The Week Singapore's AI Choices Got Wider
If last week was about AI starting to do things, this week was about the structure underneath it shifting — quietly but unmistakably.
The Microsoft–OpenAI partnership got rewritten. Big Tech penciled in $700 billion of AI spend for this year alone. Alphabet posted its fastest revenue growth in nearly four years. And PM Lawrence Wong took to the stage at home to talk directly to workers about AI.
The headlines look like Silicon Valley plumbing. The downstream effect lands here.
1. Microsoft and OpenAI Reset the Most Important Deal in AI
On April 27, Microsoft and OpenAI announced a major restructuring of their partnership.
The three things that matter:
- Microsoft's exclusive cloud arrangement is over. OpenAI can now serve its products on Amazon Web Services and Google Cloud — not just Azure.
- The Azure revenue share Microsoft was paying OpenAI ends.
- OpenAI keeps paying Microsoft a capped 20% revenue share through 2030.
In plain language: the most consequential AI lock-in of the past five years just loosened. OpenAI can now sell into customers on whatever cloud those customers are already on. Enterprises that standardised on AWS or GCP no longer have to route through Azure to use GPT.
For businesses, the meta-story is vendor optionality. The frontier labs are no longer single-cloud creatures. The cloud you pick for your data and the model you pick for your AI are decoupling.
2. Big Tech Will Spend $700 Billion on AI This Year
Fortune reported on April 30 that the world's hyperscalers — Microsoft, Google, Amazon, Meta, Oracle — are now on track to spend roughly $700 billion on AI infrastructure in 2026, with no clear ceiling in sight.
For context: that's more than the GDP of Singapore, Switzerland, and Norway combined. In one year. On data centres, chips, power, and cooling.
Why this matters for an SME owner reading this in Singapore: none of that money makes sense unless the supply creates demand. The companies building this capacity are betting that, within 24 months, AI workloads will look as ordinary as cloud workloads do today — embedded in every reporting tool, CRM, support desk, and back-office process.
The pricing curve continues to bend down. The only question left is who captures the productivity gains first.
3. Alphabet's Cloud Hit a New Milestone
Alphabet reported quarterly results this week showing 22% revenue growth — its fastest in nearly four years. The standout: Google Cloud crossed $20 billion in quarterly revenue for the first time, driven almost entirely by AI workloads.
This lines up with what we covered last week — Google's $40B commitment to Anthropic, the new Gemini Enterprise Agent Platform, and the latest TPUs. The investment is now showing up in the financials, which means more capacity, more sales pressure, and more enterprise-grade tooling aimed squarely at mid-market businesses.
The practical takeaway: if you've been quoted enterprise pricing on AI six months ago, ask again. The economics have shifted twice since then.
4. AI Agents Started Closing Real Deals
A quietly important data point this week: Anthropic published early results from "Project Deal" — an internal experiment where Claude-powered agents conducted commerce on behalf of real users. The result: 186 completed deals across e-commerce, bookings, and services.
Meta separately disclosed its business AI is now handling roughly 10 million conversations per week. Amazon rolled out AI-powered audio Q&A on product pages. Yelp expanded transactional booking via its assistant (covered last week).
The pattern is consistent: the assistant layer is no longer answering questions. It's pulling out a wallet.
If your business depends on a customer journey that ends in a booking, an order, or a quote — the relevant question is whether an AI agent can actually complete that journey on a customer's behalf. If your flow assumes a human clicking buttons, you're invisible to a fast-growing channel.
5. The Quieter Risk: Prompt Injection in the Wild
The security story of the week, but rarely on the front page: researchers and CISOs are flagging a new class of attack where malicious actors seed public web pages with hidden instructions. Any enterprise AI agent that scrapes those pages can be redirected — with the agent's real, approved credentials — to leak data, send emails, or execute actions inside its own company.
Traditional security tooling sees nothing wrong, because nothing technically is wrong. The agent is doing what it was told.
For SMEs deploying agents: the lesson isn't to stop. It's to scope tightly. An agent that can read should not also be able to write. An agent that can draft should not also be able to send. Deny by default; permit deliberately.
Closer to Home: Singapore's AI Conversation Got Personal
While Big Tech rewrote contracts, Singapore's leadership turned the conversation toward people.
- On May 1, PM Lawrence Wong addressed workers directly on AI fears, framing AI as a force the country will navigate with workers, not around them. The pitch: training, transition support, and an explicit commitment that productivity gains should be shared.
- DBS continued expanding its Spark GenAI programme for SMEs with a new tiered structure and a free assessment tool — a practical on-ramp for businesses still figuring out where to start.
- Budget 2026 measures, which kicked in on April 1, have now been live for a full month. The headline change for outward-looking SMEs: the Market Readiness Assistance grant has been raised from 50% to 70% of eligible costs through March 2029.
- Microsoft's $5.5B Singapore AI investment continues its rollout, including free Microsoft 365 Premium with Copilot for every tertiary student in the country.
The theme connecting these: Singapore is not asking SMEs to find the AI wave. It's putting the wave on the doorstep, with co-funding attached.
What This Means for Your Business
Three practical takeaways from the week:
1. Vendor lock-in is worth re-examining
The Microsoft–OpenAI restructure means your cloud and your AI no longer have to be from the same vendor. If you're paying enterprise tax for that bundling, the leverage just shifted in your favour. Renegotiate.
2. AI is a buyer now — make sure it can buy from you
Project Deal, Yelp transactions, Amazon's Q&A on product pages. The customer at your checkout is increasingly an agent representing a person. If your booking, ordering, or enquiry flow can't be completed by software, you'll quietly lose volume — without ever seeing the lost customers.
3. Agents need scope, not just power
As you deploy agents internally, give them the narrowest useful permission set. The week's prompt-injection stories aren't theoretical — they're operational. The companies that win with agents will be the ones that ship them with explicit, auditable boundaries from day one.
The Practical Question
The headlines this week were about $700B of AI infrastructure and rewritten partnerships. The news cycle treats those as financial events. They aren't. They're a signal.
The relevant question for an SME owner in Singapore isn't "Which model should I be using?"
It's:
"What part of my business would I want an agent to do — and what guardrails would let me trust it?"
A year ago that was a hypothetical. This week, the tooling, the pricing, and the local funding to answer it all moved in your direction simultaneously.
At The Empyrean, we help Singapore SMEs cut through the noise around AI and identify the practical, scoped tasks where it actually delivers value — without ripping up the systems you already run on. If you'd like an honest read on where to start, we're happy to take a look.