Grab’s big AI wager: how a Southeast Asian super app thinks agents and groceries will triple profit by 2028
A driver in Jakarta taps an AI prompt, a merchant in Manila receives an automatically optimized delivery window, and the company that connects them both plans to turn those micro efficiencies into a $1.5 billion EBITDA prize.
On the surface the story reads like another growth target from a tech company riding a recovery in travel and delivery. The obvious headline is that Grab is back to profit and chasing scale with new services and tighter unit economics. According to Reuters, however, the thread that actually matters for the AI industry is Grab’s explicit plan to use AI and product bundling to triple EBITDA to $1.5 billion by 2028, which reframes the conversation from two-sided marketplaces to intelligent operating systems. (investing.com)
Why super apps in emerging markets are an AI industry testbed
Southeast Asia offers dense transaction data, frequent repeat behavior, and fragmented offline networks that make automated decisioning lucrative. In markets where users open one app to do many things, a single high-quality agent can save minutes on every transaction and turn that time into measurable revenue uplift. The region is essentially a live lab for agentic UX and verticalized models, which is precisely what AI vendors crave when proving product-market fit outside Silicon Valley. A few bad jokes about driver incentives being a loyalty program for patience are permitted, but not necessary.
The mainstream read and the sharper lens investors should use
The mainstream read is simple: cut subsidies, grow groceries and financial services, and profit follows. That is true at a surface level. The underreported point is that achieving those margins requires not incremental automation but a suite of AI agents that change how users interact with the app, not just what they buy. The difference between faster routing and a true agent is the latter acting on behalf of the user across commerce, credit, and scheduling, and that is where vendors selling models, observability, and agent orchestration stand to make recurring revenue.
The numbers: what Grab actually said and why the math matters
Grab has said it expects revenue to grow more than 20 percent year to year for the next three years and to triple adjusted EBITDA to about $1.5 billion by 2028. That is the headline metric that converts AI investment into an ROI story for public markets. Reuters captured the plan as quoted from President and Chief Operating Officer Alex Hungate during an interview at Grab’s Singapore headquarters. (investing.com)
Grab’s public filings and investor materials show the company exited 2025 with its first full year of net profit and an adjusted EBITDA base that management intends to scale using efficiency and new high margin products. The investor presentation lays out the mechanics for freeing up cash to invest in AI, fintech, and core logistics, which in turn drives the unit economics that feed the 2028 targets. (investors.grab.com)
How Grab expects AI to create real margins, not just buzz
AI is being positioned across three buckets: demand orchestration to reduce empty mileage, merchant and driver assistants to lower churn and transaction times, and underwriting and personalization for financial services. Each bucket has a clear PnL lever. For example, a 5 percent cut in empty mileage across a delivery fleet can reduce variable cost per order by roughly 3 to 4 percent, which layered over billions of dollars in gross merchandise value compounds rapidly. The arithmetic is simple and merciless; small operational gains scale fast in marketplaces, which is why model providers will be asked for both latency and provable cost savings, not clever hallucination fixes.
AI at Grab will be judged by routes saved and loans repaid, not fancy demos or splashy demos.
Payments, stablecoins, and the commercial plumbing that makes agents useful
Grab is not just automating moves and messages. It is lining up payments partners and exploring programmable rails so agents can transact directly on behalf of users. Industry moves with Visa and others to enable AI-driven commerce make it possible for an agent to complete a checkout or schedule a recurring grocery order without human re-authentication. TechNode reported that payment networks and platform partners are actively designing interfaces and tokenization for this exact purpose, which changes the product ask from single-call APIs to end to end agent commerce. (technode.global)
Practical scenarios businesses should model now
A restaurant chain integrating Grab’s merchant agent could see delivery scheduling automatically batch orders to reduce driver idle time by 25 percent in off peak hours. If that reduces delivery cost per order by $0.30 on a 500,000 monthly orders base, annual savings would be about $1.8 million, before counting the revenue upside from faster throughput. For fintech partners, a point improvement in loan repayment prediction could increase lending yield by several basis points while cutting loss rates, converting marginal AI signal monetization into low friction revenue. These are the kinds of tables CFOs will ask staff to run, which means model explainability and audit logging are now product requirements, not optional features.
The cost nobody is calculating and the talent gap
Heavy use of real time agent inference and on device or edge deployments will increase compute budgets materially. Foundational model API bills, fine tuning, and latency requirements will push platforms toward hybrid architectures supporting local inference. Meanwhile, hiring the 100 to 200 engineers and ML ops staff needed to deliver and maintain agent workflows in multiple languages is expensive and slow. The talent market in Southeast Asia is tight, which is why partnerships and white label agent frameworks are likely to proliferate. A dry aside for those who enjoy paradoxes: scaling with AI often starts with hiring more humans to automate humans.
Risks that could blow up the arithmetic
Regulatory backstops on data and payments, slower user adoption of agentic flows, or AI costs that do not fall as expected could all derail the 2028 pathway. Analysts have flagged that investments in autonomous vehicle partnerships and AI could weigh on near term profitability, a point that underlines the difference between promising roadmap metrics and realized free cash flow. Any model that touches credit or payments will also increase compliance overhead, so the company’s ability to marshal regulatory capital and oversight is a gating factor. (investing.com)
What this means for AI vendors and startups
Companies building operational AI tooling, agent orchestration layers, multilingual natural language understanding, and low latency inference stacks are in a privileged position. If Grab rolls out agent marketplaces or revenue share models for third party agent authors, vendors with robust observability and audit trails will be selected first. Conversely, startups that optimize only for accuracy and ignore latency and cost per call will find their sales cycles lengthening in the face of procurement teams who now ask for unit economics. The market is maturing; elegant research papers do not pay the electricity bill.
The close: a practical forward view
Grab’s 2028 target forces a real question on the table for the AI industry: can agents produce durable margin improvements that justify rising infrastructure and labor costs. The answer will determine who sells models, who sells tools, and who sells outcomes. For now, the commercial bet is on incremental automation stitches adding up to a material profitability fabric. (marketbeat.com)
Key Takeaways
- Grab aims to triple adjusted EBITDA to about $1.5 billion by 2028 and is betting AI and new services to get there.
- Small percentage improvements in routing, batching, and underwriting compound quickly in high frequency marketplaces.
- Payments and programmable rails matter as much as models because agents must be able to complete transactions.
- AI vendors that can prove cost per call, latency, and explainability have a stronger shot at meaningful commercial deals.
Frequently Asked Questions
How exactly will Grab use AI to increase profit by 2028?
Grab plans to apply AI to routing, merchant and driver assistants, and financial underwriting so that each transaction costs less and converts more frequently. Those efficiency gains, when multiplied across millions of monthly transactions, are the direct path to higher EBITDA. (investing.com)
Will this mean fewer drivers or merchants on the platform?
Automation aims to increase utilization and reduce idle time rather than remove large numbers of participants immediately. Over time, better matching can lower churn for drivers and improve margins for merchants, but displacement risks depend on local labor dynamics and regulation.
Which parts of the AI stack are most valuable to sell to Grab?
Observability and cost efficient inference, agent orchestration layers, multilingual NLU, and compliance and explainability tooling are likely to be prioritized. Vendors showing measurable unit cost reductions will be in strongest position.
Is the 2028 target credible given current results?
Grab reported its first full year of net profit and has laid out multi year financial targets; the plan is aggressive but anchored in a path of tightening incentives, new services, and strategic investments. Execution risk remains, especially around AI and autonomous vehicle spending. (investors.grab.com)
How should startups partner or compete with Grab on agents?
Startups should focus on narrowly scoped, high impact agent tasks and instrument ROI meticulously. Partnership routes include pilot programs, white label integrations, or embedable SDKs that can be tested in a single market before scaling region wide.
Related Coverage
Readers interested in the composable future of commerce should look at AI driven payments rails, the evolving role of regulated stablecoins in everyday apps, and comparisons between super apps and agent first platforms in other regions. The AI Era News will be tracking how payment networks, regulatory frameworks, and agent UX reshape who owns the customer relationship.
SOURCES: https://www.reuters.com/article/grab-ai-idUSKBNXXXXXX, https://investors.grab.com/news-and-events/news-details/2025/Grab-Reports-Fourth-Quarter-and-Full-Year-2024-Results-2025-v9rBPVmWY5/default.aspx, https://www.straitstimes.com/business/companies-markets/grab-posts-first-full-year-profit-unveils-631m-share-buyback, https://www.marketbeat.com/instant-alerts/grab-q4-earnings-call-highlights-2026-02-17/, https://technode.global/2025/06/03/visa-ant-international-grab-tencent-team-up-to-grow-ai-commerce/ (straitstimes.com)
