Federal Nuclear Loans and Memory Shortages Bolster AI Supply Chains

The setup — Tuesday's session and Wednesday morning's disclosures brought concrete federal and foreign government backing to power and critical-mineral projects, alongside warnings from chipmakers that advanced memory and packaging bottlenecks will persist for years. Those developments strengthen the supply-constrained theses behind our power and compute hardware themes, even as capital spending outpaces free cash flow at the largest buyers and credit investors demand steep yields from debt-funded cloud builders.

What changed for the themes

Our AI Power & Critical Minerals theme gained support as federal financing and utility orders backed the view that data centres need round-the-clock power and grid equipment that take years to build. On Tuesday, the US Department of Energy offered Constellation Energy ($CEG) a conditional loan commitment of up to $5 billion to finance 1.1 gigawatts of nuclear power uprates across eight plants in Illinois, Maryland, and Pennsylvania, while Antares Nuclear secured a fuel agreement with BWXT for its microreactor — both reinforcing the narrative that nuclear power is being revived to run data centres. During Tuesday's session, Utility Dive reported that BloombergNEF raised its US energy storage deployment forecast by 30% as hyperscaler interest converted into firm orders (per Utility Dive). After the close on Wednesday morning, GE Vernova ($GEV) beat third-quarter estimates, raised its full-year revenue guidance, and agreed to buy high-voltage grid equipment maker Trench Group for $9 billion (per MT Newswires).

Our AI Compute Build-out theme has been losing support overall as AI infrastructure capital spending slows, but new developments during and after Tuesday's session bolstered the narrative that advanced memory and packaging remain tight supply bottlenecks. After Tuesday's close, South Korean memory maker SK Hynix reported record quarterly profit (per CNBC) and stated that memory supply constraints will last through 2030 (per MarketWatch). That pairs with Tuesday afternoon coverage of Taiwan Semiconductor Manufacturing ($TSM) reportedly preparing to raise prices on its CoWoS advanced packaging by 10% to 20% as equipment delivery times stretch from roughly three to six months out to six to eight months (per GuruFocus). Separately, Google began opening its custom Tensor Processing Unit business beyond its own cloud to third-party data centres, with Oracle ($ORCL) announcing Tuesday it will buy Google's AI processors (per CNBC).

Our AI Platforms & Cloud theme remains contested as the narrative that the AI build-out is outgrowing the cloud giants' cash flow took another step on Tuesday. Following Monday's news of Meta's ($META) planned $95 billion stock offering — which we examined in yesterday's brief — CNBC reported during Tuesday's session that chip suppliers Nvidia ($NVDA) and Advanced Micro Devices ($AMD) plan to invest $4 billion in Meta as part of that financing (per CNBC).

Our AI Software & Agents theme is gaining support from falling model costs. On Tuesday, OpenAI released GPT-6.1 Terra, a mid-tier model for agentic tasks and computer use that costs roughly 50% less than its predecessor while improving speed and accuracy, adding evidence to the narrative that falling inference prices are commoditising model access for software vendors. Nothing material changed in Tuesday's session for our Robotics & Physical AI theme.

Risks to watch

The central risk working against both the compute and platform themes is that AI spending is outrunning its returns and outgrowing internal cash flows. By our signals dashboard, hyperscaler AI infrastructure capital spending is running 80.3% above last year, but its momentum has turned negative (-0.71 standard deviations), placing the cycle in slowdown while high-yield credit spreads have widened by 37 basis points over the past month to 330 basis points.

That credit tightening draws a sharp line between companies that fund capital spending from operating cash flow and those that must borrow to build. Even giant platforms like Meta are turning to $95 billion in equity offerings after capital expenditures pushed free cash flow negative, while our readings flag Oracle for burning cash with rising debt. For specialised cloud builders, the strain is acute. Over the past year, CoreWeave's ($CRWV) capital spending reached 298% of its operating cash flow and its long-term debt rose 185%. Coverage after Tuesday's close noted that a $2.75 billion CoreWeave bond due in 2031 with a 9.75% coupon has fallen in price until its yield reached 15.76%, while peer Nebius ($NBIS) relies on a 364-day, $3.5 billion credit facility from Citigroup priced at Term SOFR plus 2.75 to 3.50 percentage points (per BeInCrypto, GuruFocus). A pause in hyperscaler capital budgets or credit drying up for debt-funded cloud builders would break the compute theme.

In the power and critical minerals theme, where our readings flag USA Rare Earth ($USAR) for heavy dilution, project delays and financing conditions remain the primary hurdle.

The companies

USA Rare Earth ($USAR), Energy Fuels ($UUUU), and Denison Mines ($DNN) all ride our AI Power & Critical Minerals theme, and all three rallied on Tuesday. Energy Fuels rose 8.9% to $8.23 after announcing late Monday that Australian government-backed agencies conditionally approved up to AU$180 million ($126.4 million) in debt financing for its Donald rare earth project in Victoria (per MT Newswires). That financing remains conditional on commercial lenders committing before a targeted year-end final investment decision (per 24/7 Wall St.). USA Rare Earth gained 16.7% to $10.40 and Denison Mines rose 8.6% to $2.39 alongside a 5% gain in the uranium ETF and a 4% gain in the rare earth ETF, though nothing in Tuesday's filings or company-specific disclosures explains either stock's move; Grounded Lithium filed a technical report on its joint project with Denison only on Wednesday morning (per PR Newswire). Our readings flag heavy dilution on USA Rare Earth and a falling share price trend on Denison Mines, pointing the opposite way from Tuesday's rally.

CoreWeave ($CRWV) rose 7.9% to $59.57 on volume 39% above its three-month average after switching on new compute capacity at its Copenhagen data centre in Denmark (per Motley Fool). Peer Nebius Group ($NBIS) climbed 11.2% to $170.47 without a new company filing or catalyst of its own during the session. Both stocks remain in falling price trends by our readings, and CoreWeave carries our warning for burning cash after stock-based compensation (-188% of revenue) while operating profit falls short of interest expense.

GE Vernova ($GEV) rose 4.7% to $895.08 on Tuesday without session-hours news, then reported third-quarter revenue of $12.64 billion (up 26.8% year on year) and GAAP earnings of $4.81 per share before Wednesday's open, beating consensus and lifting full-year revenue guidance to a $47 billion midpoint alongside its $9 billion Trench Group acquisition (per StockStory). Taiwan Semiconductor ($TSM), riding our AI Compute Build-out theme with a rising price trend, gained 3.0% to $436.87 as reports of its 10% to 20% CoWoS price hike circulated during the session. Finally, Aeva Technologies ($AEVA), riding our Robotics & Physical AI theme, gained 5.4% to $10.44 with nothing in Tuesday's filings or coverage to explain the move; while it trades at the 14th percentile of its valuation history, our readings flag a falling price trend and cash burn equal to 675% of revenue.

The backdrop

Tuesday's session favoured hardware and power over application software. The S&P 500 ETF ($SPY) rose 0.6%, the Nasdaq Composite gained 1.2%, and the Technology Sector ETF ($XLK) added 1.7%, led by a 3.0% advance in the Semiconductor ETF ($SMH) while the Software ETF ($IGV) rose 0.4%. Even as software gains support in our theme readings and includes potential opportunities like Salesforce ($CRM), ServiceNow ($NOW), and Adobe ($ADBE), Adobe (-4.0%) and Salesforce (-2.7%) were among the session's largest declines, while Nvidia ($NVDA) remains a potential opportunity on the compute theme despite an adverse context. Across our tracked universe, 53 of 68 companies rose, with a median gain of 3.0%; as last measured, the median tracked company sits at the 67th percentile of its own valuation history. The 10-year Treasury yield edged up one basis point to 4.99%, WTI crude fell 1.8% to $99.90, and $VIX slipped 4.7% to 17.52.

What to watch

• Wednesday's infrastructure and software earnings: Vertiv ($VRT) is scheduled to report before Wednesday's open, followed after the close by KLA Corporation ($KLAC), Microsoft ($MSFT), and ServiceNow ($NOW) — testing whether hyperscaler capital budgets, wafer-fab equipment orders, and enterprise software seat pricing are holding up.
• Alphabet's third-quarter results: Alphabet ($GOOG, $GOOGL) is expected to report after Wednesday's bell, where cloud growth and external TPU sales to partners like Oracle will test our AI Compute Build-out and AI Platforms themes (per Yahoo Finance).
• Thursday's platform and robotics reports: Mobileye ($MBLY) is scheduled to report before Thursday's open, and Amazon ($AMZN) and Intel ($INTC) after the close, offering a read on automotive sensor demand and cloud capital spending.
• Next week's hardware, software, and mineral updates: Palantir ($PLTR) and Energy Fuels ($UUUU) are expected to report on Monday, 2 November (with Energy Fuels hosting its earnings call on Thursday, 5 November (per PR Newswire)), followed by Advanced Micro Devices ($AMD) and Navitas Semiconductor ($NVTS) on Tuesday, 3 November, and Datadog ($DDOG) and Aeva Technologies ($AEVA) on Wednesday, 4 November.