As enterprises across Asia-Pacific accelerate AI adoption, the conversation is moving beyond experimentation to structural transformation. While software-as-a-service (SaaS) has defined enterprise IT consumption for more than a decade, industry leaders now argue that a new paradigm is taking shape, one centred on intelligence, autonomy and real-time decision-making.
In conversation with iTNews Asia, Kenneth Lai, Vice President, ASEAN at Cloudflare, shared insights on how AI-as-a-Service is emerging not merely as an extension of SaaS, but as a fundamentally new operating model for enterprises.
“AI-as-a-Service is one of several operating models that will coexist within organisations as they adopt AI in different ways,” Lai said.
“We have seen it streamline operations, take on administrative workloads, and automate routine processes, freeing up teams to focus on higher-value work. This represents another industry shift in how organisations are going to consume software moving forward,” he added.
From applications to outcomes
According to Lai, traditional SaaS models are likely to decline in dominance as AI agents increasingly become the primary interface for enterprise workflows.
“Traditional SaaS models only make sense in scenarios where humans remain the primary operators of software,” he said. As organisations expect AI to execute tasks, make decisions and operate in real time on their behalf, spending shifts from paying for application access to paying for “intelligence, outcomes and insights.”
This evolution, he noted, is driven by the convergence of changing user expectations, economics and infrastructure capabilities that now support autonomous operations at scale.
Network becomes mission-critical infrastructure
As AI agents embed deeper into workflows, the underlying Internet architecture takes on greater strategic importance.
“With this shift, the robustness of the Internet's underlying architecture matters more than ever. Building a better Internet now means embedding compute, security, and trust directly into the network so AI can operate safely and efficiently everywhere,” Lai said.
AI workloads demand real-time responsiveness and consistent policy enforcement across clouds and geographies. Lai pointed to trust and latency as two bottlenecks enterprises often underestimate.
AI applications are only as effective as their responsiveness. High latency quickly erodes the value of real time inference, agents and decision making.
- Kenneth Lai, Vice President, ASEAN at Cloudflare
He cautioned that organisations must safeguard data privacy as information moves between regions, clouds and models, placing new pressure on globally distributed networks to reduce latency while maintaining consistent security controls.
A larger cybersecurity shift than SaaS
The transition to AI-as-a-Service also introduces new cybersecurity challenges. Instead of securing a fixed set of applications and endpoints, enterprises must now protect autonomous agents operating at machine speed.
“Security must be embedded directly into the architecture, enabling systems to evaluate intent, adapt to changing behaviours and respond at machine speed rather than relying on static rules or manual intervention,” Lai said.
He added that as organisations adopt third-party and open AI models, governance gaps may introduce systemic risks. Enterprises should implement guardrails including access controls, clear data boundaries and auditability even from day one.
Security, in this new paradigm, must become “invisible” and integrated into the network fabric itself, particularly as verifying every device or AI driven system individually becomes impractical
Will AI agents replace SaaS licensing?
As AI agents demonstrate value, Lai expects enterprises to expand their deployment and align budgets accordingly. Over time, he anticipates cloud spending could shift toward AI agent orchestration, inference and network-delivered intelligence rather than traditional SaaS licensing.
Vendor selection, however, will depend on context, he added. In regulated industries such as healthcare, finance and critical infrastructure, trust, accuracy and auditability will dominate decision making. In consumer-facing or internal productivity use cases, latency and cost may outweigh perfect accuracy.
Long term, enterprises are unlikely to choose on a single dimension. Instead, the balance between trust, responsiveness, cost efficiency and ecosystem openness will shape procurement strategies.
Productivity gains or complexity shift
While AI-as-a-Service is gaining momentum, Lai described the industry as still being in an exploratory phase.
“We’re only at the beginning of AI-as-a-Service. Over the next five years, the focus will rapidly shift from initial experimentation to a phase of rigorous trial and integration, with the objective of translating these early efforts into scalable, real-world impact across every enterprise workflow,” Lai said.
Whether AI-as-a-Service ultimately delivers productivity gains beyond what SaaS achieved will depend on how effectively enterprises embed intelligence, security and network resilience into their core architecture, he added.
