Now reading:

It’s still early in the AI revolution

It’s still early in the AI revolution

  • June 2026
  • By OCBC
  • 10 mins

We believe AI-related companies as a group will continue to deliver earnings growth ahead. AI adoption is still at an early stage with a meaningful runway ahead.

Eli Lee
Managing Director,
Chief Investment Strategist,
Bank of Singapore


Semiconductor: Outperformance underpinned by a broadening of AI trade beyond GPU to AI inference beneficiaries

Semiconductors (Semis) have driven technology stocks to an all-time high in recent months, supported by chip supply chain constraints and strong demand amid upward revisions in hyperscalers’ capex to new highs. Initial fears that the US-Iran war could disrupt semi production also did not materialise.

Notably, the AI trade is broadening beyond GPUs to inference-specific ASICs, CPUs, memory, storage and networking with the rise of agentic AI. Our analysis shows that GPUs were previously the clear winner and related stocks more than doubled in about half a year following the release of ChatGPT on 30 November 2022; in a similar period after Claude Cowork was released on 12 January 2026 - arguably a turning point for agentic AI - CPU, storage and memory sub-segments saw related share prices at least doubling while GPU-related shares grew less aggressively. Though industry GPU demand would likely remain resilient, each model call now requires more coordination, memory and system-level compute. The CPU-to-GPU ratio could even increase from 1:4–8 to as much as 1:1 according to AMD, creating a seismic shift in the semis landscape.

Internet: Easing concerns over ability to monetise capex, though profit growth would have to catch up with semis for re-rating

1Q2026 earnings for major internet players broadly point to accelerating cloud revenue along with backlog growth, suggesting compute demand continues to outweigh existing data centre capacity. The healthy end-user take-up rate is further corroborated by the surge in native AIs’ sales, led by Anthropic whose annualised revenue had leaped from less than US$2bn at the start of 2025 to US$30bn in April 2026. Moreover, the internet players are actively monetising their AI investments through product expansion, moving beyond services to include chip sales. These should ease earlier concerns around their ability to monetise the massive amount of AI capex. Nevertheless, internet players’ profit growth has lagged that of semis, and a subsequent revenue catch‑up as AI applications are rolled out could serve as a potential re‑rating catalyst.

Software: Be selective amid AI disruption risks

While the software segment’s near-term earnings have been resilient, valuation multiples have declined significantly from the 2025 peak on concerns about vulnerable terminal values, as AI increases competitive pressures and erodes software moats. For selective vulnerable software companies, it is possible that such market behaviour is similar to what happened to the newspaper industry when it was being disrupted by the internet — share prices started falling in early 2000s, preceding a structural decline in earnings several years later. The pressure is unlikely to abate anytime soon in our opinion, as plans by AI natives to IPO could lead to an accelerated rollout of commercial applications to demonstrate monetisation of their large language model (LLM) investments.

We are therefore selective in software and investors should be prepared to ride through a potentially protracted period of weakness and volatility for selective vulnerable sub-segments. We think cybersecurity stands out for its moat defensiveness because AI increases security risks and regulatory requirements, thereby providing structural tailwinds for the industry. Larger security firms should benefit disproportionately, as cybersecurity clients have been consolidating spending to reduce the number of vendors. Meanwhile, native AI companies could play a complementary role instead of competing directly with existing cybersecurity players. This is illustrated by Project Glasswing, under which technology partners will get early access to Anthropic’s Mythos, an AI model that excels at identifying and exploiting software vulnerabilities.

Investors’ Playbook: Diversify into inference for the next leg of AI, stay disciplined on entry points

Overall, we think the broadening of the AI trade to AI inference beneficiaries (i.e. ASIC, CPU, memory, storage, networking) is a positive development, as it allows participation in AI growth with diversification across more sub-segments and names. That said, investors should also take valuations into account in assessing the risk-reward offered by AI inference beneficiaries as the recent strong momentum could suggest stretched positioning. Semiconductor foundries should remain beneficiaries as long as long-term compute demand continues to rise. Major internet players also remain attractive, given their lagging share price performance relative to semi-conductor players and their growth potential from both external and internal AI application rollouts. Lastly, we would be selective in software, preferring large cybersecurity players that can benefit from both the AI tailwind and vendor consolidation trend.