Amazon plans more than $200 billion in AI spending

Amazon plans more than $200 billion in AI spending: the analysis positions NVDA, AVGO and MU to benefit and AMZN, MSFT and CRWV under pressure; confidence high.

Analyzed 2026-09-09 — point-in-time, never revised. Outcomes are scored against SPY and published either way.

Ranked exposures

▲ NVDA Nvidia — impact +88 · confidence high · horizon quarters · 1st order
Amazon's expanding AI clusters increase NVIDIA accelerator orders, driving data center revenue through sustained premium GPU pricing.
Chain: Amazon commits over 200 billion to AI capex → most spending targets accelerated compute racks → AWS places larger GPU cluster orders → Nvidia data center revenue and pricing hold firm
What would break the call: AWS discloses a majority shift of new capacity to Trainium, cutting Nvidia order volumes.
▲ AVGO Broadcom — impact +82 · confidence high · horizon quarters · 1st order
Amazon's faster AI cluster deployments increase Broadcom custom accelerator and networking volumes, expanding chip revenue.
Chain: Amazon funds custom accelerator programs at scale → Broadcom supplies ASIC design and packaging → cluster scale-out requires Tomahawk switching silicon → Broadcom AI semiconductor revenue rises
What would break the call: Amazon moves custom silicon design to Marvell or Alchip, removing Broadcom accelerator content.
▲ MU Micron — impact +78 · confidence high · horizon quarters · 2nd order
Amazon's AI server expansion tightens high-bandwidth memory demand, raising Micron's contract prices and memory margins.
Chain: AI server builds expand at Amazon → each accelerator needs high bandwidth memory stacks → HBM and server DRAM supply tightens → Micron contract prices and margins rise
What would break the call: Samsung qualifies large HBM volumes at low prices, breaking the contract price uptrend.
▲ VRT Vertiv — impact +76 · confidence medium · horizon weeks · 2nd order
Amazon's denser AI facilities increase Vertiv orders for power and liquid-cooling equipment, expanding its backlog.
Chain: Amazon builds denser AI data halls → rack power density exceeds air cooling limits → operators order liquid cooling and power distribution → Vertiv orders and backlog expand
What would break the call: Amazon standardizes on in-house cooling designs sourced directly from contract manufacturers.
▲ ETN Eaton — impact +69 · confidence medium · horizon weeks · 2nd order
Amazon's expanding data centers drive Eaton switchgear and transformer orders, extending its electrical backlog at firm pricing.
Chain: new data center campuses need electrical infrastructure → switchgear and transformers are long lead items → Eaton wins content per megawatt → electrical segment backlog and pricing stay firm
What would break the call: Data center project delays from grid interconnection queues push electrical orders past 2027.
▲ MRVL Marvell — impact +68 · confidence low · horizon quarters · 1st order
Marvell earns custom silicon and optical interconnect revenue from Trainium programs, raising data center content per rack.
Chain: Amazon expands Trainium program volumes → Marvell provides custom silicon and optical DSPs → interconnect content per rack increases → Marvell data center revenue grows
What would break the call: Amazon awards next generation Trainium interconnect to Broadcom, cutting Marvell content.
▲ AMD AMD — impact +65 · confidence low · horizon quarters · 1st order
AMD chip orders rise after Amazon seeks diverse suppliers for its massive AI computing expansion.
Chain: Amazon seeks accelerator supplier diversity → AWS qualifies Instinct GPUs for inference → order volumes shift partly to AMD → AMD data center GPU revenue rises
What would break the call: AWS declines to offer Instinct instances at scale, leaving AMD share negligible.
▲ ANET Arista Networks — impact +64 · confidence medium · horizon weeks · 2nd order
Amazon's larger AI training clusters increase demand for Arista's high-bandwidth Ethernet switches and back-end fabric ports.
Chain: larger training clusters need back-end fabrics → Ethernet ports scale with accelerator count → Amazon remains a major Arista customer → Arista switching revenue rises
What would break the call: Amazon shifts fabric purchases to internally designed white box switches.
▲ SMCI Super Micro — impact +64 · confidence medium · horizon weeks · 2nd order
Amazon's larger AI infrastructure purchases increase Super Micro's GPU server platform volumes.
Chain: Amazon increases AI server unit purchases → some rack integration is outsourced → Super Micro ships GPU platforms → server revenue volumes increase
What would break the call: AWS keeps rack integration with ODM partners in Taiwan, bypassing Super Micro entirely.
▼ AMZN Amazon — impact -63 · confidence medium · horizon immediate · 1st order
Amazon's larger AI capital budget increases depreciation and interest expenses before new AWS capacity generates revenue, reducing free cash flow.
Chain: capital budget exceeds 200 billion → assets enter service and start depreciating → debt issuance raises interest expense → free cash flow falls before revenue arrives
What would break the call: AWS revenue acceleration and backlog conversion outrun depreciation, expanding operating margin.
▲ TSM TSMC — impact +61 · confidence high · horizon quarters · 2nd order
Amazon's AI chip demand keeps TSMC's leading-edge wafer and advanced-packaging capacity highly utilized.
Chain: Amazon orders accelerators from Nvidia and partners → all designs need leading edge wafers → advanced packaging capacity stays booked → TSMC utilization and pricing hold high
What would break the call: A broad AI order digestion cycle leaves leading edge capacity underutilized.
▲ CEG Constellation Energy — impact +60 · confidence low · horizon quarters · 2nd order
Constellation signs long dated nuclear supply contracts with hyperscalers, converting spare output into higher realized power prices.
Chain: AI campuses need firm carbon free power → hyperscalers sign long dated supply contracts → Constellation locks premium nuclear pricing → realized power prices and earnings rise
What would break the call: Regulators block behind the meter deals, forcing sales at merchant market prices.
▲ GEV GE Vernova — impact +59 · confidence medium · horizon quarters · 3rd order
Amazon's rising AI data-center electricity demand increases GE Vernova orders for turbines and grid equipment.
Chain: data center load growth strains regional grids → utilities order new generation and grid gear → GE Vernova books turbines and transformers → equipment revenue and backlog margins improve
What would break the call: Utilities defer generation orders after load forecasts for data centers are cut.
▼ MSFT Microsoft — impact -56 · confidence medium · horizon quarters · 2nd order
Amazon's expanded AWS AI capacity intensifies cloud pricing competition, slowing Microsoft's Azure AI sales growth.
Chain: AWS adds large AI capacity → supply of cloud AI compute increases → enterprise buyers gain pricing leverage → Azure AI revenue growth slows
What would break the call: Azure AI demand stays supply constrained with backlog growing faster than capacity.
▼ CRWV CoreWeave — impact -55 · confidence medium · horizon weeks · 2nd order
Amazon's self-funded capacity expansion intensifies competition for CoreWeave's GPU cloud customers, pressuring rental pricing.
Chain: Amazon self funds massive GPU capacity → neocloud rental supply faces bigger rival → GPU hourly rates come under pressure → CoreWeave contract pricing and margins compress
What would break the call: CoreWeave signs new multiyear take or pay contracts at stable rates.
▼ GOOGL Google — impact -49 · confidence medium · horizon quarters · 2nd order
Amazon's expanded AWS compute capacity captures enterprise AI workloads from Google Cloud, reducing Alphabet's cloud revenue growth.
Chain: AWS expands AI capacity and instance choice → enterprise AI workloads shift toward AWS → Google Cloud win rates slip → Alphabet cloud revenue growth moderates
What would break the call: Google Cloud backlog and Gemini adoption accelerate, showing no share loss to AWS.
▲ PWR Quanta Services — impact +45 · confidence low · horizon quarters · 3rd order
Data center campuses require transmission and substation construction, lifting Quanta electric infrastructure backlog and skilled labor pricing.
Chain: Amazon campuses need new interconnection capacity → utilities contract transmission and substation buildouts → Quanta wins electric infrastructure work → backlog and revenue per crew rise
What would break the call: Interconnection permitting delays freeze utility capital plans and push transmission awards out.
▼ NBIS Nebius — impact -44 · confidence medium · horizon weeks · 3rd order
Amazon's global AI capacity expansion raises Nebius's accelerator costs by intensifying competition for scarce GPUs.
Chain: Amazon absorbs scarce accelerator supply → GPU allocation tightens for smaller buyers → Nebius pays more per accelerator → capacity costs rise and margins compress
What would break the call: Nvidia supply loosens and Nebius secures allocation at declining unit costs.
▲ FIX Comfort Systems USA — impact +44 · confidence low · horizon quarters · 3rd order
Mechanical contracting for cooling systems in new hyperscale builds expands Comfort Systems backlog and project margins.
Chain: Amazon accelerates data center construction → builders subcontract mechanical and cooling installation → Comfort Systems books technology sector projects → backlog and gross margin rise
What would break the call: Amazon slows construction starts, causing technology segment backlog to decline sequentially.
▼ ORCL Oracle — impact -42 · confidence low · horizon quarters · 2nd order
Oracle Cloud faces pricing pressure for AI infrastructure when AWS adds capacity faster.
Chain: AWS adds AI infrastructure capacity quickly → OCI faces broader competing supply → AI infrastructure pricing softens → Oracle cloud margins and RPO quality weaken
What would break the call: Oracle reports rising OCI backlog with stable cloud infrastructure gross margin.
▲ ALAB Astera Labs — impact +42 · confidence low · horizon quarters · 2nd order
Rack scale accelerator deployments increase demand for Astera connectivity chips inside AWS custom silicon systems.
Chain: Amazon scales custom accelerator racks → PCIe and Ethernet signal integrity limits bind → Astera retimers and fabric switches ship → content per rack revenue grows
What would break the call: Broadcom integrates retimer functions on switch silicon, removing standalone connectivity sockets.
▼ DLR Digital Realty — impact -40 · confidence low · horizon quarters · 2nd order
Digital Realty faces lower Amazon colocation demand when internal data center construction replaces leased capacity.
Chain: Amazon shifts toward self build campuses → leased colocation requirements grow more slowly → hyperscale leasing pipeline narrows for landlords → Digital Realty releasing spreads soften
What would break the call: Digital Realty signs record hyperscale leasing bookings with rising cash rent spreads.
▼ INTC Intel — impact -40 · confidence low · horizon quarters · 2nd order
Intel server chip volume decreases because Amazon prefers GPUs and custom silicon for AI workloads.
Chain: Amazon capex concentrates on accelerators → general purpose server budget share shrinks → Graviton covers remaining CPU needs → Intel data center unit volumes decline
What would break the call: AWS expands Xeon instance fleets or awards Intel Foundry a large accelerator contract.
▲ ARM Arm — impact +40 · confidence low · horizon quarters
Arm royalty payments accelerate after Amazon designs additional custom processors around its core architecture.

Sources

What happened

Horizon	Verdict	Top pick	Excess vs SPY
1 trading days	Wrong	NVDA	-0.4% vs SPY
5 trading days	Wrong	NVDA	-4.9% vs SPY

Related scenarios

Stocks in this scenario

NVDA · AVGO · MU · VRT · ETN · MRVL · AMD · ANET · SMCI · AMZN · TSM · CEG · GEV · MSFT · CRWV · GOOGL · PWR · NBIS · FIX · ORCL · ALAB · DLR · INTC · ARM · COHR · EQIX · HPQ · CSCO · CCJ · MARA · RIOT · CENX · IBM · NTAP · NUE

News events · All scenarios · Pricing · Track record · Methodology · Map your own scenario