The shape of AI in 2026

Ensuring AI models can continue to deliver will require an overhaul of infrastructure.

"AI success will not be determined simply by who has the biggest model or the most data, but who has the most unified, governed, and accessible data. APAC organisations are moving from passive data storage to active data management, and this means that they will need to reassess their data infrastructure strategy.

"Intelligent data infrastructure will underpin this shift by automating the curation, vectorisation, and access to data needed to train, tune, and operationalise AI at scale," suggested Elaine Chan, Director and APAC Head of AI Sales & GTM, NetApp.

"Most AI initiatives fail not because of weak models, but weak data foundations. The next generation of AI success stories will solve for the data first." 

It is no longer about size, agreed Mark Russinovich, CTO, Deputy CISO and Technical fellow for Microsoft Azure. Instead of building more and bigger data centres, the next wave is about making every ounce of computing power count.

“The most effective AI infrastructure will pack computing power more densely across distributed networks,” Russinovich said, suggesting that 2026 will see a new generation of linked AI 'superfactories' — that will drive down costs and improve efficiency.

AI will be “measured by the quality of intelligence it produces, not just its sheer size,” Russinovich said.

Russinovich also talked about orchestration of AI workloads that ensures nothing sits idle. If one job slows, another moves in instantly — ensuring every cycle and watt is put to work. This shift will translate into smarter, more sustainable and more adaptable infrastructure to power AI innovations on a global scale, Russinovich said.

Gartner's predictions for 2026 include trends towards AI supercomputing platforms, defined as platforms that combine powerful processors, massive memory, specialised hardware, and orchestration software to tackle data-intensive workloads in areas like machine learning, simulation, and analytics.

According to Gartner, AI supercomputing platforms are all-in-one devices that integrate CPUs, GPUs, AI ASICs, neuromorphic and alternative computing paradigms, enabling organisations to orchestrate complex workloads while unlocking new levels of achievement. 

"These systems combine powerful processors, massive memory, specialised hardware, and orchestration software to tackle data-intensive workloads in areas like machine learning, simulation, and analytics," the consultancy said.

By 2028, Gartner predicts that over 40% of leading enterprises will have adopted hybrid computing paradigm architectures into critical business workflows, up from the current 8%.

“This capability is already driving innovation across a diverse range of industries,” said Tori Paulman, VP Analyst at Gartner. 

“For example, in healthcare and biotech, companies are modelling new drugs in weeks instead of years. In financial services, organisations are simulating global markets to reduce portfolio risk, while utility providers are modelling extreme weather to optimise grid performance.”

Hybrid architectures 

Source: Lenovo. Bhatia.

Sumir Bhatia, President, Asia Pacific, Infrastructure Solutions Group, Lenovo, also touched on hybrid AI infrastructure for 2026. "Enterprises are now shifting from oversizing centralised environments to ‘right‑sized’ hybrid architectures and as‑a‑service models that allow them to scale up or down with demand. This marks the steady decline of monolithic, one‑size‑fits‑all IT strategies in 2026," said Bhatia. 

"Across Asia, customers are converging on a similar architectural answer: hybrid AI. Some workloads belong in the public cloud, majority at the edge or on‑prem/data centre, often in the same workflow. This distributed model is driven by data sovereignty, latency needs, cost predictability, and, increasingly, sustainability.  

"At Lenovo, we strongly believe that hybrid AI will be the default architecture for enterprises that want flexibility across diverse Asia-Pacific (APAC) markets." 

Private AI

Source: Cloudera.
Lim.

"As global regulations tighten and data sovereignty concerns grow, private AI will rise in importance, said Remus Lim, Senior VP, Asia Pacific & Japan, Cloudera. Private AI refers to AI that operates in a controlled environment.

"This is especially critical as cybersecurity continues to be a top enterprise priority. Microsoft’s Digital Defense Report 2025 revealed a 32% surge in identity-based attacks in the first half of the year, showing the increasing use of AI to craft highly convincing social engineering lures. As threat actors adopt AI, enterprises must match that sophistication with AI-driven defence.

"Private AI frameworks will play a pivotal role here, enabling organisations to deploy models in controlled environments, detect anomalies faster, and minimise exposure to public cloud vulnerabilities. The enterprises that invest in secure, compliant AI now will be the ones to innovate with confidence later."

Lim expects highly-regulated industries such as financial services, healthcare, and the public sector to accelerate adoption of private AI architectures as they can then harness the power of generative and agentic AI without exposing sensitive data.  

Sovereign clouds 

Martin Creighan, VP, Asia Pacific at Commvault, called cloud sovereignty "the new strategic frontier". "Sovereignty is about control and choice. In a multicloud, multiregion world, enterprises need the freedom to decide where data resides – on-premises, in a private cloud, a local hyperscaler region, or a global cloud – while still maintaining visibility into under whose laws it sits, and how it can be recovered without crossing borders," he pointed out.

"Architectures are becoming sovereignty-aware by default, with encryption, access policies, and compliance rules moving with the data - across borders and clouds. When sovereignty is built into design, compliance becomes a competitive advantage. In 2026, this combination of sovereignty and freedom of choice will allow organisations to innovate confidently within trusted boundaries."

Source: Cohesity.
Poonen.

Sanjay Poonen, Cohesity's CEO, said AI initiatives will fuel sovereign cloud growth. Data in sovereign clouds must be stored in specific geographical locations. 

"As AI initiatives mature, enterprises are turning to sovereign cloud environments to balance innovation with control. We’ve heard from leaders across the globe, in markets such as the EU, the Middle East, and APAC, that sovereign clouds are a significant push for organisations looking to protect their data," he said.

"I anticipate that in 2026, we will see a marked increase in enterprise AI workloads driven by sovereign clouds. As regulators intensify scrutiny of AI data workflows, IT leaders will continue to deepen partnerships with hyperscalers to maintain control of data while generating AI-powered insights."

Geopatriation, moving company data and applications out of global public clouds into local options such as sovereign clouds, regional cloud providers, or an organisation’s own data centres will increase due to perceived geopolitical risk, Gartner said. 

“Shifting workloads to providers with an increased sovereignty posture can help CIOs gain more control over data residency, compliance and governance,” said Gene Alvarez, Distinguished VP Analyst at Gartner.

“This greater control may improve alignment with local regulations and build trust with customers who are concerned about data privacy or national interests.” 

Hybrid architecture 

Brian Spanswick, CIO, Cohesity, noted that annual IT budgets are growing at 2–3% on average. This has led to technology leaders facing mounting pressure to deliver greater value without significant increases in spending, he said. 

Source: Cohesity.
Spanswick.

"This dynamic is accelerating the shift toward hybrid cloud strategies that optimise both cost and capability. In 2026, I expect enterprises to allocate more than half of their infrastructure investments to hybrid models—designed to balance data sovereignty requirements with the agility and scalability of cloud services.

"By embracing hybrid architectures, IT leaders can extend the life and value of existing assets while enabling innovation at scale, all without overextending financial commitments. This approach is becoming critical as organisations demand measurable outcomes, operational resilience, and flexibility from every technology investment."  

Sovereign edge

The sovereign edge will continue to evolve, said Rajiv Ramaswami, President and CEO of Nutanix. "AI is a force for more distributed infrastructure as AI moves out to process data generated at the edge.  Enterprises will need to consider the global management, distributed security, and remote recovery/destruction policies available for the sovereign edge and rely more on platform engineering to successfully achieve this.  

"As AI continues to skyrocket in adoption, businesses will look to find ways to process AI-related data locally. As a result, organisations will look to global management solutions with integrated security and edge resiliency to help keep this in check."   

AI or no AI, cloud vulnerabilities mean an edge strategy is critical for businesses today, said Gopi Duddi, CTO, Couchbase. "Edge devices will grow more powerful and begin to take on meaningful AI processing rather than sending everything to a central service. Data will be created and consumed at the edge which reduces dependence on cloud availability," he said. 

"When networks fail, businesses without an edge strategy will stall - as seen when retail locations like Starbucks have had to close because their cloud-based checkout systems went down - while resilient systems continue to operate. This matters for everyday services and retail as much as it does for emergency services where downtime can result in life or safety consequences. Organisations with strong edge designs will have a clear advantage in reliability and uptime."  

Security implications

“Issues around organisations and government agencies being supplied and serviced from unfavourable locations are continuing to hit the headlines. It is not just about data sovereignty, it is about understanding where the last mile of your service is provided and answering: who built the technology? Where are they? Who’s supporting my applications? Scrutiny over who is using what and where it is coming from is only going to increase,” said Jon Scolamiero, Americas Field CISO, Mendix.

Dmitry Volkov, CEO, Group-IB, described the paradoxical lack of global visibility in cyberdefence as a result of data sovereignty. "Criminals continue attacking without borders while defenders limit detection due to regional visibility constraints," Volkov said.

"Attackers operate globally, but data localisation trends are inadvertently creating intelligence blind spots for defenders. As regions embrace data sovereignty and localisation measures, the regionalisation of data is slowing down global collaboration and threat intelligence sharing.  

"To succeed at protecting critical infrastructure, businesses and regions should prioritise frameworks that enable threat indicator sharing over raw data movement."  

Gartner has a related prediction: By 2027, 35% of countries will be locked into region-specific AI platforms using proprietary contextual data. "Sovereign AI is no longer a concept — it’s a competitive lever. The lines between governments and vendors are blurring, and the implications stretch far beyond tech. Once locked in, getting out won’t be easy," Daryl Plummer, VP, Distinguished Analyst & Gartner Fellow, said in a blog post.

Platform wars

"We're entering a decade of platform wars, where success won't hinge on individual features but on the strength and flexibility of the entire platform. Enterprises will increasingly seek platformisation focused on solutions that provide freedom of choice while delivering on three core priorities: resiliency, hardware modernisation, and software modernisation," added Ramaswami.

"The fastest path to innovation will come from platforms that embrace openness: choice of containers, choice of LLMs, and choice of GPUs. Vendors that can integrate these options seamlessly and rapidly will be the ones that win in this new era. And customers that pick the most complete platforms can integrate management across traditional and new applications deployed across hybrid cloud locations, including the sovereign edge." 

Adaptability

"By 2026, AI leadership in APJ will be defined by an organisation’s ability to reconfigure its AI
infrastructure models, inference engines, and deployment platforms regularly," said Nathan Hall, GM and VP of Asia Pacific and Japan (APJ), Pure Storage. 

"Companies that build modular, provider agnostic AI stacks will outpace those that bet on large fixed infrastructure or single vendor architectures. As foundation model capabilities leapfrog every few months and cost/performance curves shift unpredictably, enterprises will no longer gain advantage by scaling static AI systems," Hall explained. 

"Instead, the winners will implement infrastructure that allows them to: 

● Swap models and inference providers within weeks 

● Orchestrate workloads across cloud, on-prem, and edge based on cost and capability 

● Absorb rapid changes in GPU hardware, quantisation formats, and model architectures 

● Redeploy AI copilots or agents to production multiple times per quarter 

"In 2026, competitive advantage won’t come from how large your AI infrastructure is, but how quickly you can adapt it." 

The AI divide 

Steve Yen, Co-Founder, Couchbase, said that the gap between GPU-rich and GPU-poor companies will define the next phase of AI. "The GPU-rich players are the hyperscalers and model labs that can afford data centres full of (NVIDIA's) H100s and Blackwells. Everyone else is left figuring out what they can run without that level of compute. If you don’t have the GPUs, there are entire areas of AI you simply can’t touch right now," he explained.

"But this divide won’t last. If the AI bubble cools, all that GPU capacity and all those new data centres don’t vanish. They become available to the broader market at far more reasonable prices. That shift creates space for teams that aren’t training giant models but still want to build meaningful AI features on top of the expanding infrastructure stack. 

"It also puts more attention on platforms that can store, sync and reshape the large volumes of data AI produces without consuming half the budget on compute."

Yen added that businesses should absolutely prepare for the popping of the AI bubble. "Whether the AI bubble bursts in 2026 or 2027, forward-looking teams can prepare now for the opportunities that follow. As technologists, we’ve seen this pattern before: once the hype cools, infrastructure becomes dramatically more accessible," he said. 

"Smart organisations will treat this as a moment to plan, so they’re ready to capitalise on the windfall on the other side of the bubble." 

"At Couchbase, we expect this GPU surplus to accelerate what our database can deliver. We see a future of incredibly fast, massively parallel processing for operational workloads, differentiated data analysis capabilities, highly scalable vector indexing and search, and new ways to serve AI-powered applications at global scale. 

"With GPUs, memory, storage and networking becoming cheaper and more widespread, the price-performance curve shifts in favour of builders. And we couldn’t be more excited about what that unlocks," Yen said. 

Demand and supply was also Nigel Green's focus. The CEO of the deVere Group commented: “The pace of AI infrastructure expansion is colliding with physical limits in global supply chains, creating shortages of critical components and forcing markets to confront a new reality.

“This imbalance is already pushing up prices, concentrating market gains, and raising the risk of volatility. In 2026, this dynamic is likely to move from background noise to centre stage for investors.”

Green observed that the buildout of data centres, cloud capacity, and AI computing power is driving intense demand for advanced chips, high-bandwidth memory, networking hardware, and power systems. “Supplies of several of these components are tight, with prices for advanced memory and specialised silicon rising sharply as manufacturers prioritise large AI customers," he said.

“This is where the AI story turns hard-edged. It becomes about bottlenecks, allocation, and who controls supply. Those bottlenecks are already altering behaviour across markets. Hyperscalers and enterprise buyers are locking in long-term supply agreements and committing capital well in advance, effectively reserving capacity for themselves. This enhances revenue visibility for some suppliers but leaves others exposed.”

Green also pointed out that consumer electronics producers rely on many of the same components now being diverted into data centres. With higher input costs and constrained supply, prices are likely to rise for consumers as well. “In 2026, this divergence is likely to be more pronounced as AI-related demand scales further and supply constraints persist," he said.

Hashtag: #2026Predictions 

*An ASIC is an application-specific integrated circuit, or specialised chip. CPU stands for central processing unit, the main chip that runs devices, while GPU refers to the graphics processing unit, which originally complemented the CPU in handling the mathematical calculations for displaying graphics, and is now also used to manage the calculations required for AI. LLM stands for large language model.