Newsletter comparison
The Diligence Stack vs Damnang’s Substack
Short answer: Choose The Diligence Stack for an investment map of AI capacity and suppliers. Choose Damnang's Substack for power, memory, and near-term market pressure. Keep both when physical infrastructure and public-market expectations need to be tested together.[1][2][3][4][5][6]
An independent guide. Scottie isn't affiliated with, endorsed by, or sponsored by The Diligence Stack and Damnang’s Substack. Each publication owns its name, writing, and subscription terms.
A brief from both newsletters · July 31, 2026
See both newsletters in one brief
See what made the brief, what didn't, and the original links behind every included story.
In this brief: Scottie kept the selected stories separate because they added different value.
Reader priorities
An investor following the AI buildout needs evidence about physical capacity, suppliers, capital spending, and the market's changing expectations.
These are illustrative priorities, not a customer’s data.
What made the brief
- The Diligence Stack3 read · 3 included
- Damnang’s Substack3 read · 3 included
Scottie
Scottie example brief
July 31, 2026 · Executive brief
01 / The rundown
- Hyperscalers are adopting behind-the-meter power solutions to bypass multi-year grid delays and accelerate immediate physical data center capacity.
- Contracted cloud demand is outstripping capital expenditure growth, forcing tech giants to lease third-party physical infrastructure through 2030.
- Fears of a memory downturn are overblown because high-bandwidth memory production consumes quadruple the wafer capacity of commodity DRAM.
02 / The briefing
01 / main
Hyperscalers Turn to Off-Grid Power to Speed Up AI Deployments
The takeaway: Hyperscalers are embracing behind-the-meter power systems to sidestep multi-year utility grid queues. Driven by manageable inference power draws and new rack-level smoothing tech, Bloom Energy reports all major U.S. hyperscalers are now actively pursuing off-grid capacity.
Concrete details
- Bloom Energy Q2 revenue jumped 165.5% year-over-year to $1.065 billion with a 33.4% GAAP gross margin.
- Interactive inference workloads cause maximum power spikes of just 9%, compared to 37.5% for heavy AI training.
Why it matters for this reader: Track how off-grid generation suppliers enable faster capacity expansion for your portfolio companies without waiting on delayed local electric utility connections.
Original sourcesThe Diligence Stack
02 / main
Power Deficits Force Hyperscalers Into Leased Physical Infrastructure Partners
The takeaway: Electric grid studies taking up to a year and substation builds taking three years will keep power constraints tight through 2030. Cloud giants are increasingly relying on leased data center shells to secure physical footprints while retaining full ownership of servers.
Concrete details
- The contracted demand ratio of capex to backlog dropped from 44% in 2024 down to 37% in 2026.
- Core Scientific has 437 megawatts actively billing out of 1.1 gigawatts currently leased to infrastructure partners.
Why it matters for this reader: Evaluate how physical shell landlords preserve hyperscaler margins while expanding total addressable compute capacity ahead of owned facility delivery schedules.
Original sourcesThe Diligence Stack
03 / main
Early AI Forecasts Vastly Underestimated Supporting Component Expansion Costs
The takeaway: Initial market models failed to predict how fast hardware around AI accelerators would scale up. Blended semiconductor prices are expanding rapidly, while advanced packaging, networking, and rack power requirements continue to far outpace original 2030 industry forecasts.
Concrete details
- Average semiconductor selling prices are modeled to jump from $0.75 today to over $2.10 by 2028.
- Frontier AI rack power density has already hit 120 to 142 kilowatts, exceeding old 2030 targets.
Why it matters for this reader: Recalibrate your capex expectations across the supply chain as non-logic infrastructure components capture an expanding share of hardware budgets.
Original sourcesThe Diligence Stack
04 / main
Memory Capex Worries Overlook High-Bandwidth Wafer Intensity Constraints
The takeaway: Market anxiety over upcoming memory supply gluts misjudges how high-bandwidth memory reduces total bit output per wafer. As new HBM generations require up to four times the physical wafer area of commodity DRAM, bit oversupply risks remain low.
Concrete details
- Shipping 16-high HBM4 requires four times the physical wafer area of standard commodity DRAM memory modules.
- Current selloff prices imply a severe 70% drop in peak quarterly memory supplier earnings over four quarters.
Why it matters for this reader: Use structural HBM wafer consumption penalties to filter out unjustified cyclical selloff noise across your memory supply chain holdings.
Original sourcesDamnang’s Substack
05 / main
CXMT Debut Highlights Deep Cost Disadvantages Against Incumbent Memory Leaders
The takeaway: Despite CXMT's massive domestic stock debut, the manufacturer faces significant cost and yield hurdles. Producing memory costs CXMT 30% more than global leaders, while its transition into HBM absorbs substantial wafer capacity and squeezes standard DRAM supplies.
Concrete details
- CXMT processes 13% of global DRAM wafers but yields only 6% of worldwide usable memory capacity.
- Samsung and SK Hynix trade at low forward multiples of 4.32 and 4.69 times earnings, respectively.
Why it matters for this reader: Protect your positions in incumbent memory suppliers as market fear over Chinese market-share dilution overlooks vast production cost and yield gaps.
Original sourcesDamnang’s Substack
06 / main
Korean Memory Sector Shifts Toward Custom Architectures and Long-Term Agreements
The takeaway: The traditional commodity memory model is evolving into custom logic-integrated architectures like HBM4 and near-memory computing. Field insights from Korea show multi-year supply contracts are providing earnings visibility, though customer power dynamics persist as technical integration challenges mount.
Concrete details
- Next-generation 12-high and 16-high HBM4 stacks introduce major physical and manufacturing yield bottlenecks for fabricators.
- Long-term supply agreements are expanding remaining performance obligations across leading Korean memory component manufacturers.
Why it matters for this reader: Monitor supplier progress in custom HBM integration to identify memory makers capable of commanding structural pricing power over buyers.
Original sourcesDamnang’s Substack
Action items
- Audit your infrastructure portfolio for off-grid power partnerships that bypass multi-year utility grid interconnection queues.
- Stress-test memory holdings against HBM wafer supply penalties to separate cyclical noise from structural earnings potential.
- Model capex expansion rates across non-logic hardware suppliers to capture rising semiconductor average selling price targets.
See every issue behind this brief
Behind-the-Meter Power Acceleration
- The Behind-the-Meter AI BuildoutThe Diligence Stack · Included
Hyperscaler Capacity and Utility Constraints
- The Hyperscaler Capacity Partner HierarchyThe Diligence Stack · Included
AI Capex and Semiconductor Stock Concerns
- What Would Stop This Selloff?Damnang’s Substack · Included
Underestimating Accelerator Ecosystem Scaling
- In 2023-2024 We Weren't Bullish EnoughThe Diligence Stack · Included
CXMT Listing and Memory Industry Impact
- After CXMT’s Listing, Do the Memory Three Have Further to Fall?Damnang’s Substack · Included
Korean Semiconductor Industry Insights
- I Met with Semiconductor Experts in KoreaDamnang’s Substack · Included
The useful difference
The difference that matters
The Diligence Stack follows capacity partnerships, storage demand, and the investment chain around hyperscalers. Damnang's Substack emphasizes power infrastructure, memory, and market reactions to the buildout.[1][2][3][4][5][6]
What each is best for
The Diligence Stack
The Diligence Stack traces demand through storage, capacity partners, hyperscalers, and the suppliers receiving capital.[1][2][3]
Damnang’s Substack
Damnang's Substack adds the electrical and memory constraints, then connects them to price moves and changing investor sentiment.[4][5][6]
When it’s worth reading both
The pair can reveal when a believable infrastructure need is already overstated—or misunderstood—by the market. Scottie can retain both the physical evidence and the valuation signal.[1][2][3][4][5][6]