El superciclo de semiconductores de IA: los ganadores en la construcción de una infraestructura de billones de dólares

The global AI semiconductor industry is entering a transformative phase, driven by a confluence of surging demand for high-performance computing, regulatory reshaping, and capital-intensive infrastructure expansion. As hyperscalers commit to multi-hundred-billion-dollar AI buildouts by 2026, the semiconductor value chain is witnessing a supercycle of investment and innovation. At the forefront are NVIDIA, Broadcom, and TSMC, whose strategic positions in chip design, networking, and manufacturing are cementing their dominance. Meanwhile, emerging challengers like AMD and Marvell are carving niche roles, though their long-term trajectories remain contingent on navigating regulatory headwinds and market concentration risks.

NVIDIA: The Uncontested AI Compute Leader

NVIDIA's dominance in the AI semiconductor market is unparalleled. In Q3 2025, the company generated $57.0 billion in revenue, maintaining an 80% share of the AI GPU market. This leadership is underpinned by its $100 billion investment in OpenAI for a 10-gigawatt data center buildout and a $10 billion commitment to Anthropic, leveraging its Vera Rubin chip architecture. These bets align with the projected 40% compound annual growth rate in AI chip revenue through 2028, positioning NVIDIA as the de facto standard for AI training and inference.

NVIDIA's competitive edge is further reinforced by its exclusive access to TSMC's 3nm production capacity, securing 30% of the foundry's advanced node output for its Rubin chips. This vertical integration ensures supply stability amid global chip shortages and geopolitical tensions. However, regulatory scrutiny looms as competitors like Broadcom gain traction, prompting antitrust concerns over market concentration.

Broadcom: The Networking and Custom Silicon Challenger

Broadcom is emerging as a formidable rival, leveraging its strengths in custom ASICs and networking solutions to capture a 50% market share in AI semiconductors. The company's Tomahawk 6 and Tomahawk Ultra processors, which double the performance of prior generations and enable rack-scale AI systems, are critical to hyperscalers' infrastructure upgrades. Revenue growth of 16% quarter-over-quarter in 2025, reaching $17.4 billion, underscores its ability to monetize this demand.

Broadcom's acquisition of VMware in 2025 has also strengthened its software infrastructure, enabling seamless integration with AI/ML workloads. With the AI infrastructure market projected to reach $60–90 billion by 2027, Broadcom's dual focus on hardware and software positions it to challenge NVIDIA's hegemony. Yet, its reliance on TSMC for manufacturing exposes it to similar supply risks, though its diversified customer base mitigates some of these pressures.

TSMC: The Foundational Enabler of the AI Era

As the world's largest chip foundry, TSMC is the linchpin of the AI semiconductor supercycle. Its 2025 capital expenditures of $38–42 billion are directed toward advanced fabrication processes, including 3nm and 2nm nodes, to meet demand for AI accelerators from NVIDIA, AMD, and others. TSMC's geographic expansion-new foundries in Arizona, Germany, and Japan-not only diversifies its risk profile but also aligns with U.S. and EU policies prioritizing domestic semiconductor production.

The company's pricing power remains robust, with hyperscalers projected to spend $500 billion on AI infrastructure by 2026. However, rising operational costs from capacity expansions and geopolitical tensions could pressure margins. For investors, TSMC's role as the sole enabler of next-generation AI chips ensures its long-term relevance, even as end-market competition intensifies.

AMD and Marvell: Niche Players in a Crowded Field

AMD and Marvell represent alternative investment avenues but face structural challenges. AMD's 22% share of the data center AI training market, driven by its Instinct MI350 and MI400 series, positions it as a credible NVIDIA alternative. Its 6-gigawatt partnership with OpenAI and Oracle further diversifies its client base. However, AMD's reliance on TSMC for manufacturing and its smaller R&D budget compared to NVIDIA limit its scalability.

Marvell, meanwhile, has surged on 78% year-over-year data center revenue growth, bolstered by its $3.25 billion acquisition of Celestial AI reported by market analysts. The company's focus on electro-optical interconnects and custom ASICs aligns with AI infrastructure needs, but its exposure to regulatory export controls poses risks. While Marvell's AI revenue is projected to double from $1.8–2.0 billion in FY2025, it remains a long shot to displace the industry's titans.

Regulatory Dynamics: A Double-Edged Sword

The AI semiconductor landscape is increasingly shaped by regulatory interventions. The Trump administration's rescission of the Biden-era AI Diffusion Rule has relaxed export controls on advanced chips but introduced new compliance burdens, such as location verification features on AI chips. The GAIN AI Act's domestic-first export restrictions could further constrain AMD and Marvell's global sales, prioritizing U.S. demand over international markets.

For NVIDIA and Broadcom, these policies create both opportunities and risks. While they benefit from U.S. government support for domestic production, they also face antitrust scrutiny as market concentration intensifies according to industry analysts. TSMC, meanwhile, navigates a delicate balance between complying with U.S. and Chinese demands, with its geographic diversification offering a buffer against geopolitical shocks.

Investment Thesis: Strategic Entry Points in the AI Supercycle

The AI semiconductor supercycle presents a rare confluence of secular growth and structural tailwinds. NVIDIA's leadership in compute, Broadcom's innovation in networking, and TSMC's foundational role in manufacturing form a trinity of long-term value creation. Investors should prioritize these three as core holdings, given their entrenched positions and ability to scale with demand.

AMD and Marvell, while promising, are better suited as satellite positions. Their success hinges on executing strategic partnerships and navigating regulatory complexities. For risk-averse investors, TSMC's stable cash flows and critical role in the AI supply chain offer a more predictable path.

As the industry races toward a $1 trillion market by 2030 according to industry forecasts, the winners will be those who can balance innovation with regulatory agility. The AI semiconductor supercycle is not just a technological revolution-it is a redefinition of global economic power.