The engineers behind Cloudflare’s new edge hosting expansion

Cloudflare engineers use different partners to expand their footprint. The company is increasing its US presence through two separate batches of new locations. The first batch utilizes Vapor IO for deployment. The second batch uses EdgeMicro. This strategy moves computing infrastructure closer to internet users. Cloudflare already operates in 200 cities across more than 90 countries.

The scale of US expansion

Cloudflare expansion in the US involves dozens of new locations. Vapor IO provides the first group of sites. EdgeMicro provides the second group. Vapor IO has active edge data centers in Chicago, Pittsburgh, Atlanta, and Dallas. The company is building 16 more sites and planned to finish another 16 cities by the end of 2021. EdgeMicro has live sites in Austin, Raleigh, and Tampa. EdgeMicro expects to launch in four more markets by the end of the third quarter. These startups build small data center sites specifically for edge computing.

The deployment strategy focuses on proximity. Cloudflare aims to reduce latency for a maximum number of people. Nitin Rao, the head of global infrastructure at Cloudflare, says the network is within 100 milliseconds of 95 percent of the world’s population. This figure reaches 99 percent for internet users in the developed world. Accessing a hyperscale cloud data center from a city like New York or Seattle can take up to 200 milliseconds. Matt Trifiro, the chief marketing officer at Vapor IO, says Cloudflare moves aggressively to the edge. He notes that Cloudflare can get lower latency in specific regions by deploying its existing equipment into Vapor IO facilities.

Network architecture and middle-mile fiber

Low-latency workloads depend on the middle mile. This is the network section between access networks and the interconnection point for regional data centers. Vapor IO spends time on site selection and the assembly of middle-mile fiber routes. The company draws a ring around a city population center and overlays assets like land and fiber routes. They use 150 criteria as risk-mitigation factors. These factors include population location and the locations of cell towers and exchange points. Vapor IO picks typically three sites in a city. These sites interconnect with multiple, redundant fiber routes to form a distributed data center. Vapor IO also connects to last-mile networks, cable head ends, and the telco baseband unit using dark fiber that they assemble themselves.

Vapor IO uses a software-defined networking platform to cross-connect any two networks that terminate at or connect within one of its pods. This platform works based on agreements between customers like Cloudflare and providers like Verizon, Comcast, or Charter. This method results in sub-five-millisecond latency between any two Vapor locations in a city. In many metro areas, traffic usually backhauls to large internet drains. For example, all internet traffic in Austin, Texas, backhauls to Dallas. Major metros like Chicago, New York, New Jersey, Miami, Atlanta, and Los Angeles also have seven or nine internet drains.

EdgeMicro chooses sites closest to fiber interconnects. This includes core routers for the internet exchange or internet partners like ISPs and MSOs. EdgeMicro does not buy dark fiber to interconnect its own data centers. Instead, the company brokers network-connection partnerships for Cloudflare. This helps avoid the network backhaul that usually happens in secondary markets. You should understand that edge computing reduces the distance between code and users. EdgeMicro works with network partners to ensure traffic stays localized in a market.

Power density at the edge

Vapor IO sites have between 150kW and 180kW of power capacity. These sites use high-density 35kW racks. This density is comparable to power density in hyperscale cloud data centers. Vapor IO designs its pods to accommodate multiple customers per rack. EdgeMicro does not focus on high density. EdgeMicro clients typically request 8kW to 10kW per rack. Jason Bourg, the VP of strategy at EdgeMicro, says the Cloudflare deployment at EdgeMicro is smaller.

Site Type	Rack Power Density	Primary Use Case
Vapor IO	35kW	Multi-customer support
EdgeMicro	8kW – 10kW	Cloudflare deployment

Developer platform leadership

Cloudflare is a Leader in The Forrester Wave: Edge Development Platforms, Q1 2026 report. The report gives Cloudflare the highest score in the Strategy category. Cloudflare also received the highest possible score for the Vision, Innovation, and Roadmap criteria. Forrester describes Cloudflare as a strong fit for organizations seeking edge-first applications and performance-critical user experiences. The report cites Cloudflare’s investment in AI application development. This includes the AI Gateway for routing and governance and an emerging agent-first development ecosystem.

Dane Knecht, the CTO at Cloudflare, says the company strips away the complexity of distributed systems so developers can focus on building applications. Cloudflare aims for widest possible distribution to improve the developer experience. Developers no longer choose between running code in Ashburn or Seattle. The adoption of serverless, such as Cloudflare Workers, makes it likely that code runs in many locations. Cloudflare’s platform allows for the use of WebAssembly modules and a beta Python runtime.

Category	Cloudflare Score
Strategy	Highest
Vision	Highest
Innovation	Highest
Roadmap	Highest

Performance benchmarks and cold starts

Cloudflare Workers runs code inside V8 isolates. This technology allows Cloudflare to multiplex thousands of tenants on one machine. There is no container to boot and no VM to warm up. AWS Lambda@Edge uses a container-based execution model. This model replicates full Lambda functions across CloudFront edge locations.

Cloudflare Workers cold starts take less than 1 millisecond. Lambda@Edge running Node.js has cold starts between 250 and 800ms. Lambda@Edge cold starts remain high.

Metric	Cloudflare Workers	AWS Lambda@Edge
Cold Start	< 1ms	250ms – 800ms
TTFB (US East)	8ms – 12ms	12ms – 18ms
TTFB (APAC)	15ms – 25ms	25ms – 40ms
Locations	320+	600+

Ease Cloud’s study found Cloudflare Workers produced lower P50 latency in every region tested. For example, Workers produced 15-25ms latency in APAC while CloudFront produced 25-40ms. While Cloudflare offers unmetered bandwidth for cacheable HTTP traffic on lower tiers, teams pushing 50 TB of video through a Pro plan risk having their traffic deprioritized or receiving a request to upgrade to an enterprise contract.

Storage and state management

Edge functions require storage to read configuration or persist session state. Cloudflare provides several storage primitives. Workers KV provides fast, eventually consistent key-value reads. Durable Objects provide strongly consistent stateful coordination. D1 is a serverless SQL database. R2 is an S3-compatible object storage with no egress fees. These services are built to be called from Workers with minimal added latency.

AWS provides access to the full AWS service catalog. This includes DynamoDB, S3, SQS, and Step Functions. Every call from an edge location back to a regional AWS service adds network round-trip time. This can reduce the edge advantage. Cloudflare’s Workers KV trades strong consistency for global read performance. Values written in one region can take up to 60 seconds to propagate globally.

Service Type	Cloudflare Option	AWS Option
Key-Value	Workers KV	DynamoDB
Object Storage	R2	S3
SQL Database	D1	Aurora/RDS (via connection)
Stateful Coordination	Durable Objects	Step Functions

Pricing and bandwidth costs

Cloudflare structures pricing around four tiers. The Free and Pro tiers include unmetered bandwidth for cacheable HTTP traffic. Enterprise contracts negotiate bandwidth commitments. For Enterprise plans, the effective rate is between $30 and $50 per TB.

Cloudflare egress pricing on Free, Pro, and Business plans is zero for cacheable HTTP traffic. However, metered billing begins for services like Cloudflare Images, Stream, or R2 egress beyond the free tier. R2 read operations beyond the free allocation cost $0.36 per million requests.

Provider	Low Volume (per GB)	High Volume (per GB)	Effective per TB
Cloudflare (Enterprise)	~$0.05	~$0.03	$30 – $50
Amazon CloudFront	$0.085	$0.02	$20 – $85
BlazingCDN	$0.004	$0.002	$2 – $4

CloudFront pricing varies by region. In Asia-Pacific, CloudFront egress costs $0.12 per GB in India and $0.14 per GB in certain zones. Cloudflare does not charge regional surcharges.

Will developers eventually abandon the centralized cloud entirely for the edge?

I recommend Cloudflare Workers for developers who need the lowest possible latency for user-facing applications.