GPUs & AI Chips
GPUs and AI-specific accelerators are the core compute layer of the AI infrastructure buildout, with Nvidia, AMD, and Broadcom operating as leaders and Intel, Micron, and Tesla functioning as enablers across compute, memory, and custom silicon. Competitive dynamics span direct GPU and accelerator rivalry among Nvidia, AMD, Broadcom, and Intel, memory supply from Micron into GPU-based systems, and rising exposure to U.S. and Chinese export-control regimes across the leader cohort.
Active institutional exposure flipped from a net outflow of $14.4 billion in Q1 2026 to a net inflow of $29.0 billion in Q2 2026. 77.8% of companies in this sub-theme saw increased active positions in Q2 2026.
Overview
GPUs and AI chips comprise purpose-built graphics and tensor processors used for training and running AI models, spanning discrete GPU accelerators, custom SoCs, and the memory and networking silicon that support them. Nvidia, AMD, and Broadcom compete directly across GPU, accelerated-computing, SoC, and AI-networking product categories, while Intel competes in data-center CPU/GPU server products and supplies CPUs to a broad enterprise customer base including HPE, Oracle, Dell, and Cisco. Micron supplies memory components that underpin GPU-based systems, with disclosed supply relationships into Nvidia, Super Micro Computer, Arista Networks, and Penguin Solutions. Tesla develops its own SoC products internally, which its own filings identify as a direct competitor to Nvidia in chip design, even as Tesla separately sources analog and mixed-signal components from other semiconductor suppliers. FuriosaAI is classified as a leader in this sub-theme, though no verified relationship, activity, or constraint data is currently on record for the company.
Investment Thesis
Investment thesis details are part of the Pro plan.
Bottleneck Tracker
Bottleneck details are part of the Pro plan.
Key Companies
Related Events
A large fintech reports about a 10x inference cost cut by defaulting to self-hosted open-weights GLM instead of commercial coding services such as Claude Code and Codex. This is a data point against frontier-lab token pricing power, and may pressure Anthropic and OpenAI coding revenue if other enterprises follow. Nvidia benefits from the on-prem hardware demand, since 56 B200 GPUs were deployed and the serving optimizations ran on NVFP4. Z.ai gains credibility for Chinese open-weights models in regulated Western enterprises, though data-residency concerns are mitigated only by on-prem or regional hosting. The model-router pattern, with vision tasks sent to Gemini or GPT, suggests a hybrid enterprise architecture with frontier models used selectively.
Brookfield and its Radiant neocloud platform plan a UK AI campus at Drax Power Station that could exceed £20 billion and 1 GW of capacity. Drax shares rose about 8.8% as investors priced in monetization of its grid-connected site. Phase 1 (up to 100 MW from Q1 2028, £2.5 billion) still needs planning, regulatory and network approvals, and a tenant, and no final investment decision has been made. The project tests whether the UK can host large-scale AI compute despite high power costs and grid connection delays, which pushed Microsoft out of the Nscale deal in October. If delivered, it would add sovereign-style UK compute capacity and demand for power, cooling, chips and construction, though execution risk remains high.
The reported purchase is a major external validation of Google's TPUs as a merchant alternative to Nvidia GPUs, since OpenAI is one of the largest AI compute buyers. It strengthens Alphabet's case for returns on AI infrastructure spending, with TPU revenue expected mostly in 2027. Nvidia keeps its dominant position, but the move shows its largest customers diversifying across Google TPUs and custom silicon. Broadcom stands to gain from both the TPU and OpenAI custom-chip programmes. The undisclosed size of the order limits near-term estimate changes, and investor scrutiny of roughly $600 billion in 2026 hyperscaler capex remains.
Anthropic's filing shows a compute obligation base of about $895 billion, and $437 billion of that sits in non-cancellable deals with Microsoft, Google, Amazon and Broadcom. Revenue growth is slowing (46% in Q3 versus roughly 2.5x in Q2) while compute costs grew 77%, which raises concern about whether revenue can cover the commitments. Suppliers and hyperscalers gain long-dated contracted demand visibility, but they also take on counterparty concentration risk if Anthropic's growth stalls. About 44% of Anthropic's revenue flows through cloud marketplaces, which ties its fortunes to partners that also compete with it. The IPO could set the valuation and financing benchmark for frontier AI labs, and investors will watch whether capital markets can fund spending on this scale.
SK Hynix's warning that the memory shortage will persist through 2030, alongside record profit, signals prolonged pricing power for HBM and DRAM suppliers. Long-term agreements covering more than half of 2027-2028 capacity give the company unusual earnings visibility. Electronics makers such as smartphone, PC and console vendors face continued margin pressure and price hikes. Korean memory shares rallied, with SK Hynix up 10.9%, Samsung up 9% and the KOSPI up 6.6%. The upside is tempered by investor fears of an AI spending slowdown: SK Hynix remains about 50% below its mid-year peak, it postponed the Solidigm IPO, and the Bank of Korea flagged up to $100 billion of export downside.