THOUGHT OF THE DAY
AI infrastructure is being repriced around execution, not announced capacity. Microsoft’s reported cancellation of a major Nscale data-center deal amid grid-connection delays, alongside a steep selloff and insider share sales at CoreWeave, puts delivery risk and financing credibility in sharper focus. This updates the power-constraint story: access to capital, permits, and a credible path to energization now determine which announced projects can become billable capacity.
The semiconductor outlook is diverging from the financing outlook for AI infrastructure. The news points to continued strength at TSMC and ASML, while infrastructure providers face tighter funding conditions and Broadcom’s CEO has reportedly identified capital—not wafers—as a key limit on expansion. Chip demand can remain strong even as the ability to finance and deploy new compute becomes the bottleneck.
AI-generated software is sharpening the challenge to incumbent application vendors. Adobe was downgraded on concern that open-source models and AI coding tools could make some software alternatives cheaper to build. The catalyst is a more credible path from model capability to usable substitutes; the investment question is whether incumbents can defend pricing through workflow integration and customer value.
COMPUTE & SEMICONDUCTORS
- AMD and Sterling/BitRipple are integrating LT3 transport software exclusively with AMD GPUs in Sterling’s AI Factory offerings. The claimed benefit is better use of existing clusters without new hardware purchases, a useful pitch for cost-sensitive enterprise and public-sector buyers. Large deployments and independently validated performance gains remain the proof points.
- TSMC reported a 68.7% increase in net profit and a 66.4% gross margin, while ASML said it cannot meet demand beyond 2027. That supports continued pricing power for leading-edge manufacturing and lithography, but does not remove capital and deployment constraints elsewhere in the AI chain.
- Broadcom’s AI custom-silicon business is reported to be growing rapidly, but access to advanced TSMC capacity remains a key execution dependency. The reported warning that capital, rather than wafers, may constrain AI expansion highlights a financing risk alongside physical supply limits.
DATA CENTERS & INFRASTRUCTURE
- Microsoft reportedly withdrew from a major Nscale data-center deal after grid-connection delays pushed power availability far into the future. Separately, reports cite widespread project delays or cancellations amid permitting and community opposition. Announced capacity is not equivalent to deliverable capacity.
- CoreWeave reportedly saw its shares fall 25% while insiders sold $3.3 billion in stock. The combination raises questions about investor confidence and the financing durability of fast-growing, capital-intensive compute providers.
- Drax Group committed more than £500 million to battery storage, while Volex is positioned to supply high-speed copper interconnects for data centers. These investments address different infrastructure needs, but their commercial value depends on projects proceeding and securing power.
ADOPTION & MONETIZATION
- Accenture reported an $11.5 billion AI backlog and $84.5 billion in annual bookings, pointing to substantial demand for AI-related services. Backlog is a demand signal, not proof of realized revenue or customer returns.
- Enterprise-software sentiment remains mixed: Salesforce was cited as receiving a double upgrade, while Adobe faced a downgrade tied to potential AI-enabled software substitutes. The market is distinguishing between vendors that can embed AI into customer workflows and those exposed to cheaper alternatives.
POSITIONING IDEAS
Bullish
- ASML — Reported inability to meet demand beyond 2027 supports pricing power and the strategic value of scarce lithography capacity.
- AMD — The Sterling/BitRipple integration offers a route to differentiate on utilization and software, though broad adoption is not yet established.
Bearish
- CoreWeave and highly leveraged AI-infrastructure developers — The reported share decline, insider sales, and project delays highlight financing and execution risks for capacity that is not yet reliably powered or monetized.
- Adobe — The downgrade reflects a credible competitive risk if AI coding tools and open-source alternatives reduce the cost of building or replacing parts of its software suite.