What's the read on semiconductors and the AI compute build-out, and how is positioning shifting?

The AI compute build-out is driving a historic capital investment cycle, projected to be an order of magnitude larger than the cloud era with an estimated **$10 trillion in total expenditure** [8, 24]. This has fueled a supercycle where the semiconductor industry's revenue run rate is projected to hit $1 trillion annually by 2026, four years ahead of previous forecasts [8, 24]. Demand from hyperscalers and enterprises is insatiable, with the top six cloud providers alone projected to spend over $1 trillion on GPUs by 2027 . This explosive demand is creating severe and persistent supply chain bottlenecks that are expected to last for years, if not a decade [3, 10, 12, 26]. Initially centered on chip fabrication, the primary constraints are now shifting to high-bandwidth memory [11, 14, 17] and physical infrastructure, particularly the availability of power and data center capacity, which is throttling even the largest players [7, 22].

As the market matures, the nature of AI workloads is shifting from compute-intensive training to the recurring revenue streams of inference, reasoning, and agentic AI [2, 9, 10]. This evolution is causing a fundamental change in data center architecture and broadening the demand for hardware beyond GPUs [1, 25]. The rise of agentic AI, which requires CPUs to "harness" the GPU "brain," is creating a renaissance for CPU manufacturers [12, 23]. This is projected to shift the typical CPU-to-GPU ratio in servers from 1-to-8 toward **1-to-1**, creating a massive new growth vector for companies like Intel and AMD . Furthermore, the emergence of "reasoning" models, which can be up to 50 times more compute-intensive per query, ensures a long-term, escalating demand profile for advanced hardware . This shift indicates that long-term, defensible value in the AI stack will accrue primarily at the semiconductor and infrastructure layers, a reversal from the software-centric cloud era [19, 24].

NVIDIA maintains a dominant position, controlling over **90% of the market** and running an estimated 98% of non-Google AI workloads, protected by a powerful moat of integrated software, hardware, and networking . The company is aggressively defending its share by accelerating its product roadmap to an annual cadence and strategically adjusting margins . However, the sheer scale of demand is creating opportunities for a wider ecosystem [1, 4]. The acute shortage of NVIDIA GPUs has forced customers to seek alternatives from AMD and Amazon . Intel is experiencing a resurgence, buoyed by CPU demand and growing confidence in its foundry business, which was bolstered by a strategic $5 billion investment from NVIDIA [5, 6]. While the market is expanding, there is tension regarding the potential for new competition; some analysts see the shift to inference as an opening for specialized chips , while others contend it is **too late for new entrants** to challenge incumbents in core chip architecture .

The AI build-out is increasingly intertwined with geopolitical strategy and systemic risk [4, 8]. Semiconductors are now treated as critical national assets, leading to industrial policies aimed at reshoring manufacturing and intensifying the US-China tech war through sanctions and efforts by firms like Huawei to develop a domestic supply chain [6, 8, 10]. The global supply chain remains highly concentrated, with firms like ASML, TSMC, and NVIDIA representing critical points of failure . This intense investment has inflated the semiconductor sector to **17% of the S&P 500** , raising concerns that a future correction, potentially triggered by a slowdown in enterprise AI adoption or a failure to realize ROI, could have a broader macroeconomic impact than the dot-com bust [4, 24].