AI Demand Is Squeezing Memory—What SG Firms Do Next

Applied Materials just told the market a blunt truth: AI is consuming so much compute capacity that it’s starting to reshape the memory market—and the companies that supply chipmaking tools are cashing in. In its latest outlook (reported by Reuters via CNA on 13 Feb 2026), Applied forecast Q2 sales of about US$7.65B (±US$500M) versus US$7.01B estimated, and adjusted EPS of about US$2.64 (±US$0.20) versus US$2.28 estimated. Shares jumped over 12% after hours.

If you’re running a business in Singapore, it’s tempting to file this under “semiconductor industry news.” Don’t. This matters because the same AI build-out that’s driving chip equipment demand is also tightening memory supply, pushing costs and lead times into places your business actually feels—cloud bills, device pricing, procurement delays, even how quickly your team can ship AI features.

This post is part of the AI Business Tools Singapore series, where we look at what AI shifts mean for real operations: marketing, planning, customer service, and finance. The headline here isn’t “buy chip stocks.” It’s: plan for volatility—and use AI business tools to stay fast when infrastructure isn’t.

A useful rule for 2026: if AI infrastructure is constrained, your AI roadmap becomes a planning problem, not just a product problem.

What Applied Materials’ forecast is really signalling

Answer first: Applied’s upbeat forecast is a leading indicator that AI infrastructure spending remains aggressive in 2026, and that memory (especially advanced DRAM) is becoming a bottleneck.

Applied Materials sells the equipment used to manufacture chips. When Applied says demand is strong, it’s often because chipmakers are expanding or upgrading production lines. In this update, two drivers stand out:

• AI processors (the “brains” like GPUs/CPUs and accelerators)
• A worldwide memory shortage, as AI systems absorb huge volumes of memory

The CNA report highlights that AI infrastructure build-outs have “absorbed much of the world’s memory chip supply,” which then drives higher memory production capacity investments—and that feeds right back into Applied’s sales.

Why memory is the pinch point (and why it’s not just “more RAM”)

Answer first: Modern AI workloads are memory-hungry, and the market is prioritising high-bandwidth memory (HBM) and advanced DRAM, which are harder to scale than standard components.

HBM is built by stacking DRAM layers. It’s often paired with high-end AI processors (Nvidia is the common reference point) because it moves data faster and supports the throughput AI training and inference need.

Applied’s CEO called out that demand for higher performance and more energy-efficient chips is driving growth in:

• Leading-edge logic (compute)
• High-bandwidth memory (HBM)
• Advanced packaging (including techniques like 3D chiplet stacking)

That’s a strong signal that 2026 isn’t about generic compute capacity. It’s about specific, premium AI infrastructure.

What this means for Singapore businesses using AI tools

Answer first: Expect AI costs and timelines to stay uneven in 2026, because infrastructure constraints show up as cloud pricing pressure, hardware lead times, and model deployment trade-offs.

Singapore companies are heavy cloud users, and many are now adding AI across functions: customer support copilots, demand forecasting, document automation, fraud detection, personalised marketing. When memory supply tightens, you’ll see second-order effects:

1) Cloud AI spend becomes less predictable

Even if you never buy servers, your vendors do. A memory squeeze can flow through to:

• Higher per-token or per-request costs for premium models
• More aggressive pricing tiers for “fastest” endpoints
• Capacity limits for certain regions or instance types

Practical implication: finance teams need better forecasting, and product teams need cost guardrails.

2) Hardware refresh cycles may slip

Retail, healthcare, logistics, and manufacturing in Singapore often run edge devices: scanners, kiosks, cameras, rugged tablets, industrial PCs. If components tighten or prices spike, refresh plans get messy.

Practical implication: treat device strategy as part of the AI roadmap, not separate from it.

3) Model choice becomes an operations decision

When compute is tight or expensive, teams get pushed toward smarter engineering:

• Smaller or distilled models
• Retrieval-augmented generation (RAG) to reduce context and compute
• Caching, batching, and async pipelines
• Hybrid setups (some tasks on-device, others in cloud)

This is where AI business tools stop being “experiments” and start being operational infrastructure.

The better way to plan: treat AI as supply-chain exposed

Answer first: The companies that win with AI in 2026 will run AI adoption like a supply-chain discipline: scenario planning, vendor diversification, and tight measurement.

Most businesses plan AI like software—ship features, iterate, scale. That’s only half the picture. AI depends on scarce inputs: compute, memory, skilled talent, and data readiness.

Here’s a planning approach I’ve found works better for Singapore SMEs and mid-market teams.

Build three scenarios for AI capacity and cost

Answer first: You don’t need perfect forecasts; you need decision-ready scenarios that tell you what to cut, pause, or accelerate.

Create three scenarios for the next 2–3 quarters:

1. Base case: steady costs, stable capacity
2. Tight capacity: premium endpoints restricted, cost up 15–30%
3. Relief case: more competition/capacity, cost down 10–20%

Then map each scenario to specific actions:

• Which use cases stay (customer support automation, internal search, forecasting)?
• Which use cases pause (high-cost experimentation, large-scale generation)?
• What’s your fallback model/provider?

This isn’t theoretical. Applied’s forecast and the memory shortage narrative are telling you that tight capacity is a real planning input.

Put measurement where it belongs: unit economics

Answer first: AI projects fail quietly when teams track “accuracy” but ignore cost per outcome.

For each AI workflow, track one “business unit metric,” such as:

• Cost per resolved support ticket
• Minutes saved per invoice processed
• Cost per qualified lead
• Stockout reduction per dollar spent on forecasting

When memory and compute costs swing, you can still decide quickly because you know what you’re buying.

Where AI tools help most during shortages: forecasting and operational agility

Answer first: AI can’t manufacture DRAM, but it can reduce the damage of shortages by improving forecasting, procurement timing, and workload efficiency.

This is the bridge from semiconductor headlines to everyday Singapore operations.

AI for forecasting: demand, inventory, and cash flow

In the same way Applied uses demand signals to project revenue, businesses can use AI to forecast:

• Sales demand by SKU/channel
• Inventory reorder timing (especially for imported items)
• Cash flow stress under different cost assumptions

If you’re in Singapore and rely on regional supply chains (Malaysia, China, Vietnam, Indonesia), forecasting matters even more because lead times can change quickly.

A solid starting point:

• Use time-series models or automated forecasting tools
• Add external signals where relevant (promotions, seasonality, shipping lead times)
• Run weekly rolling forecasts instead of quarterly “big updates”

AI for workflow efficiency: do more with less compute

Not every AI feature needs the biggest model. If infrastructure stays tight, efficiency becomes a competitive advantage.

Practical tactics:

• Route requests: simple queries go to cheaper models; complex ones go premium
• RAG-first design: retrieve relevant internal docs before generating
• Summarise once, reuse often: cache summaries for repeated documents
• Batch non-urgent tasks: run overnight to reduce peak costs

This is where AI business tools in Singapore should be evaluated on control as much as capability: cost controls, monitoring, routing, and governance.

“People also ask” (Singapore context)

Will the memory shortage affect SMEs in Singapore?

Answer: Yes, indirectly. SMEs feel it through cloud AI pricing, device costs, and service capacity limits, even if they never buy chips.

Does this mean AI projects should pause in 2026?

Answer: No. It means AI projects should be prioritised. Focus on use cases with clear ROI and measurable unit economics, and design for model/provider flexibility.

What should I do if my AI tool costs suddenly rise?

Answer: Have a playbook: cap spend, switch to smaller models, add caching/RAG, batch workloads, and renegotiate enterprise pricing based on volume and predictability.

How to turn this moment into a lead for better AI operations

Applied Materials’ forecast is a reminder that AI is now a physical economy story—factories, memory stacks, packaging techniques, and capacity constraints. That’s not “someone else’s problem.” It shows up in your budgets and timelines.

If you’re building AI into operations in Singapore, the next step is to make your AI stack more resilient:

• Decide which workflows must run on premium models
• Engineer everything else for cost and flexibility
• Build scenario forecasts into finance and ops reviews

The question worth asking your team this quarter: If AI capacity tightens again, which workflows keep running—and which ones break first?

Source article (CNA landing page): https://www.channelnewsasia.com/business/applied-materials-forecasts-upbeat-results-ai-demand-memory-shortage-5927826