Wedbush's hardware team published a note covering three developments that, taken together, paint a picture of an AI infrastructure market running hotter than supply can support. Blackwell GPUs are becoming difficult to procure. Memory pricing is rising. Intel CPUs are scarce. Components across the board, from PCBs to capacitors, are constrained.
The demand is not slowing. It is accelerating into a supply chain that was already stretched.
Supermicro's $7bn raise
Supermicro announced concurrent equity-linked offerings totalling approximately $7bn: $1.3bn in common stock, $3.78bn in depositary shares representing mandatory convertible preferred stock, and a $2bn at-the-market programme beginning no earlier than the third quarter.
The proceeds are earmarked for one purpose: buying components against roughly $39bn in AI server orders received in recent weeks from more than 20 customers.
Wedbush analyst Matt Bryson described the order momentum as positive, particularly given lingering questions about how customers would react to earlier reports of illicit shipments to China. But the financing is dilutive, and the margins on the orders will determine the financial impact.
The more significant read-through is for the broader market. Bryson noted that Supermicro's announcement is "directly indicative of continued momentum" for peers including Dell, HPE and Gigabyte. The vast bulk of the demand is tied to Nvidia-based platforms, supporting expectations of robust GPU momentum in coming quarters.
The supply picture is tightening further. Blackwell pricing has lifted or is expected to lift in certain segments. Secondary market prices for memory modules have taken another leg up after appearing to stabilise. Intel CPUs remain particularly scarce. The component shortage is broad-based, not limited to any single part of the stack.
China's $295bn domestic push
Beijing is drafting a plan to spend 2 trillion yuan, roughly $295bn, over five years on a nationwide AI data centre network. The critical detail is the sourcing requirement: at least 80% of the technology, including chips, must be domestically produced. China Mobile and China Telecom would link the facilities into a single computing grid by 2028.
The 80% rule locks Western suppliers out of the Chinese AI infrastructure build. For companies like Nvidia, AMD and Intel, the direct revenue impact is negative.
Wedbush's view is counterintuitive: the near-to-intermediate-term effect on the Western supply chain is paradoxically positive. Chinese manufacturers attempting to fill the domestic supply gap are consuming foundry capacity, particularly at SMIC, whose most advanced node is already running above 93% utilisation. That tightens global component supply, particularly for mature foundry processes used across the industry.
The one Western company that could benefit directly is AXT, which has significant domestic China capacity and would see increased demand for indium phosphide substrates as Chinese infrastructure buildout drives demand for optics.
OpenAI's 10-gigawatt campus
The Information reported that OpenAI is in advanced negotiations to lease a proposed 10-gigawatt data centre campus in Ohio. SoftBank would develop the facility, with an eventual price tag exceeding $500bn at full deployment. Nvidia could provide financial backing.
Wedbush sees this as a strategic shift for OpenAI, which has relied on third parties to build and operate data centre capacity until now. The move towards owning infrastructure mirrors what hyperscalers did a decade ago, transitioning from rented capacity to purpose-built facilities.
For hardware makers, OpenAI and SoftBank builds would more likely use OEM services than the ODM-first approach favoured by hyperscalers, which benefits Dell, Supermicro and HPE. For Cerebras, a shift towards owned infrastructure could mean later tranches of its OpenAI agreement move from cloud capacity leases to equipment purchases.
Unified picture
The three stories describe the same phenomenon from different angles. Demand for AI compute is accelerating across every major geography and every major customer category. The supply chain cannot keep pace. Prices are rising. Lead times are extending. Companies are raising billions in dilutive equity to secure components.
The AI infrastructure boom is no longer a forecast. It is a procurement crisis.
