Semiconductors race toward $1.6T

The headline number is no longer theoretical

The semiconductor industry has spent years talking about the road to a trillion-dollar market. On September 20, that debate moved sharply forward: the global chip market is now projected to exceed $1.6 trillion in 2026, according to reporting on a Korea Eximbank research-center forecast that drew on industry analysis including TrendForce [1]. Seoul Economic Daily separately reported the same core forecast: about $1.6 trillion in 2026, roughly double last year’s market size and 60% above the bank’s own second-quarter projection of about $1 trillion [2].

That is the story: not just a bigger semiconductor market, but a faster and more concentrated one. The new forecast says memory chips could represent 55% of total semiconductor revenue in 2026, up from about 30% last year and slightly above an earlier 50% estimate [2]. Yonhap described the shift as memory moving to more than half of the global chip market as AI systems process larger volumes of data [1].

This matters because it changes the map of bottlenecks. For much of the AI build-out, the public narrative has centered on accelerators, GPUs and leading-edge logic. Those remain critical. But the latest numbers say the tightest value pools are also in the components that feed and surround compute: DRAM, NAND and especially the memory bandwidth required to keep AI processors busy.

Memory becomes the center of gravity

Seoul Economic Daily reported that Korea Eximbank expects the DRAM market to reach $519 billion in 2026, or 3.4 times last year’s size, while NAND could expand 4.9 times to $360 billion [2]. The same report said the bank raised its growth forecasts to 250% for DRAM and 395% for NAND, reflecting both AI-driven demand and persistent supply shortages [2].

Those price dynamics are not cosmetic. Yonhap reported that, as of the end of August 2026, DRAM and NAND flash prices stood at $25 and $30.5 respectively, up 4.4-fold and ninefold from a year earlier [1]. In other words, this is not just a unit-growth story. It is a market-value story driven by sharply higher average selling prices, with AI infrastructure absorbing more memory even as supply struggles to catch up.

McKinsey’s September 18 analysis points in the same direction. The consultancy wrote that the semiconductor market had crossed $1 trillion in revenue for the first time in 2026 and that AI has lifted average selling prices for leading-edge chips and memory well above earlier expectations [3]. It now projects a $2.3 trillion semiconductor market by 2030 in its base case, up from the $1.6 trillion estimate it used in autumn 2025 [3].

The most important line in that analysis may be its description of leading-edge logic and memory as the “twin engines” of growth through 2030 [3]. Memory, in this framing, is not a commodity afterthought. It is a structural limiter for AI scaling. McKinsey notes that high-bandwidth memory sits close to GPUs or CPUs and supplies the bandwidth needed to keep expensive compute resources operating efficiently [3]. If the memory pipe is too narrow, the accelerator sits underused; if the pipe expands, the whole data-center economics model changes.

The AI slowdown debate has not slowed spending

The timing of the forecast is important because it lands during renewed debate over whether AI development should slow. Seoul Economic Daily reported that the $1.6 trillion forecast came despite recent arguments by major AI developers for slowing model development because of safety concerns, which had fed worries about weaker demand for chips [2]. The report also said executives at Anthropic, OpenAI, xAI and Google had recently warned about AI risks and argued for a slower pace of model development, pressuring shares of chipmakers including Samsung Electronics and SK hynix [2].

Yet the capacity signal points the other way. Seoul Economic Daily reported that capital expenditure by the world’s nine largest hyperscalers is expected to rise 98% in 2026 from last year, above an earlier forecast of 80% [2]. McKinsey’s analysis is even broader: it says five major hyperscalers have announced planned capital expenditures of about $800 billion for 2026 and $1 trillion for 2027, with most of that directed toward AI infrastructure [3].

That combination creates a paradox for planners. Public concerns about AI risk may produce volatility in chip stocks and policy debates, but the physical build-out of AI infrastructure continues to pull memory and leading-edge logic through the system. The silicon spice must keep flowing because data centers do not scale on algorithms alone. They scale on wafers, packaging, power, interconnects and long-term supply commitments.

TSMC’s moat remains the logic-side anchor

The memory surge does not erase the importance of foundry power. It makes it more visible. The Motley Fool reported on September 20 that TSMC owned about 72.5% of global chip foundry revenue in the second quarter, citing TrendForce, while the second-place foundry held 5.9% [4]. The same article argued that no competitor can easily take large share from TSMC because no rival has comparable capacity [4].

That is the supply-chain moat. If memory is the segment where revenue is exploding, TSMC remains the manufacturer that anchors much of the leading-edge logic used by AI companies. The Motley Fool reported that 66% of TSMC’s second-quarter revenue came from high-performance computing, a segment that includes AI demand [4]. The same article noted that TSMC makes chips for nearly every leading big-tech firm [4].

The practical implication is that the semiconductor boom is bifurcated but interdependent. Memory makers such as Samsung Electronics and SK hynix gain leverage as DRAM, NAND and HBM become scarcer and more valuable; foundry leaders such as TSMC gain leverage because advanced AI processors still require leading-edge manufacturing scale. The bottleneck may move, but it does not disappear.

Long-term contracts signal a less familiar cycle

Memory has historically been cyclical: shortages lift prices, new supply arrives, prices fall. The current cycle looks different because hyperscalers appear willing to secure multi-year access. Seoul Economic Daily reported that big tech firms are pursuing three- to five-year long-term agreements with memory producers such as Samsung Electronics and SK hynix, a shift Korea Eximbank said could stabilize industry volatility compared with past cycles [2]. Yonhap also reported that long-term supply agreements with those memory makers could help stabilize the semiconductor industry [1].

There is still a timing problem. Seoul Economic Daily reported that DRAM shortages may not ease until the second half of 2027 or 2028, while NAND shortages may persist until after the first half of 2027 [2]. McKinsey similarly argues that few short-term solutions exist for memory shortages and that supply may begin catching up around 2027 if new fabs, advanced packaging and higher bit density succeed [3].

That is why the $1.6 trillion figure is not just a victory lap. It is a warning label. Equipment makers will need to follow the capacity migration from logic to memory and packaging. Hyperscalers will need to budget for higher component prices and longer commitments. Governments will need to decide whether chip policy focused mainly on leading-edge fabs is sufficient when memory, power and packaging can become the next limiting factor.

What to watch next

The near-term test is whether memory prices keep rising into late 2026 and early 2027 as forecast [2]. If they do, the 55% memory share will not look like a one-off spike; it will look like a reset in where semiconductor value is captured. The medium-term test is whether new supply arrives without collapsing prices in the classic memory-cycle pattern. The strategic test is whether the industry can expand memory, leading-edge logic and advanced packaging at the same time.

For now, the market is sending a clear signal. AI demand is no longer measured only in accelerator backlogs. It is showing up in DRAM, NAND, HBM, foundry utilization, hyperscaler capex and long-term supply agreements. Semiconductors are racing toward $1.6 trillion, and the race is increasingly being won by whoever controls the scarce layers of the AI stack.

Sources from the last 72 hours

1. [1]Global chip market projected to exceed $1.6 tln this year despite AI slowdown concerns: reportSep 20, 2026, 5:38 AM UTC
2. [2]Chip Market to Double to $1.6 Trillion This Year Despite AI Slowdown CallsSep 20, 2026, 6:50 AM UTC
3. [3]The $2.3 trillion horizon: How AI is rewriting the semiconductor storySep 18, 2026, 12:00 AM UTC
4. [4]Taiwan Semiconductor Manufacturing's Foundry Market Share Is a Massive Moat Nobody Talks AboutSep 20, 2026, 9:03 AM UTC

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.