AI Pulls the Industry into the Trillion-Dollar Era

We remember—well it was not that long ago—when we were holding our breath for the moment the global semiconductor industry crossed the line into $600 billion in annual sales. That moment has come and gone.  We barely got out of bed to note its recent passage into $700 billion territory. The next stop is a $1 trillion a year.

As always in this industry, the incremental growth in sales and revenues far underestimates the rising power of the technology itself. When we pass into $1 trillion in revenues, it will be because artificial intelligence has triggered an epochal transformation of the economy. If $600 billion in global semiconductors sales supported a $100 trillion global economy, expect that number to double again as semiconductors claim their first $1 trillion in annual sales.

As for the industry itself, expect the entire silicon ecosystem to boom as well. We don’t see the sector as overpriced but held back by a failure to fully imagine the future we are promised.

The entire industry is being reshaped by AI-led capability and demand—from hyper scalar capital expenditures to foundry product mix to advanced packaging, memory pricing, and the power analog layer that feeds next-generation data centers.

Expect at least 45% of the total semiconductor market to be accounted for by high performance compute and artificial intelligence by 2030, as advances in packaging density and energy per performance accelerate. TSMC (NYSE: TSM), through which more than half the world’s semiconductors now pass, expects 2025 revenue growth of around 30%. It expects a CAGR In the mid-40s for AI accelerator revenues through 2028.

For now, while we await the utter disruption of the industry by already accelerating wafer scale technology, advanced packaging has become the new center of gravity. CoWoS (Chip-on-Wafer-on-Substrate) capacity is the gating factor for top-tier AI accelerators and HBM (high bandwidth memory) systems. Our analysis indicates monthly CoWoS output will reach roughly 75,000 wafers per month by the end of 2025, with another significant lift in 2026. This is not an incremental tweak—it is the structural loosening of a bottleneck that has defined high-end AI chip availability for the past two years.

TSMC is pushing the packaging envelope aggressively. CoWoS-L will move to a 9.5× reticle-size “super-interposer” in 2027 with SoIC die-to-die stacking and support for up to a dozen HBM stacks, while a 5.5× reticle-size step arrives in 2026. In parallel, a System-on-Wafer (SoW) platform is in development with an unprecedented reticle field of roughly 40×, designed for ultra-large chip-on-wafer assemblies. The direction is unmistakable: larger substrates, more HBM channels, and denser 2.5D/3D integration to serve ever-larger model sizes and context windows.

Table 1 shows the global semiconductor market accelerating into mid-2025. Q2 2025 sales reached $179.7 billion, a 7.8% increase over Q1 2025. June alone posted a year-on-year gain of 19.6%. First-half billings of $346 billion mark a near 19% rise versus the first half of 2024. The upward revision in the 2025 forecast to $728 billion and the 2026 forecast to $800 billion reflect confidence in intensifying AI and HPC (high performance compute)-driven demand.

Table 2 shows hyperscaler capital expenditure plans for 2025 have expanded sharply, implying a sustained multi-year demand for semiconductor content to equip the centers. Alphabet increased its forecast by $10 billion to $85 billion, driven by AI infrastructure expansion. Meta’s $66–72 billion range suggests another year of similarly significant spending in 2026. Amazon’s trajectory toward approximately $118 billion underscores the intensity of the current investment cycle.

Advances in both volume and complexity are driving growth across industry. Table 3 shows high and rising foundry utilization rates. These gains reflect in part renewed demand in analog, power management integrated circuits, and AI-related compute, with the most resilient pull coming from advanced processes feeding HPC and AI workloads.

As Table 4 suggests, in the memory market momentum is driven by HBM (high bandwidth memory) demand and a recovering flash memory market. Micron’s HBM revenue rose by roughly 50% sequentially in its fiscal third quarter, while SK hynix delivered record revenue and profit with expanded capex for HBM production. HBM demand is driven by AI.

Table 5 suggests the analog and power sectors are transitioning from stabilization to infrastructure-led growth, once again thanks to AI. Texas Instruments (NASDAQ: TXN) and Monolithic Power Systems ( NASDAQ: MPWR) both reported sequential revenue growth, with Monolithic’s strength tied directly to AI/server power management. NVIDIA’s (NASDAQ: NVDA) 800 volt DC datacenter architecture, supported by Innoscience’s high-voltage GaN solutions, indicates a multi-year expansion path for power semiconductors as next-generation AI data centers require higher-voltage, more efficient power delivery systems.

Table 6 tells us that outsourced semiconductor assembly and test (OSAT) providers are entering a period of rapid growth tied to AI packaging demand. These companies thrive on complexity. ASE (NYSE:ASX) is guiding for a 12–14% sequential revenue increase in Q3 2025, while Amkor (NASDAQ: AMKR) expects a ~27% rise, supported by both high-end smartphones and AI edge modules. These gains signal robust order pipelines for advanced packaging and 2.5D/3D assembly for the remainder of the year.

Investor Takeaway

The semiconductor industry’s first-half 2025 performance confirms that the AI-driven growth cycle marks not a passing phase but a structural shift in market composition, capital allocation, and technology roadmaps. Overall sales growth is accelerating, for which the hyperscaler investment wave outlined in Table 2 is the crucial driver. Happily for investors, this creates multi-year visibility for the supply chain, from foundry wafers to advanced packaging substrates.

Packaging capacity will be the critical gating factor for AI chip availability into 2026, positioning OSAT providers, substrate manufacturers, and advanced packaging material suppliers for outsized earnings leverage. HBM remains the performance and profit engine of the memory sector, with NAND recovering in parallel. Tables 5 and 6 highlight two additional secular themes: the transformation of datacenter power architectures to 800V DC, which opens a multi-year content cycle for GaN and SiC power devices, and the acceleration in OSAT demand as AI packaging becomes more complex and capacity-intensive.

For investors, the priority should be overweight exposure to advanced packaging, memory leaders in HBM and NAND, and power semiconductor companies aligned with high-voltage datacenter upgrades, while maintaining select positions in foundries with visible AI-driven wafer demand.

We report in detail on the ascendent firms in these fields in the Gilder Technology Report.