The AI Capex Cycle: $775–800B Hyperscaler Buildout and the Five High-Conviction Positions

The AI capex cycle is the coordinated surge in capital expenditure by hyperscale cloud operators to build the physical infrastructure — data centres, power, accelerators and memory — required to train and serve large AI models. In 2026 the five largest operators — Amazon, Alphabet, Microsoft, Meta and Oracle — have guided to a combined $775–800 billion, roughly three times the ~$238 billion deployed in 2024, of which approximately 75% is AI-specific. Data as of August 7, 2026, from each company's Q2 CY2026 earnings disclosures.

Hyperscaler capex tracker — 2026 guidance

Data as of August 7, 2026 · each figure from that company's own Q2 CY2026 disclosure

McKinsey's research shows global demand for data center capacity could almost triple by 2030, with approximately 70% of that demand driven by AI workloads. Total projected capital expenditure: $6.7 trillion, of which $5.2 trillion is attributable to AI processing loads and $1.5 trillion to traditional IT applications.

The AI capex cycle is defined by the hyperscalers — the companies that build and operate the cloud infrastructure on which AI runs. Amazon, Alphabet, Meta, Microsoft, and Oracle collectively spent ~$238 billion on capital expenditure in 2024. The Big-5 estimate for 2025 is $429 billion, a 73% YoY increase. Q1 2026 earnings (reported April 29, 2026) confirmed the combined 2026 figure at approximately $775–800 billion — a ~64% increase over 2025 — with approximately 75% of that spend (~$545 billion) directed at AI-specific infrastructure: GPUs, servers, data center construction, power systems, and cooling. Alphabet raised guidance to $195–205B (Q2, partially due to the Intersect acquisition); Meta raised its floor to $130–145B, explicitly citing memory price inflation; Microsoft guided to ~$175B CY2026 (lease reclassification, not a cut) ($25B attributed to higher component pricing); Amazon raised to ~$220B. See Exhibit 1 for the full company-level breakdown. Amazon's 2026 capex of $200 billion alone exceeds the combined annual capex of the entire publicly traded US energy sector. As a share of GDP, AI-related capital formation now sits at approximately 5% — a level last seen during the late-1990s technology boom, but with a structurally different demand foundation (see Section 02).

McKinsey constructed three scenarios ranging from constrained to accelerated demand, shaped by semiconductor supply constraints, enterprise AI adoption rates, efficiency improvements, and regulatory challenges. The base case — $5.2 trillion in AI data center capex — assumes continued growth without runaway acceleration or structural constraints.

The analogy to the late-1990s telecommunications infrastructure bubble is compelling in one dimension — the scale of capital deployment — and misleading in every other. Fiber in the 1990s was built speculatively, with virtually unlimited capacity once laid and zero refresh requirement. Data centers are physically constrained, contractually committed before construction, and subject to accelerated depreciation cycles that naturally absorb any temporary excess. The evidence is visible in vacancy data: North American colocation vacancy has fallen from 9.8% in 2020 to ~1.4% by YE 2025 (JLL/CBRE Research), while the fiber glut post-2001 saw vacancy exceed 20%.

The key structural difference McKinsey identifies is the cost of carrying excess capacity. Fiber, once laid, is nearly free to maintain. Data centers are the opposite: power, cooling, and maintenance are ongoing high costs regardless of utilization. But crucially, AI accelerators have 3–4 year refresh cycles — meaning any overcapacity is rapidly converted into obsolescence, and new workloads pull spare capacity well before it becomes stranded.

The strongest bear case, stated fairly. By mid-2026 the debate had shifted from whether AI demand is real to the financial mechanics beneath it — and four arguments deserve to be taken seriously. Depreciation: hyperscalers depreciate AI hardware over 5–6 years while its economic life may be 2–3, which Michael Burry and others estimate understates true depreciation by roughly $176 billion across 2026–2028, flattering reported earnings. Circular financing: interlocking commitments among NVIDIA, OpenAI, Oracle and CoreWeave — vendor equity stakes, take-or-pay compute deals, and debt-funded GPU purchases — can make end-demand look larger and more independent than it is, echoing the Lucent/Nortel vendor-financing dynamics of 1999–2001. Cash-flow strain: PIMCO estimates combined hyperscaler capex will consume ~94% of operating cash flow by 2026–2027, and Moody's flags ~$662 billion of signed-but-not-commenced data-center leases sitting off balance sheet. Runway: Barclays has flagged a potential hyperscaler cash-runway constraint by late 2026 if the current trajectory holds.

The rebuttal is not that these risks are imaginary — it is that they sit on the buyers' balance sheets, not the suppliers'. The overbuild question is legitimate; the demand-does-not-exist question is not, with ~1.4% vacancy, sold-out HBM and CoWoS, and 12–18-month GPU lead times. This is precisely why A.L. Capital Advisory's five High Conviction positions sit on the supply side: NVDA, VRT, EQIX, CEG and MU are paid on delivery, whether or not each hyperscaler's own AI ROI ultimately clears. The bear case compresses hyperscaler multiples; it does not un-sign the contracts the suppliers are already shipping against.

The AI capex cycle has introduced a structural change to hyperscaler financial models that was not present in prior technology investment waves: the shift from cash-funded to debt-funded infrastructure. Amazon, Alphabet, Meta, and Microsoft spent a combined ~$238 billion on capex in 2024. Q1 2026 earnings confirmed the combined 2026 figure at approximately $775–800 billion. Internal free cash flow cannot scale at that rate. As a result, hyperscalers have collectively turned to debt markets at a scale not seen since the 2001 telecom build-out — but with materially stronger balance sheets underpinning the leverage.

Morgan Stanley and J.P. Morgan project the technology sector will need to issue approximately $1.5 trillion in new debt over the next three years to fund the AI infrastructure build-out. The hyperscalers now spend 31–83% of revenue on capex — ratios previously seen only in capital-intensive industrial utilities and telecommunications companies. Amazon's 2026 capex guidance of $200 billion alone exceeds the combined annual capex of the entire publicly traded US energy sector.

Q2 CY2026 — the quarter self-funding ended. The July 2026 reporting cycle turned this from a projection into a disclosure. Alphabet spent $44.9bn of capital expenditure against $39.1bn of operating cash flow, producing free cash flow of −$5.9bn — the first negative free-cash-flow quarter in Alphabet's history as a public company. It repurchased no stock, against $13.2bn a year earlier, and long-term debt more than doubled from $46.5bn at December 2025 to $98.2bn at June 2026. Amazon's disclosure is the more durable one because it is not a single quarter: trailing-twelve-month free cash flow through 30 June 2026 was −$7.6bn, against +$18.2bn a year prior, which Amazon attributes to a $66.1bn year-over-year increase in property and equipment purchases. Meta's free cash flow of $784m, down from $8.55bn, did not turn negative but was exceeded by its own dividend. Microsoft remains the outlier, covering $35.8bn of property and equipment additions with $55.4bn of operating cash flow. Sources: company Q2 2026 earnings releases, 22–30 July 2026.

Two figures that are widely quoted and should not be. The first is Alphabet's “$84.75bn equity raise” of June 2026. The headline is Alphabet's own, but it aggregates $18.0bn of underwritten stock, $16.75bn of mandatory convertible preferred, a $10bn private placement including Berkshire Hathaway, and $40bn of potential proceeds from an at-the-market programme that Alphabet's own Q2 filing confirms it has not drawn on — and whose stated purpose is administering tax obligations on employee equity grants, not capital expenditure. Net proceeds actually received were $49.6bn. Quoting the $84.75bn overstates externally financed capex by roughly seventy per cent. The second is the widely circulated pairing of $433.9bn of hyperscaler property and equipment purchases against ~$149bn of depreciation. We could not source either figure to a filing or to named institutional research; both trace to analyst blogs that concede their own denominator mixes incompatible definitions. The same argument is available from a single primary document and is stronger for it: Alphabet's Q2 capital expenditure of $44.9bn against depreciation of property and equipment of $7.1bn is a ratio of 6.3×. Today's income statements carry a fraction of today's build-out. That is the depreciation wall, and it is arithmetic, not opinion.

Why this is a timing argument, not a demand argument. The bear case is frequently overstated as “AI capex has no revenue behind it.” The Q2 CY2026 disclosures do not support that. Microsoft's commercial remaining performance obligation reached $678bn, up 84% — contracted customer commitments, not forecasts. Cloud revenue accelerated in the same quarter capital expenditure peaked: AWS +37%, its fastest in roughly eighteen quarters; Azure +43%, ahead of its own guide; Google Cloud +82%. Both Amazon and Alphabet described themselves as supply-constrained, which is the opposite of an overbuild. FactSet's finding that incremental debt rose from 9% of capex in FY2024 to 32% on a trailing basis by mid-2026 is real, but the same analysis puts total debt to EBITDA at approximately 1× or below for four of the five. The honest bear case is therefore narrower and harder to dismiss: the depreciation schedule lands in 2027–2028 regardless of how well that backlog converts, and it lands on reported earnings rather than on cash. Sources: company filings; FactSet Insight, Kama & Rajput, 23 July 2026.

This analysis is updated each quarter as the hyperscalers report. The Portfolio Fragility Brief carries the revisions — what changed, what was wrong, and what it means for positioning.

The investment implication is dual-edged. Debt-funded hyperscaler capex raises the structural demand floor for AI infrastructure suppliers — contracts are signed, purchase orders placed, and delivery timelines locked in regardless of short-term sentiment shifts. However, rising leverage also introduces a new risk layer for hyperscaler equity positions themselves: if AI revenue monetisation disappoints, the debt service burden will suppress free cash flow precisely when investor patience is shortest. The debt burden is precisely why Microsoft (MSFT) and Alphabet (GOOGL) carry a Selective rather than High Conviction rating in A.L. Capital Advisory's framework — the infrastructure beneficiaries (NVDA, VRT, EQIX, CEG) capture the upside without carrying the balance sheet risk of the buyers.

In January 2025, OpenAI, SoftBank, and Oracle announced Project Stargate — a $500 billion AI infrastructure programme targeting US deployment, with an initial $100 billion committed within the first four years. Stargate represents a category of demand that does not appear in McKinsey's April 2025 base-case model: sovereign and government-adjacent AI infrastructure, funded at national-strategy scale rather than commercial return logic alone. By mid-2026 the programme was simultaneously advancing and re-sequencing: OpenAI, Oracle and SoftBank added new US sites (including a ~$16 billion campus in Saline, Michigan), while OpenAI walked away from a planned 600 MW Abilene expansion to hold out for next-generation NVIDIA silicon — a reminder that even sovereign-scale demand is now gated by chip-generation timing, not by capital.

Stargate is not an isolated event. Saudi Arabia's Public Investment Fund has committed to a $40 billion AI infrastructure programme. The UAE has established G42 as a sovereign AI entity with data centre commitments across three continents. The European Union's AI Continent Action Plan targets €200 billion in AI investment through 2030. Each of these programmes represents demand outside the hyperscaler-driven model — demand that is contractually committed, politically supported, and insensitive to short-term ROI calculations.

The structural implication for the AI capex cycle is clear: the demand floor is higher than McKinsey's base case assumed, because McKinsey's model was built on commercial hyperscaler logic alone. Sovereign AI programmes add a second, non-correlated demand layer. Morgan Stanley's projected US capacity shortfall — with Morgan Stanley projecting ~49 GW by 2028 — understates the total demand gap when sovereign programmes are included.

Among the five high-conviction positions, Constellation Energy (CEG) is the most direct Stargate beneficiary: nuclear baseload is the only power source that meets both the carbon-free and uninterruptible specifications that sovereign AI programmes require. Equinix (EQIX) benefits through its hyperscale campus footprint in the geographies where sovereign programmes are concentrating — Virginia, London, Singapore, and the Gulf. NVIDIA (NVDA) benefits from GPU procurement at sovereign scale. Vertiv (VRT) benefits through the thermal management requirements of the dense compute clusters Stargate's architecture demands.

McKinsey's analysis maps the $5.2 trillion AI capex envelope across five distinct investor archetypes. Understanding this architecture is essential: the investment case, risk profile, and return dynamics differ fundamentally across archetypes.

The $5.2–$6.7 trillion capex envelope flows through a defined set of public equities. But raw exposure to the AI theme is not sufficient — the archetype, moat, and balance sheet quality of each company determine whether they capture compounding returns or get crushed in the shake-out.

Every dollar of hyperscaler AI infrastructure capex ultimately flows through a semiconductor. The $660–690 billion committed for 2026 does not build itself — it must become physical chips, packaged onto boards, slotted into racks, and cooled. The supply chain is the root constraint on whether the cycle delivers McKinsey's projected $6.7 trillion of capacity by 2030. Three chokepoints define it: GPU allocation, advanced packaging at TSMC, and the EUV lithography monopoly held by ASML.

NVIDIA GPU Shortage — Why 90% Market Share Persists Despite Competition

NVIDIA Corporation (NASDAQ: NVDA) enters 2026 in a position with few historical precedents: a near-monopolist in a market growing at 36% annually, constrained not by demand but by its own supply chain. The Blackwell B200 architecture — 2.5× the inference throughput of the H100 at the same power envelope — carries reported order lead times of roughly 8–12 months (36–52 weeks, SemiAnalysis) for hyperscaler allocations as of May 2026. The binding constraint is CoWoS-L (Chip-on-Wafer-on-Substrate with Local) advanced packaging rather than TSMC fab capacity for the die itself: CoWoS capacity expands from approximately 40,000 wafers per month in 2024 (~75,000 in 2025) to an estimated ~130,000 by end of 2026 (TSMC CEO: "sold out through 2026"), and hyperscaler demand still tracks above the build rate.

NVIDIA's Vera Rubin architecture (H2 2026 full production) is confirmed, with the first rack already running at Microsoft Azure. Vera Rubin moves to HBM4 at 22 TB/s per GPU — 2.75× Blackwell — and delivers a 10× inference cost reduction, extending the CUDA platform advantage through the 2027–2028 cycle.

The CUDA software moat is as material as the hardware lead. NVIDIA's Compute Unified Device Architecture — the programming model underlying virtually every production AI training workload — has been in continuous development since 2007. The libraries (cuDNN, cuBLAS, TensorRT), the developer tooling, and two decades of academic and commercial code written natively for CUDA constitute a switching cost that Advanced Micro Devices (NASDAQ: AMD) is dismantling only slowly with ROCm. A.L. Capital Advisory estimates enterprise migration from CUDA to ROCm at current pace would require 3–5 years for non-latency-sensitive inference workloads, and meaningfully longer for training.

AMD MI300X vs NVIDIA B200 — Deep Technical and Commercial Comparison

AMD's MI300X is the most credible GPU challenger in the AI data center market. It integrates CPU and GPU chiplets in a unified HBM3 memory pool — 192GB of shared memory versus the H100's 80GB. For very large model inference (70B+ parameter LLMs) that capacity advantage is architecturally significant: models that would require tensor parallelism across eight H100s run on four MI300X units, cutting interconnect overhead. Microsoft Azure and Meta have both announced MI300X inference deployments, validating the commercial thesis.

The competitive gap nonetheless remains wide on three dimensions: software maturity (ROCm operator coverage against CUDA is estimated at 85–90% for inference, substantially lower for cutting-edge training kernels), supply chain reliability (TSMC allocates CoWoS capacity to NVIDIA first as the larger revenue customer), and ecosystem lock-in (the dominant MLOps toolchain — PyTorch, JAX, TensorFlow — all optimise natively for CUDA). A.L. Capital Advisory's base case: AMD captures 8–12% of the AI accelerator market by 2027, up from approximately 5–6% in 2025. At that share and current data center GPU ASPs ($25,000–$35,000 per unit wholesale), AMD Data Center revenue could reach $15–20 billion annually by FY2027 — material but sub-consensus.

Intel, ARM Architecture & the CPU Transition

Intel Corporation (NASDAQ: INTC) is pivoting from integrated device manufacturer to pure-play foundry (Intel Foundry Services) while defending its CPU franchise against AMD and the ARM architecture wave. CPUs matter at the margin in the AI data center: every GPU cluster needs host CPUs for ingestion, preprocessing, orchestration, and inference serving. Intel's Xeon Scalable 6th Generation (Granite Rapids) and AMD's EPYC Genoa compete for that socket, and EPYC has outgrown Intel in data center CPU share for three consecutive years — an estimated 33–35% of new server deployments versus Intel's 65%.

Arm Holdings plc (NASDAQ: ARM) is the deeper structural story. Arm's v9 architecture — licensed to Apple, Qualcomm (QCOM), Amazon (Graviton), Google (Axion), and NVDA (Grace CPU) — delivers 30–40% better performance-per-watt than x86 at comparable workloads. In the inference layer, where power efficiency determines cost per token, x86's dominance is structurally eroding. Arm earns a royalty on every chip shipped using its architecture — compounding leverage as ARM-based designs proliferate across data centers, edge infrastructure, and AI accelerators.

ASML — The Single-Point-of-Failure in Global AI Chip Supply

ASML Holding N.V. (NASDAQ: ASML) manufactures every extreme ultraviolet (EUV) lithography machine on Earth. There is no second supplier. EUV is required to pattern the sub-7nm transistors in every leading-edge AI chip — NVIDIA's B200 on TSMC N3E, AMD's MI300X on TSMC N5, Intel's Gaudi 3 on Intel 7. A single EUV tool costs approximately €200 million, weighs 180 tonnes, and requires 40 shipping containers and a dedicated Boeing 747 to transport. ASML ships approximately 50–60 EUV systems per year, on 18–24 month lead times from order to installation.

The investment case for ASML is the AI capex cycle expressed through the supply chain's deepest chokepoint. Every new fab built for AI demand — TSMC Arizona, Samsung Taylor, Intel Ohio — requires ASML EUV machines. The order book extends through 2027 and includes High-NA EUV tools (approximately €380 million per unit) required for sub-2nm nodes. No competitor has a functioning EUV tool; industry analysts estimate 15–20 years and multiple billions to build one from scratch. The Dutch government's US-pressured restriction of EUV exports to China removes the largest potential demand overhang and creates a China-exclusion premium benefiting Western fabs.

Custom Silicon — Broadcom & Marvell as the ASIC Layer

Custom silicon is one of the most consequential supply-chain developments of the 2026 cycle. Google (TPU v7 "Ironwood", GA April 2025 — 4.6 PFLOPS), Amazon (Trainium3 on TSMC 3nm, GA December 2025 — 2.5 PFLOPS FP8), Meta (MTIA 400), and Microsoft (Maia 200) are all deploying in-house AI accelerators manufactured at TSMC, explicitly to reduce dependence on NVIDIA and capture gross margin. They cannot design these chips unaided: Broadcom Inc. (NASDAQ: AVGO) and Marvell Technology Inc. (NASDAQ: MRVL) are the two dominant third-party custom ASIC architects for AI workloads.

Broadcom's AI semiconductor revenue reached $10.8 billion in Q2 FY2026 (+143% YoY, reported June 3, 2026), with AI bookings exceeding $30 billion in the quarter alone and Q3 FY2026 AI revenue guided to $16 billion (+200% YoY). Management reiterated a line of sight to more than $100 billion of AI revenue in 2027. Design wins now extend beyond Google, Meta, and Microsoft to OpenAI and Anthropic (confirmed early 2026), putting AVGO across an estimated 60% of hyperscaler custom silicon programmes. TrendForce shows custom ASIC shipments growing 44.6% in 2026 versus 16.1% for merchant GPUs; Goldman Sachs projects ASIC demand matching GPU demand by 2027, a roughly 50/50 split. Marvell's custom AI silicon is smaller but growing fast (data center now 75% of revenue, +63% YoY in Q1 FY2026), with wins at Amazon (Trainium) and Microsoft; its December 2025 acquisition of Celestial AI ($3.25B) extends it into optical photonic interconnect — a $6B TAM by 2030 as bandwidth demands exceed what copper sustains at 1.6T+ speeds. Both benefit from a dynamic NVIDIA cannot easily disrupt: hyperscalers have a strategic incentive to fund custom silicon as a CUDA countermeasure regardless of cost premium. The ASIC layer is no threat in training, where custom parts cannot match CUDA's flexibility — but it is an accelerating inference share shift AVGO and MRVL are positioned to capture.

Server Integration Layer — Super Micro Computer & Dell Technologies

Super Micro Computer Inc. (NASDAQ: SMCI) and Dell Technologies Inc. (NYSE: DELL) occupy the final assembly layer, converting raw silicon into deployable rack-scale systems. Super Micro's liquid-cooled DGX-H100 and MGX platforms are NVIDIA reference designs, and it has been first to market with new NVIDIA server generations three cycles running. Dell's PowerEdge XE9680 competes for the same enterprise and colocation buyer, with the added advantage of a global direct sales force and ProSupport services.

Both are gated by NVIDIA's GPU allocation — when B200 supply is constrained, SMCI and DELL backlog builds while gross margins compress on pre-sold orders. The risk is margin: server integration earns 5–8% EBIT at SMCI and 4–6% on AI infrastructure servers at Dell, and price competition intensifies during shortages. The structural opportunity is the move from Phase 1 (Build) to Phase 2 (Deploy), when hyperscalers shift from direct NVIDIA procurement to third-party integrators.

The binding constraint on NVIDIA's Blackwell B200 output in Q1–Q2 2026 is not the GPU die. It is High Bandwidth Memory 3E (HBM3E) — the stacked DRAM that separates a usable AI chip from a theoretical one. A single B200 die requires six HBM3E stacks of approximately 8GB each, 192GB per chip. At current shipment volumes, the global HBM market must produce and package more advanced memory in 2026 than in all previous years combined. There are exactly three HBM suppliers on Earth: SK Hynix, Samsung, and Micron Technology Inc. (NASDAQ: MU) — the smallest of the three, and the most investable for Western investors.

HBM3E — The Architecture of Scarcity

High Bandwidth Memory is not DRAM as conventionally understood. DDR5 transmits serially over a narrow bus; HBM stacks 8–12 DRAM dies vertically using through-silicon vias, achieving 1.2 TB/s per stack versus DDR5's 0.1 TB/s per channel. For AI training — feeding terabytes of weights and activations per second to thousands of CUDA cores — that bandwidth is the ceiling on throughput. Running a 70B parameter training job on HBM3E rather than GDDR6 is roughly 4–5× faster, which at hyperscaler compute costs is tens of millions of dollars per run.

HBM production also requires TSMC (TSM) CoWoS packaging to integrate the stacks with the GPU die — a circular dependency, since GPU and memory compete for the same scarce capacity. SK Hynix holds an estimated 50–55% of the HBM market, Samsung 35–40%, and Micron 8–12% and ramping. NVIDIA has qualified all three for HBM3E, but SK Hynix retains a generation lead: Hynix began HBM3E production in Q3 2024, Micron qualified in Q4 2024, and Samsung's yield issues delayed qualification into Q1 2026 — leaving Micron a qualified second supplier in a market where the leader cannot meet demand and the third player has quality problems.

Micron Technology — The High-Conviction AI Memory Investment Case

Micron Technology Inc. (NASDAQ: MU) is A.L. Capital Advisory's fifth High Conviction position, and the one with the widest gap between consensus expectations and structural opportunity. Three independent pillars compound through 2028.

First, the HBM revenue inflection — now confirmed. In fiscal Q3 2026 (reported June 24, 2026), Micron posted record revenue of $41.5 billion (+346% YoY) at a record ~85% non-GAAP gross margin, guided fiscal Q4 to a record ~$50 billion, and signed 16 multi-year take-or-pay Strategic Customer Agreements covering roughly half of future revenue — lifting market capitalisation past $1 trillion. What was a forward thesis a year ago is a reported inflection: HBM3E is sold out, HBM4 is ramping, and HBM has moved from a negligible FY2024 line to Micron's primary growth engine. At SK Hynix's disclosed HBM gross margins (50–55%), that is transformative against a blended profile historically averaging 25–35% across the DRAM/NAND cycle. A single HBM3E 8-Hi stack now carries a blended ASP of approximately $30–40 per GB in 2026 (HBM contract prices raised ~20% amid broad memory “memflation”).

Second, the DRAM pricing cycle. The 2022–2023 oversupply has fully resolved. All three major producers have deliberately constrained bit output growth, redirecting capital expenditure into HBM — capacity that becomes unavailable for standard DRAM. Server DDR5 contract prices are up ~50% quarter-on-quarter (“memflation”) as data center demand — each AI rack carries $100,000–$400,000 of standard DRAM alongside the GPU — grows faster than supply can respond. Micron captures both the volume increase and the ASP uplift.

Third, the NAND recovery. Western investors often model Micron as a pure DRAM company, but NAND flash — SSDs, storage arrays, training data pipelines — is approximately 35% of revenue. The cycle bottomed in Q3 2023 at prices that pushed every producer, including Western Digital (NASDAQ: WDC), into EBITDA-negative operations. Enterprise SSD pricing has since recovered 60–80% from the trough as AI training datasets, model checkpoints, and inference caches drive unprecedented enterprise NVMe demand. Western Digital is the pure-play expression of that recovery; Micron is the more complete AI memory story.

The Memory Wall — AI's Hidden Bottleneck

The "memory wall" is the gap between the growth rate of AI model complexity — parameters, context window, batch size — and the growth rate of memory bandwidth. A GPT-4-scale model feeds approximately 140GB of weights to GPU cores each forward pass; at 1.2 TB/s per stack and six stacks per B200, peak sustainable throughput is approximately 7.2 TB/s per GPU. Projected 2026–2027 training runs (1–10 trillion parameter models) would need 8–12 stacks per GPU, beyond the current B200 architecture. Every new generation — Rubin/R100, AMD's MI400 — therefore requires more HBM, more advanced packaging, and more bandwidth: a structural demand escalator for Micron, SK Hynix, and the HBM ecosystem.

Per-rack memory content shows the scale. One NVIDIA DGX H100 system (8× H100) holds 640GB of HBM2e, 2TB of DDR5 system DRAM, and 30TB of NVMe SSD — an estimated $80,000–$120,000 of the system cost (A.L. Capital Advisory estimate; SemiAnalysis teardowns imply the lower end), comparable to a single H100 GPU. Each gigawatt of AI data center capacity contains roughly 10,000 racks, so 10,000 × $120,000 implies $1.2 billion of memory spend per gigawatt. Against the 49 GW US shortfall Morgan Stanley projects, that is $48 billion of cumulative incremental memory demand — before Europe, Asia, and Project Stargate are counted.

Geopolitical Risk — China Memory and Export Controls

Micron's most significant risk is regulatory. China's Cyberspace Administration banned Micron products from "critical information infrastructure" operators in May 2023, retaliating against US restrictions on Chinese advanced chip imports. China represented approximately 16% of Micron's FY2023 revenue. Redirecting supply to India, Southeast Asia, and European data center expansion has partially absorbed the impact, but approximately $800M–$1.2B of annualised revenue remains displaced. Resolution — or further escalation — of the US-China technology trade war is the key binary risk in the Micron case.

Chinese domestic memory is a structural headwind, but further from competitive parity than commonly believed. CXMT (ChangXin Memory Technologies), China's domestic DRAM producer, is shipping DDR4 and early DDR5 at an estimated 25–35% yield against an industry standard of 85–90%. Yangtze Memory (YMTC) reached 232-layer NAND in 2023 but faces ASML EUV import restrictions that will prevent progression to the sub-10nm nodes next-generation HBM requires. China is not a 2026 threat to Micron's HBM franchise; it is a 2029–2032 risk to commodity NAND share.

Portfolio Construction Framework — Five principles for building the AI infrastructure position without getting burned:

The AI capex cycle is not a theme. The AI capex cycle is a decade-long structural reallocation of capital — from consumption to physical infrastructure — at a scale not seen since the electrification of the United States economy in the 1920s. Q1 2026 earnings confirmed the Big-5 hyperscalers will spend approximately $775–800 billion on AI infrastructure in 2026 alone, nearly tripling the ~$238 billion deployed in 2024. Goldman Sachs projects US data centre power demand rising from 41 GW in 2026 to 66 GW in 2027; Morgan Stanley projects it growing to ~49 GW by 2028. North American colocation vacancy has tightened to ~1.4% (JLL/CBRE, YE 2025) — a level at which pricing power is structural and durable.

This paper has mapped the full capital flow across the cycle's three phases. In Phase 1 (Build, 2024–2026), semiconductor procurement and thermal management are the primary value-capture layer — hence NVIDIA and Vertiv as peak-conviction positions. In Phase 2 (Deploy, 2026–2028), contracted infrastructure operators with physical scarcity moats begin to compound: Equinix's interconnect estate and Constellation Energy's 20-year nuclear power purchase agreements. In Phase 3 (Compound, 2028–2030), all five positions benefit simultaneously as AI-driven cloud revenue scales and contracted revenue compounds across multi-year agreements. A static "AI basket" allocation misses the Phase 2–3 rotation entirely.

The structural bull case rests on three pillars that the bear case cannot dislodge without a fundamental change in physical reality: construction lead times of 18–30 months mean supply cannot respond to short-term sentiment shifts; AI accelerator refresh cycles of 3–4 years mean overcapacity converts to obsolescence faster than it becomes stranded; and the contracts-first structure of the build-out means virtually every dollar of hyperscaler guidance is committed before a shovel enters the ground. These are not financial projections — they are engineering constraints.

The honest risks are equally structural. If enterprise AI deployment fails to materialise at scale by 2027, the demand curve reverts to the constrained scenario ($3.7T vs $5.2T in AI-specific capex). If semiconductor efficiency gains (in the tradition of DeepSeek V3) suppress training demand, NVIDIA's backlog clears faster than consensus models. If hyperscaler ROI discipline breaks down under debt pressure, the capex/revenue ratio remains above 50% indefinitely — compressing free cash flow precisely when patient capital needs a return. These risks are real. The risks are precisely why A.L. Capital Advisory distinguishes High Conviction infrastructure suppliers (NVDA, VRT, EQIX, CEG) from Selective hyperscaler equity positions (MSFT, GOOGL) — infrastructure suppliers are paid regardless of which ROI scenario resolves; hyperscaler equity positions are not.

The monitoring framework is straightforward: vacancy below 3% is healthy; above 6% is the early warning. Capex/revenue ratios declining from 2026 onwards signal the ROI inflection the market is waiting for. Enterprise AI deployment rate — currently in the early phase — is the key leading indicator for whether the McKinsey base case ($6.7T by 2030) is achieved or exceeded. A.L. Capital Advisory updates these readings quarterly.