Challenges in data center construction and open-source models may be overblown, with underlying demand for semiconductors remaining solid.

On Tuesday, the semiconductor sector, led by memory chip makers, surged over 5%, marking its second consecutive day of outperforming the broader market.

Analysis suggests that while short-covering provided some short-term support, a more compelling argument is that neither local opposition to data center construction nor the rise of open-source large language models is sufficient to signal the end of the AI capital expenditure cycle.

Recent export data from Asia and price increases from manufacturers like Taiwan Semiconductor Manufacturing Company Ltd (NYSE: TSM) confirm that underlying demand for computing power, memory, networking, and power infrastructure remains intact. Market sentiment had previously become excessively pessimistic.

As these two major concerns are gradually disproven, the rebound from last Friday's low may represent more than just a technical recovery from oversold conditions.

Data Center Construction Hurdles Likely Mean Delays, Not Cancellations

Local resistance to data center construction does present a real challenge.

According to data cited by Morgan Stanley, approximately $156 billion in data center projects were canceled or delayed in 2025, with another $130 billion impacted in the first quarter of 2026.

The firm estimates AI capital expenditure for 2026 at $877 billion, facing multiple downside risks including grid limitations, construction moratoriums, and stricter regulations on power and water usage.

However, given the strategic importance of AI, the most likely outcome of this pushback is project delays and redesigns, not wholesale cancellations.

From another perspective, construction constraints could actually boost semiconductor procurement within existing sites.

Operators can generate more computing output within the same physical footprint and power supply by replacing inefficient servers, increasing rack density, and introducing advanced cooling systems.

On-site power generation, battery storage, and grid support services offer another path to bypass transmission bottlenecks.

These constraints will objectively shorten the lifespan of older equipment, thereby increasing replacement demand for newer chips, memory, and cooling systems.

Lower API Pricing for Open-Source Models Does Not Equate to Reduced Hardware Demand

The impact of China's latest wave of open-source large models on hardware demand also appears limited.

Taking Kimi K3 as an example, reports indicate each service instance of the model consumes significant amounts of high-bandwidth memory (HBM) and numerous accelerators. While the model may complete individual tasks with lower computing power, its overall demand for memory capacity remains substantial.

Further research from Bank of America notes that falling API prices in China should not be interpreted as a signal of declining hardware costs.

Cheaper pricing reflects improved architectural efficiency, along with China's more efficient electricity, labor, and land costs, coupled with aggressive market share strategies, rather than a systemic collapse in semiconductor hardware costs.

From a longer-term perspective, the proliferation of open-source models could even expand the overall addressable market for semiconductors.

Closed models concentrate hardware deployment in a few cloud computing facilities. Open models, however, can be downloaded and independently deployed by enterprises, governments, and sovereign clouds, creating new, incremental demand for HBM, DRAM, and NAND memory installations at each end-node and driving additional hardware purchases.

In summary, the rebound in AI capital expenditure beneficiaries from recent lows may be driven by factors beyond a simple technical recovery from oversold conditions.

While earnings expectations may have peaked for the current cycle, the underlying economic rationale for AI infrastructure build-out appears more resilient than the market had recently feared, based on available data.

Market positioning had adjusted to excessively pessimistic levels, yet the fundamentals did not deteriorate materially, creating room for the rebound to continue.