If your employer rushed to roll out AI tools over the past year, the invoice is about to get interesting. Across the global tech industry, companies that treated AI like an all-you-can-eat buffet are now staring at bills they struggle to explain, and the reckoning carries a warning for every Malaysian business betting on the same technology.
Editor
Kai T chevron_right
Table of Contents
Companies are blowing through their AI budgets
TechCrunch reports that Uber burned through its entire 2026 AI coding budget by April, while Microsoft pulled back its developers' access to Anthropic's Claude Code only months after handing it out. The travel firm Priceline said a routine renewal of its Cursor coding contract came back four to five times more expensive. In one widely cited case, a single company ran up a 500 million US dollar bill from Anthropic after forgetting to set spending limits for its staff.
The cause is not higher prices. Per-token rates have actually fallen. The problem is volume. Newer agentic AI tools, which work through tasks on their own, consume far more tokens, the units providers bill by, than the chatbots they replaced. J.R. Storment of the FinOps Foundation told TechCrunch that the mood flipped almost overnight from chasing speed to asking, "we need guardrails, how do we control this?"
The response is now formal. This month the Linux Foundation said it will set up a Tokenomics Foundation, a standards body meant to bring cost discipline to AI the way an earlier effort did for cloud spending. Early backers include Microsoft, Google Cloud, IBM, Oracle, Salesforce, SAP and JPMorganChase, with a formal launch planned for July. Goldman Sachs projects that global token usage will grow about 24 times by 2030, so the bills are pointed in one direction.
What this means for Malaysia
Malaysia is pressing the accelerator at exactly this moment. Budget 2026 set aside RM2 billion for a sovereign AI cloud under the MCMC, RM53 million for the Malaysia Digital Acceleration Grant that helps smaller firms adopt AI, and RM18 million for the new National AI Office, all in service of the country's AI Nation 2030 goal. MDEC says AI adoption among Malaysian businesses has grown 35 percent in a year, with roughly 2.4 million companies now using it in some form.
That enthusiasm is the easy part. The harder part is the lesson global enterprises are learning right now: AI is cheap to start and expensive to run at scale. Most of these tools bill in US dollars, so a softer ringgit quietly inflates every invoice. A Malaysian SME that takes a grant to pilot an AI agent can still be ambushed when the bill triples once staff use the tool in earnest, the same trap that caught much larger companies abroad.
What Malaysian teams can do now
The fixes are not exotic. Set hard usage caps before a tool reaches a whole team, not after the first shock invoice. Track spending against real output, since one study found the heaviest users were about twice as productive but burned ten times the tokens to get there. Route routine work to cheaper models and reserve the expensive ones for problems that genuinely need them. Malaysian developers saw an early version of this squeeze when GitHub Copilot switched to token-based billing, and the wider market is now catching up.
Malaysia's AI ambitions are reasonable, and the public money behind them is real. The open question is whether local firms build the cost discipline to match, before the token bill, as it has overseas, comes due.
Images courtesy of İsmail Enes Ayhan and Compagnons on Unsplash.