Profit, Proof, and the SMB Aisle
If you only read one tech story this week, here's the line that sums it up:
“<em>Anthropic is on track for its first quarterly operating profit ever — roughly US$10.9B projected for Q2 — and has told investors it's open to a public listing by year-end. OpenAI's reasoning model independently produced a proof for an 80-year-old geometry problem, with Fields medalist Tim Gowers calling it a milestone. Anthropic also launched Claude for Small Business, embedding Claude directly into QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace and Microsoft 365. And Singapore's IMDA updated its Model AI Governance Framework for Agentic AI with industry input from 60+ organisations and 10+ real deployment case studies.</em>”
Last week the story was OpenAI planting its first Applied AI Lab outside the US in Singapore. The story now is what happens after that infrastructure is built — and the answer this week was unusually consistent: frontier labs are starting to act like public companies, frontier models are starting to act like research collaborators, and frontier software is starting to act like SaaS in the SMB aisle.
Here's what happened, why it matters, and what your business should do about it.
The Big Three
1. Anthropic's First Profitable Quarter — and an IPO Window Open by Year-End
Reports surfaced this week that Anthropic is projecting roughly US$10.9 billion in Q2 2026 revenue and is on track to post its first quarterly operating profit in company history. Investors and bankers have been told the company is open to a public listing by year-end, with an October 2026 window the most frequently cited. Valuation talk is circling US$900 billion.
To keep the model pipeline funded, Anthropic also expanded its compute partnership with SpaceX, agreeing to spend roughly US$1.25 billion per month through 2029 for access to the Colossus supercomputing infrastructure. And on May 22, OpenAI was reported to be preparing to confidentially file its own IPO prospectus with the SEC — at a potential valuation that would make it the largest tech listing in history.
The pattern is hard to miss. The two frontier AI labs that two years ago looked like research bets are now lining up to be the next generation of mega-cap public companies. That's a meaningful shift in risk profile: the businesses your SME relies on for AI capability are about to come under quarterly-earnings discipline, with all the upsides (predictable roadmaps, audited financials, real SLAs) and downsides (cost discipline, monetisation pressure, end of "free everything") that implies.
2. OpenAI's Model Cracked an 80-Year-Old Math Problem — Unprompted
In a result that's circulated widely among researchers this week, OpenAI announced that one of its internal general-purpose reasoning models independently produced a proof for an 80-year-old conjecture in additive combinatorics, finding an infinite family of configurations that beat the previously known bound. Princeton mathematician Will Sawin quantified the improvement; Fields medallist Tim Gowers called the result "a milestone in AI mathematics."
The model wasn't fine-tuned for the problem, wasn't guided step-by-step, and wasn't retrieving a known solution. It was handed the statement and produced the proof.
In the same week, Anthropic co-founder Jack Clark delivered the 2026 Cosmos Lecture at Oxford on May 20 and made four predictions for the next 12 months — including that AI will work alongside humans to produce a Nobel Prize-worthy discovery before May 2027. A year ago that claim would have been treated as marketing. After this week's geometry result, it's just calibration.
The practical signal for businesses: the quality ceiling of what AI can do on hard analytical work moved this week, not the floor. If your business has a category of problem that you've quietly assumed is "too judgement-heavy for AI," the half-life of that assumption just got shorter.
3. Claude Just Walked Into QuickBooks, PayPal and HubSpot
This is the story that matters most for SMEs.
Anthropic launched Claude for Small Business — a package of connectors and automated workflows that puts Claude directly inside Intuit QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace and Microsoft 365. Paying subscribers unlock 15 pre-built workflows across finance, ops, sales, marketing, HR and customer service, plus 15 skills aimed at the tasks owners flagged as their biggest time sinks. The example workflows are unusually concrete: reconcile QuickBooks cash position against incoming PayPal settlements to plan payroll, close monthly books and generate a plain-English P&L summary, chase overdue invoices automatically.
There's no extra charge beyond the existing Claude licence and whatever partner tools the business already pays for. Existing permissions inside connected tools carry through — if a staff member can't see a QuickBooks ledger today, they can't see it through Claude either.
Anthropic also expanded its strategic alliance with PwC, with Claude Code and Cowork rolling out across PwC's US teams first, then globally toward hundreds of thousands of professionals.
The shift is structural. For the last 18 months, "using AI in your business" meant copy-pasting into a chat tab in a browser. From this week onward, it means opening QuickBooks the way you always have, and finding the workflows already there. That's a much shorter behavioural distance for an SME owner to travel.
Closer to Home: Singapore Wrote the Operating Manual for Agents
While the global labs were filing IPO paperwork, Singapore did something quieter but arguably more useful for the businesses that actually deploy this stuff: IMDA updated the Model AI Governance Framework for Agentic AI.
The original framework launched in January 2026 at Davos as the world's first governance framework specifically designed for AI agents that can plan, reason and act autonomously. The May update is the substantive one:
- Industry input from 60+ organisations, including AWS, DBS, Google and Salesforce, on how to manage multi-agent systems, third-party agents, and automation bias.
- More than 10 real-world case studies of agentic deployments — contributed by Ant International, City Developments Limited (CDL), Cyber Sierra, Dayos, Google, Knovel, OCBC, PwC, Stability Solutions, Tencent, Terminal 3, Workday, X0PA and GovTech Singapore.
- New best practices for the failure modes that actually trip up agent deployments in production: third-party agent risk, multi-agent coordination, and the human tendency to over-trust an agent's output (automation bias).
Stack the timing. Anthropic is filing for IPO. OpenAI is filing for IPO. Both are pushing agentic deployment into enterprises at billion-dollar scale. And Singapore — in the same week — published the checklist for how to do that responsibly inside an organisation, with case studies from companies that have already done it.
For Singapore SMEs, that's not abstract. The IMDA framework is now the closest thing the market has to a standard reference document when you're asked "how are you governing your AI agents?" by a customer, an auditor, or a banking compliance team. It is short, English-language, and grounded in real deployments — which is unusual for any government-issued framework, anywhere.
And it lands in a market that's already moving. SME AI adoption tripled to 14.5% in the latest Digital Economy Report. The Digital Enterprise Blueprint, Grab AI Programme (10,000 F&B/retail SMEs), DBS Spark GenAI, the IMDA pre-approved AI solutions catalogue, OpenAI for Singapore, and now an updated agentic governance framework — that's a stack you can actually plan around.
What This Means for Your Business
Three takeaways for SMEs in Singapore this week:
1. "Nobody's making money on AI yet" is no longer a usable excuse
For two years, one of the soft objections to investing in AI internally has been a version of "the labs themselves aren't profitable — it'll all collapse." That argument got materially harder to make this week. Anthropic is projecting a profitable quarter on roughly US$10.9B in revenue and is openly preparing for a public listing. OpenAI is doing the same. Whatever your view on the long-term economics, the vendors you depend on are about to be public companies with audited financials and quarterly accountability. The risk of "the AI we built on disappears" went down this week, not up.
2. If your business runs on QuickBooks, Xero, HubSpot or PayPal, you're now AI-ready by default
This is the single most actionable shift of the week. Claude for Small Business plugs into exactly the SaaS stack a typical Singapore SME already pays for. You don't need to pick a model, write a prompt, or buy a new tool. The relevant SME exercise is no longer "should we try AI?" — it's:
- List the 5 software tools your business actually depends on.
- Check which ones already have a Claude (or equivalent) integration shipping in 2026.
- Pick one workflow per tool and try the pre-built agent against it for a week.
For most SG SMEs, that exercise is a 90-minute Saturday morning, not a strategy project.
3. Treat the IMDA Agentic AI Framework as a one-page internal checklist, not a regulation
Most SME owners will never read the full framework. They don't have to. The four dimensions — bound the risk upfront, keep a human meaningfully accountable, put technical controls in place, enable end-user responsibility — fit on a single A4 page. Before you let an agent send an email on your behalf, push a journal entry, or talk to a customer, run those four checks. The framework's value to an SME isn't compliance theatre — it's structured thinking borrowed from organisations like OCBC, GovTech and Ant International that have already shipped agents into production.
The Practical Question
The headlines this week were trillion-dollar IPOs, 80-year-old maths problems, and billion-dollar consulting alliances. The relevant question for most Singapore SMEs is smaller and more immediate:
"Which of my monthly back-office tasks is already living inside a tool that now ships its own AI agent — and what stops me from turning that agent on tomorrow?"
The QuickBooks reconciliation that eats the first Tuesday of every month. The HubSpot follow-up sequence that goes stale after week two. The DocuSign chase emails for the contracts sitting in someone's drafts. The Canva post that took 40 minutes longer than it should have.
This week, the lab that builds the model crossed into operating-profit territory. The model crossed into original mathematics. And the model's product crossed into the menu bar of the accounting software you opened on Monday. The interesting question for your business isn't whether to participate in that shift — it's which of the workflows you already run is the first one worth handing over.
At The Empyrean, we work with Singapore SMEs to find the practical, repeatable tasks where AI delivers value without disruption — and to set them up with the kind of governance that earns customer trust. If you're not sure where to start, we're happy to take a look at your operations and tell you honestly what would make sense.