AI Capacity Advances; Its Cost of Capital Is Visible
HBM4 shipments, a $3.5 billion project-debt close and hyperscaler approval of onsite power show the buildout moving forward under tighter capital and energy constraints.
Executive Signal
The physical AI buildout advanced overnight across three layers that increasingly have to scale together: memory, financing and power. SK hynix said it began HBM4 mass shipments in the second quarter, completed long-term agreements with around 10 customers and continues to see demand above available supply. Galaxy closed a $3.507 billion secured-note package for a 260 MW CoreWeave campus in Texas, proving that large AI projects can still attract debt—but at a 9.875% coupon and with an uncapped sponsor completion guarantee. Bloom Energy, meanwhile, reported sharply higher revenue and said its onsite power systems have been validated and approved by all major U.S. hyperscalers and more than a dozen neoclouds, AI labs and colocation operators.
The market response is less forgiving than the operating evidence. SK hynix shares fell 9.61% after record results missed elevated expectations, while long-duration yields and oil remained high. Commercial progress is real, but investors are now distinguishing between capacity that is contracted, financed and deliverable and capacity that still depends on perfect execution.
Today’s central thesis: The AI infrastructure cycle is not stalling; it is becoming more contract-heavy, more power-dependent and more expensive to finance.
Market Positioning
Indicator Latest verified reading Signal Nasdaq 100 futures +0.09% U.S. technology risk is steady ahead of megacap earnings and the Fed STOXX Europe 600 +0.13% European equities remain narrowly positive KOSPI Nearly -6% AI-memory expectations are being reset sharply SK hynix -9.61% Record earnings were insufficient against aggressive HBM4 expectations Brent crude $86.86/bbl, +3.29% Higher fuel costs raise the operating hurdle for power-intensive assets U.S. 10-year Treasury 4.6103% Long-duration financing remains expensive U.S. 30-year Treasury 5.0948% Infrastructure valuations face a persistent discount-rate burden
Market readings are time-sensitive Reuters snapshots and may change before the U.S. open. Global markets
What Changed Overnight
1. SK hynix moves HBM4 into commercial supply—but the ramp remains under scrutiny
Evidence: SK hynix said HBM4 mass shipments began in the second quarter, production will ramp in the second half, and long-term agreements have been finalized with around 10 customers. It also shipped HBM4E samples in the first half and is accelerating M15X, Yongin Phase 1 and P&T7 capacity projects. Management said customer demand exceeds current supply capability. SK hynix Q2 results · Reuters earnings response
Why it matters: HBM4 has crossed from qualification into revenue-bearing supply, strengthening demand visibility for advanced memory, packaging and equipment. Yet the share-price decline shows that commercialization alone is no longer enough; shipment pace, yields and customer allocation now determine incremental valuation.
Market exposure: SK hynix, Samsung Electronics, Micron, NVIDIA, advanced-packaging suppliers and semiconductor equipment.
Conviction: High on shipment commencement and demand; medium on the speed of the production ramp relative to market expectations.
2. Galaxy finances CoreWeave capacity—but the coupon exposes the true cost of scale
Evidence: Galaxy Helios Data Centers II completed a $3.507 billion offering of 9.875% senior secured notes due 2031, issued at 99.5% of principal. The financing supports two buildings with 400 MW of utility capacity and 260 MW of critical IT capacity in Dickens County, Texas. CoreWeave is the tenant, while Galaxy Digital Holdings provided an uncapped completion guarantee. SEC filing
Why it matters: The closing demonstrates that contracted AI infrastructure can access multi-billion-dollar project debt. It also quantifies the burden: a near-double-digit coupon, construction obligations and concentrated tenant exposure. Bankability increasingly depends on lease quality, power delivery and sponsor guarantees—not projected AI demand alone.
Market exposure: Galaxy Digital, CoreWeave, data-center lenders, electrical equipment, construction contractors and Texas power infrastructure.
Conviction: High on the completed financing terms; medium on construction timing, lease commencement and eventual debt-service coverage.
3. Bloom pushes onsite power toward standard AI-campus infrastructure
Evidence: Bloom Energy reported second-quarter revenue of $1.065 billion, up 165.5% year over year, and raised 2026 revenue guidance to $3.9–$4.2 billion. Management said every major U.S. hyperscaler and more than a dozen U.S. neoclouds, AI labs and colocation operators have validated and approved its onsite power solutions. Bloom Energy Q2 filing
Why it matters: Grid interconnection delays are turning behind-the-meter generation from a temporary workaround into a procurement category. Approval expands Bloom’s addressable market, but actual economics will depend on signed orders, manufacturing throughput, gas availability, service performance and the speed of deployment.
Market exposure: Bloom Energy, fuel-cell suppliers, gas infrastructure, utilities, turbine manufacturers and data-center developers.
Conviction: High on reported results and management’s approval statement; medium on customer-specific conversion because individual commitments were not disclosed.
EPIXCE Framework
Signals
AI demand remains broad enough to support next-generation memory, financed campuses and dedicated power. The negative equity response to SK hynix is therefore not a demand-collapse signal; it is an execution and expectations signal. Supplier fundamentals can strengthen while valuations compress.
Capital
The Galaxy financing is the clearest overnight price of AI capacity. Long-term leases and sponsor support can unlock debt, but 9.875% capital requires high utilization and disciplined delivery. Contract structure is becoming as important as headline megawatts.
Power
Bloom’s customer validation reinforces onsite generation as a control point for faster energization. At the same time, Brent above $86 and long-term Treasury yields above 4.6% raise both operating and financing costs. Rapid power is valuable precisely because conventional grid access is scarce—but fuel, emissions, permitting and reliability remain material constraints.
Today’s Market Implications
Potential beneficiaries: HBM and advanced-packaging suppliers; onsite-power providers; contracted data-center platforms; equipment vendors with funded backlogs.
Potentially exposed: Leveraged or uncontracted data-center projects; suppliers priced for flawless HBM4 ramps; hyperscalers unable to defend rising capex with revenue and utilization.
Key catalysts: The FOMC decision at 2:00 p.m. ET and press conference at 2:30 p.m. ET, followed by Microsoft and Meta earnings after the close. Investors should focus on financing conditions, AI monetization, capex guidance and power availability. Federal Reserve calendar
What would invalidate the thesis: A material HBM4 production setback; delayed delivery or weakened lease economics at the CoreWeave campus; failure of Bloom approvals to convert into orders; or broad hyperscaler capex reductions.
Watchlist
Time ET Event Why it matters U.S. session SK hynix and semiconductor reaction Tests whether the selloff is company-specific or a broader AI-supply reset 2:00 p.m. FOMC decision Sets the near-term cost-of-capital framework 2:30 p.m. Fed press conference Guidance on inflation, energy and future policy After close Microsoft and Meta results AI demand, capex, utilization and power commentary Near term Galaxy construction and Bloom order disclosures Conversion from financing and approval into operating capacity
Bottom Line
AI capacity is not waiting for cheaper money or easier grid access. It is advancing through long-term memory agreements, high-coupon project debt and behind-the-meter generation. The investable distinction is increasingly simple: assets already contracted, financed and deliverable deserve a different valuation from narratives that still require every assumption to hold.