You’ll spend time on procurement, security, uptime, and model maintenance—then still rely on cloud services.

A better first milestone: standardise 2–3 AI business tools across teams, track ROI, and tighten data access.

“Are AI tool costs going to keep rising?”

Costs will fluctuate, but usage-based pricing is here to stay. As power and cooling become strategic constraints (as described in the CNA/Reuters report), vendors will price accordingly. Your defence is governance and optimisation, not wishful thinking.

“What’s the safest place to start?”

Start where:

• data sensitivity is manageable
• output is easy to review
• metrics are already tracked

Typically: marketing drafts, internal knowledge search, meeting summaries, ticket triage (with human approval).

A practical next step for Singapore teams adopting AI tools

Applied Digital’s earnings beat is a headline, but the useful message is operational: AI is becoming infrastructure-heavy, contract-heavy, and ROI-heavy. That’s the environment Singapore businesses are adopting AI inside.

If you’re leading marketing, ops, or customer experience, your next step isn’t to chase the newest model. It’s to pick one workflow (lead gen, customer support, finance ops), instrument it with metrics, and deploy an AI tool stack with clear guardrails.

The companies that benefit most from the AI infrastructure boom won’t be the ones with the loudest “AI strategy” slides. They’ll be the ones who can answer a simple question in a board meeting: “What did AI change in our numbers this month?”