Three must-own stocks to play the second wave of AI

The initial phase of the AI boom heavily concentrated capital within a narrow hardware monopoly, primarily rewarding Nvidia as its graphics processing units captured a 90% market share in model training.

However, institutional focus is now rapidly pivoting toward the structural deployment of inference and agentic AI.

This second phase presents a broader investment landscape since the architectural moat surrounding inference hardware is less monopolized.

Consequently, capital is migrating toward specialized silicon, custom application-specific integrated circuits (ASICs), and advanced memory infrastructure, positioning three specific stocks to capture outsized returns.

AMD stock: an inference and agentic AI winner

Advanced Micro Devices is emerging as one of the clearest beneficiaries of the AI market’s pivot from model training to large‑scale inference and agentic workloads.

Its chiplet-based architecture provides a structural edge – enabling higher memory bandwidth and capacity – two constraints that increasingly define inference performance.

As data centers rebalance from an 8:1 GPU‑to‑CPU ratio toward a near‑parity model, AMD’s lead in server CPUs is becoming materially more valuable.

This dynamic was already visible in its latest quarter: Q1 sales jumped 38% to about US$10.3 billion (approx. $14.4 billion), powered by a 57% surge in data center sales.

With anchor commitments from OpenAI and Meta and the ZT Systems acquisition expanding its AI rack footprint, AMD is positioned as a foundational supplier for the 2nd wave of AI deployment.

Broadcom stock: the custom chip winner

As hyperscalers race to cut power consumption and rein in soaring AI capex, the industry is shifting decisively toward custom application‑specific chips – an area where Broadcom holds an unrivaled engineering lead.

AVGO’s long‑running partnership with Google on its Tensor Processing Units (TPUs) has already proven the commercial viability of bespoke AI silicon, and that momentum is now translating into massive direct orders, including Anthropic’s massive US$21 billion (approx. $29.3 billion) commitment for 2026.

The financial payoff is clear: Broadcom posted record Q1 revenue of US$19.3 billion (approx. $26.9 billion) (up 29%) as AI semiconductor sales more than doubled.

Management now sees ASIC revenue exceeding US$100 billion (approx. $139.4 billion) by fiscal 2027, underscoring AVGO’s central role in the second wave of AI infrastructure.

Micron stock: a memory chip winner

As AI workloads shift from compute‑heavy training to memory‑intensive inference, the industry’s bottleneck has moved squarely to DRAM and high‑bandwidth memory.

And that’s where Micron shines. Inference relies on rapid retrieval from key‑value caches – while agentic AI compounds memory requirements through longer, multi‑step reasoning chains.

With HBM demand rising in tandem with every new GPU deployment, MU is capturing a powerful secular tailwind.

That dynamic is already visible in numbers: Q2 revenue nearly tripled to US$23.9 billion (approx. $33.3 billion).

And as hyperscalers lock into long‑term supply deals, the sector’s historic cyclicality is fading.

Trading at roughly 13x forward earnings despite this explosive trajectory, Micron stock offers one of the most asymmetric valuations in the AI supply chain.