Interconnection scale, open-weight AI, and sovereign computing strengthen EQIX's data center advantage

Overview

The report analyzes four debated issues: open-weight versus closed-weight AI models, interconnection monetization, data center construction restrictions, and sovereign AI. Morgan Stanley believes colocation data centers remain well positioned under multiple scenarios, with EQIX emerging as the top pick due to its diversified enterprise customer base, global facility network, and the industry's largest interconnection ecosystem.

Core views

The report first concludes that colocation data centers do not depend on a single model or chip prevailing as AI infrastructure evolves. If open-weight models continue to proliferate, enterprises can fine-tune and deploy models in private facilities and allocate tasks across proprietary data, multiple clouds, model providers, and neocloud service providers. This would make inference architectures more distributed, increase the number of endpoints, and strengthen demand for private, low-latency connections, making incremental demand more favorable to retail colocation and EQIX, which leads in interconnection network scale. The report cites enterprise use cases to illustrate the economics and flexibility of open models: Pinterest stated that the associated transaction costs were less than 8% of those for comparable closed proprietary models, while Revolve, Spotify, and Duolingo are also expanding their use of open models or multi-model routing. Drawing on the Jevons paradox, the report argues that greater efficiency and lower token prices may accelerate AI adoption rather than necessarily weaken aggregate demand for computing power and infrastructure. If closed-weight frontier models retain a high share, inference workloads are more likely to concentrate in large, high-density environments controlled by frontier laboratories and hyperscale cloud operators, resulting in fewer but larger deployments that better fit DLR's and BXDC's large-block wholesale leasing models. The report therefore believes that a higher share of open-weight models is relatively favorable to EQIX, while closed-weight dominance is relatively favorable to DLR and BXDC. However, operators are not limited to benefiting from only one outcome: EQIX can still participate in hyperscale demand through xScale and cloud access, DLR also has retail and interconnection businesses, and BXDC has not yet acquired any data centers, with its current strategy primarily targeting wholesale leasing to hyperscale customers. The second core thesis is interconnection scale and its not-yet-fully-realized pricing potential. Interconnection services directly connect tenants within the same or different data centers through fiber or software platforms. The larger the network, the stronger the network effects, customer stickiness, and migration barriers. EQIX has 522,700 interconnection connections, far above DLR's 235,500 and CSQR's 36,000. The report also notes that EQIX's interconnection volume is more than twice DLR's and expects the gap to continue widening. Current physical cross-connects typically charge monthly fees based on the number of connections, while virtual interconnection services such as Fabric are primarily billed based on provisioned 1, 10, or 100 Gbps ports and bandwidth rather than actual data traffic. As AI agents drive growth in machine traffic and virtual connections gain share, the report believes EQIX may gradually capture more value based on actual data volume or throughput. The report decomposes interconnection revenue into “billable cabinets × connections per cabinet × monthly recurring revenue per connection.” The base case projects that EQIX will add approximately 22,000 and 30,000 billable cabinets in 2027 and 2028, respectively; connections per cabinet will grow by 1.6% and 2.1%, respectively; and MRR per connection will grow by 3.8% and 3.9%, respectively. The bull case further assumes that the pricing mechanism can capture a share of approximately 25% compound growth in data traffic, raising the 2027 interconnection revenue growth forecast from 14.5% to 16.6% and the 2028 forecast from 16.3% to 16.5%. Assuming a 90% incremental margin on interconnection revenue, every 100-basis-point increase in interconnection pricing growth could provide approximately 1.6% upside to the current 2030 AFFO/share forecast. Cloudflare's CEO's comment that non-human traffic could reach 1,000 times human traffic in five years is used by the report as supporting evidence for machine traffic growth and the opportunity for traffic-based pricing, rather than as evidence that EQIX has already implemented this pricing model. The third thesis is that development restrictions have a two-sided impact on the industry. Concerns are rising at the local level in the US regarding electricity affordability, water consumption and pollution, as well as noise, dust, and traffic. Texas is reviewing more than 474 GW of pending applications in ERCOT's interconnection process, approximately 90% of which are related to data centers. Virginia's July 31, 2026 Rider T1 order advances a “cost-causer pays” framework for direct-connect transmission, while related proposals in Ohio remain election issues and sentiment indicators rather than current law. Development yields may come under pressure if operators must bear more grid infrastructure costs. Using a conservative 25% stabilized gross profit yield, the report believes that even if this is below the historical average, EQIX's recurring revenue can still maintain double-digit year-over-year growth. On the other hand, development moratoriums, permitting delays, and power shortages constrain new supply and increase the scarcity and pricing power of existing powered assets. Historical experience in Northern Virginia shows that insufficient incremental power reduced vacancy rates and drove prices up 118% over five years. When vacancy fell below 2% in 2021, wholesale colocation price growth accelerated materially. The report's supply-demand analysis of six markets also shows that demand began exceeding newly commissioned power in 2021, coinciding with falling vacancy rates and rising prices. Political scrutiny is therefore a downside scenario for future development and growth, but may also represent a bullish pricing scenario for incumbents with operational powered capacity. The fourth thesis is that sovereign AI will increase demand for localized, redundant, and jurisdiction-specific infrastructure. Governments and enterprises are placing greater emphasis on data residency, control of sensitive data, and the location of computing and network routing, meaning infrastructure can no longer be fully interchangeable globally. EQIX, DLR, and CSQR provide power, cooling, space, security, and interconnection, while customers control equipment and workloads. Operators therefore benefit more directly from requirements governing where computing must be located and how it must connect, without having to bet on the ultimately successful chip, model, or software. The report believes EQIX is best positioned as sovereignty requirements expand from data storage to processing location and transmission paths; its Fabric Geo Zones can restrict traffic to specific countries. EQIX has approximately 280 data centers globally, including 6 IBXs in Hong Kong and 5 partner-operated sites in Shanghai. Deeper US-China decoupling would increase regulatory complexity, but could also reinforce demand for duplicated, jurisdiction-specific facilities. Based on these combined theses, Morgan Stanley maintains its Overweight view on EQIX and reiterates its status as the industry top pick and its $1,250 price target. The risk-reward framework uses forward adjusted EBITDA multiples of 29x, 26.1x, and 20.9x for the bull, base, and bear cases, respectively. The report forecasts EQIX total revenue growth of 11.6%, 12.1%, and 12.5% year over year in 2026–2028, respectively; AFFO/share growth of 12.6%, 11.2%, and 9.8%, respectively; and adjusted EBITDA growth of 16.3%, 13.8%, and 13.3%, respectively. Corresponding capital expenditures are forecast at $5.823 billion, $5.997 billion, and $6.363 billion.

Analysis framework

The report first constructs three AI architecture scenarios—open-weight, closed-weight, and hybrid models—to assess how different deployment approaches affect retail colocation, wholesale leasing, and interconnection demand. It then decomposes interconnection revenue into cabinet count, connections per cabinet, and MRR per connection, and conducts sensitivity analysis on pricing growth and AFFO/share. The report also overlays 451 Research's data center location data as of the first quarter of 2026 with construction-restriction tracking data as of August 18, 2026, and combines this with the historical supply, absorption, vacancy, and pricing data of six US markets to assess the two-sided impact of regulatory restrictions. Finally, it constructs EQIX risk-reward scenarios using adjusted EBITDA multiples.

Methodology notes

• Industry/sector analysis frameworkSupply-demand framework

  Data center power supply, demand absorption, vacancy, and pricing analysis

  The report compares newly commissioned power with demand absorption and examines changes in vacancy and pricing to illustrate how construction restrictions and power constraints compress supply and strengthen the pricing power of existing assets.

• Industry/sector analysis frameworkVolume-price decomposition

  Three-factor decomposition of interconnection revenue

  The report decomposes interconnection revenue into billable cabinets, connections per cabinet, and MRR per connection to distinguish the contributions of capacity, utilization, and pricing to revenue growth.

• Competition and strategy frameworkMoat / competitive advantage

  Interconnection network effects

  A larger interconnection network increases connection value and customer stickiness, creating a positive growth flywheel. The report uses this to explain why EQIX's scale constitutes a competitive barrier.

• Event game theory and behavioral financeEvent-driven analysis

  Scenarios involving construction restrictions, permitting, and grid cost policies

  The report analyzes the Texas review, Virginia's cost-allocation rules, and local political pressure, assessing their respective effects on development yields, incremental supply, and the pricing of existing assets.

• Valuation methodologyEV/EBITDA valuation

  Forward adjusted EBITDA multiple scenario valuation

  EQIX's risk-reward analysis uses forward adjusted EBITDA multiples of 29x, 26.1x, and 20.9x for the bull, base, and bear cases, respectively.

• (Out-of-vocabulary methodology)

  Interconnection pricing sensitivity analysis

  Holding base-case connection volumes constant and assuming a 90% incremental margin on interconnection revenue, the report estimates the impact of changes in MRR growth on the 2030 AFFO/share forecast.

• (Out-of-vocabulary methodology)

  Jevons paradox

  The report uses this logic to explain that improved model efficiency and lower token prices may expand AI usage, thereby supporting rather than necessarily weakening aggregate infrastructure demand.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Equinix (EQIX.O)

  The report's top-pick retail colocation and interconnection operator, whose business model is well aligned with open-weight AI, virtual interconnection traffic growth, and sovereign AI.

  Strengths

  Diversified enterprise customer base, global footprint, and the industry's largest interconnection network, with the ability to participate in hyperscale demand through xScale.

  Weaknesses

  Expansion may require it to bear more grid infrastructure costs, while its exposure to China also increases cross-border regulatory complexity.

  Comparison

  Has 522,700 interconnection connections, significantly more than DLR's 235,500 and CSQR's 36,000.

  Risks

  Political scrutiny, permitting restrictions, grid costs, declining development yields, and increased sovereign AI compliance obligations.

• Digital Realty Trust (DLR.N)

  If closed-weight models dominate, large, high-density deployments would better align with its hyperscale customer exposure and large-block wholesale leasing model.

  Strengths

  Has a hyperscale business as well as significant retail and interconnection operations.

  Weaknesses

  Would benefit less than EQIX if open-weight models drive distributed enterprise deployments.

  Comparison

  Has 235,500 interconnection connections, below EQIX's 522,700; the report assigns an Equal-weight rating and a $215 price target.

  Risks

  Development restrictions, grid costs, and a shift in model deployment architecture toward retail interconnection.

• Csquare (CSQR.N)

  As a colocation operator, it can benefit from sovereign AI demand for localized computing, redundancy, and jurisdiction-specific infrastructure.

  Strengths

  Directly provides power, cooling, space, security, and interconnection infrastructure.

  Weaknesses

  Its interconnection scale is only 36,000 connections, materially smaller than EQIX and DLR.

  Comparison

  The report lists a $26 price target; its interconnection scale is below EQIX's 522,700 and DLR's 235,500.

  Risks

  Smaller interconnection scale and rising development and compliance costs.

• Blackstone Digital Infrastructure Trust (BXDC.N)

  If closed-weight models retain a high share, centralized, large-scale computing deployments would be more favorable to its wholesale strategy targeting hyperscale tenants.

  Strengths

  Its target business model aligns with demand for large, high-density, hyperscale deployments.

  Weaknesses

  The report notes that it has not yet acquired any data centers.

  Comparison

  Relatively benefits under the closed-weight scenario, but is less favored than EQIX under open-weight and distributed enterprise deployment scenarios.

  Risks

  It has not yet acquired data center assets, and execution of its strategy depends on subsequent acquisitions and wholesale leasing demand.

Key data

• EQIX price target and ratingOverweight, $1,250The report reiterates EQIX as the top pick; its price on August 19, 2026 was $1,077.08
• Interconnection connection scaleEQIX 522,700; DLR 235,500; CSQR 36,000EQIX is the industry's largest interconnection provider, with more than twice as many connections as DLR
• EQIX bull-case interconnection revenue growth16.6% in 2027; 16.5% in 2028Previous forecasts were 14.5% and 16.3%, respectively
• Incremental billable cabinets in the base caseApproximately 22,000 in 2027; approximately 30,000 in 2028Capacity driver of interconnection revenue
• Growth in connections per cabinet1.6% in 2027; 2.1% in 2028EQIX base case
• Growth in MRR per interconnection3.8% in 2027; 3.9% in 2028EQIX base case
• Interconnection pricing sensitivityFor every 100 bps increase, AFFO/share rises approximately 1.6% by 2030Assumes a 90% incremental margin on interconnection revenue and base-case connection volumes
• Global data generation growthApproximately 25% through 2029IDC estimate; the report also uses approximately 25% compound growth when discussing the traffic-based pricing opportunity
• Pending grid interconnection applications in TexasMore than 474 GWApproximately 90% are related to data centers
• Northern Virginia price changeUp 118% over five yearsPower constraints lowered vacancy; wholesale colocation price growth accelerated after vacancy fell below 2% in 2021
• Conservative stabilized gross profit yield assumption25%Even below the historical average, the report still expects EQIX revenue to maintain double-digit year-over-year growth
• EQIX global facility scaleApproximately 280 data centers globallyIncluding 6 IBXs in Hong Kong and 5 partner-operated sites in Shanghai
• EQIX total revenue year-over-year growth forecast11.6% in 2026; 12.1% in 2027; 12.5% in 2028Key earnings assumption in the report
• EQIX AFFO/share year-over-year growth forecast12.6% in 2026; 11.2% in 2027; 9.8% in 2028Key earnings assumption in the report
• EQIX adjusted EBITDA year-over-year growth forecast16.3% in 2026; 13.8% in 2027; 13.3% in 2028Key earnings assumption in the report
• EQIX capital expenditure forecast$5,823 million in 2026; $5,997 million in 2027; $6,363 million in 2028Investment required for continued expansion

Impact & implications

The report believes colocation data centers can participate in demand growth whether AI deployment favors open or closed models, but the structure of benefits differs. Open-weight models and sovereign AI place greater emphasis on enterprise control, distributed deployment, and interconnection, making them most favorable to EQIX; closed-weight dominance would relatively favor DLR's and BXDC's hyperscale wholesale models. Regulation and construction restrictions raise development costs, but may also increase the scarcity and pricing of existing powered assets by limiting new supply. If EQIX can expand virtual interconnection pricing from fixed connection fees to actual traffic, its high incremental margins could translate into upside to AFFO/share.

Risks

• Operators may be required to bear more grid infrastructure costs, compressing data center development yields.
• Construction restrictions, permitting delays, and the politicization of data centers may constrain future project development and revenue growth.
• If sovereign AI rules expand to overseas computing, customer verification, cybersecurity, and routing, operators' compliance obligations and capacity-allocation constraints will increase.
• If closed-weight models dominate over the long term, more incremental demand may flow to DLR's and BXDC's hyperscale wholesale models rather than EQIX's retail interconnection model.
• Deeper US-China infrastructure decoupling may increase the regulatory complexity of EQIX's cross-jurisdiction operations.
• Charging based on actual traffic remains a bull-case assumption, and EQIX may not succeed in changing its interconnection pricing mechanism.

What to watch

• Monitor changes in the shares of open-weight, closed-weight, and hybrid models in enterprise AI deployments.
• Track EQIX's virtual interconnection mix and whether pricing shifts from fixed connection fees to actual traffic or throughput.
• Observe whether MRR per interconnection, connections per cabinet, and billable cabinet expansion meet the report's forecasts.
• Track the spread of construction restrictions, permitting rules, grid interconnection reviews, and “cost-causer pays” rules across the US.
• Monitor changes among data center supply, demand absorption, vacancy, and pricing, particularly the pricing power of existing powered capacity.
• Observe whether sovereign AI requirements expand to data processing location, overseas computing, customer verification, cybersecurity, and network routing.
• Track stabilized yields on EQIX development projects and the impact of grid infrastructure investment on recurring revenue growth.