How Many IP Addresses Are There?[ IPv4 & IPv6]

By Muninder Adavelli. Scraping, SEO, and proxies for over a decade.

The 1981 Assumption "4B Ips Is Enough" Did Not Age Well

Back in September 1981, in RFC 791, they published that IPv4 address space was fixed at 32 bits.

That 32 bits set up made the 4.29 billion IPV4 addrsses possible [2^32 = 4,294,967,296 possible addresses]. For those times, the number was huge and practically a number that would never be exhausted. In those times, it was ARPANET internet, a research network connecting a few hundred nodes at universities and government labs.

Vint Cerf, who co-designed TCP/IP, even said that they expected IPv4 to be an experiment and a production level replacement would probably arrive before the address space filled up. Seems like we got used IPV4 way too much.

As we all know 1990s saw a massive computer boom and commerical internet grew so much that the math on the 4 billion IPs was at risk. No wonder, IPv6, the 128-bit addressing was standardized in 1998.

However, for the next two decades, the industry still used IPv4 along with IPv6. IPv4 for compatibility, IPv6 for forward progress

And as they expected in 1990s, IPv4 pool drained out in 2011.

Here is the IPV4 pool exhaustion timeline

The common phrase we see around that IPv4 ran out in 2011 is technically correct. They are correct about the central pool.

However, there are some details around it:

• IANA pool exhausted but each of the other RIR members ran out at a different point

• there is a secondary market where you can buy an address for $30–60 from someone who already owns it and has been running ever since.

The bottomline is IPv4 addresses didn't disappear completely when they say they exhausted. They just became expensive.

What the engineers in 1981 thought: 4.29 billion is effectively unlimited for a research network with hundreds of nodes. They were right about the research network. They had no model for 15 billion connected devices, cloud VMs, smartphones that cycle IPs, IoT sensors, and content delivery networks needing millions of addresses per region.

4.29 Billion IPv4s : What It Actually Means

The 4,294,967,296 ips has both public and private ranges.

The Private Ranges (RFC 1918) - The Three Blocks Reserved For Your HOme and Office

RFC 1918, published in 1996, reserved three blocks permanently for private networks. These are the addresses that route inside your home or office but are invisible to the public internet:

Total RFC 1918 private ips: 17,891,328 addresses.

None of these are publicly routable. All of them are behind NAT.

The IANA Special-Purpose Blocks

In addition to the RFC 1918, there are more special blocks - IANA's special-purpose registry has these:

• 127.0.0.0/8 : The Loopback (16.8M) 127.0.0.1 is localhost. The entire /8 is reserved for it; every packet sent here stays on your machine.

• 100.64.0.0/10, the shared address space (4.2M): RFC 6598 reserved this for carrier-grade NAT. Soms ISPs may put you behind this before giving you a "real" public IP.

• 224.0.0.0/4, the Multicast (268.4M): This comapared to the above two is a large chunk and is reserved for one-to-many routing protocol traffic. These are used for routing updates, service discovery and not unicast web requests.

• 240.0.0.0/4, Reserved for future use (268.4M): This big chunk is blocked off since 1989. No one actually used them ever. As IPv4s are getting scrace and given their importances to businesses now, various proposals tried to liberate this space. But no one suceeded.

• 169.254.0.0/16, the Link-local (65K): This small pool of addresses your device can self-assigns when your DHCP fails.

• Approximately another 200k ips are for Documentation, benchmarking, IETF protocol assignments.

If we add these up, roughly around 575 million addresses are not open for public routing or usage. That leaves approximately 3.72 billion IPv4 addresses that can be assigned and reached from the open internet.

So, is that 3.72 billion is completly open and public?

Substantially part of that 3.72 billion is open for public but some of it blocked in for legacy ARIN blocks that are owned by US universities, defense contractors, and early internet companies who received huge allocations in the 1980s before allocation policies existed.

• For instnace MIT holds an entire /8 (18.0.0.0 block with 16.7 million addresses. And the funny part is it they sold part of it to Amazon in 2017.

• The US Department of Defense holds multiple /8s. Many are partially routed. Some are almost entirely dark.

Proxy.Report Note: Now that we are discussion ip pools, if you are buying a proxy service, and wondering about pool quality, it is not number 4.29 billion or 3.72 billion that matters. The key is to know how many addresses in the proxy providers pool are currently CLEAN, correctly registered to a residential ISP ASN, and not flagged by Google's fraud infrastructure.

Thats said, lets get back IANA and RIRs to know more about how IPs are managed. Indeed who manages it and how

Who Owns these 4 billion ips? IANA, the Five RIRs, and Country Breakdown

IANA an organzation that is on the top and has authority on these ips and ensures the internet runs smoothly from the IPs perspective.

What does IANA do with the IPs?

It does not assign addresses to ISP - internet service providers directly.

Its main job is to allocate blocks to five Regional Internet Registries, which then delegate to ISPs and organizations in their regions.

(Special note: These numbers are from IANA registry data, 2024. The includd legacy blocks as well and the exact "currently routed" numbers might differ. Thats an honest missing part and I did mention it in my missing data section of this article)

We can clearly see that the allocations for ARIN region is quite huge compared to other RIRs and even if we consider the population to IPs ratio. The massive allocation clearly reflects that, internet evolved in there.

And also, the numers could be huge as most of these US institutions and organization received enormous blocks of these ips, even before a formal scarcity policies ever existed. They were allocated directly by IANA's predecessor organizations in the 1970s and 1980s. In their defence, who in the 1970s thought, a concept like internet would be this much of a REALITY. or may be I can go a step further and say THE REALITY!

Also, even after that policy, most of the internet companies started in the US, flourished in the US, in the 1990s and 2000s.. For the whole 2 decades. And probably that could add to that massive number as well.

For instance lets have a look at AWS ips count. to cut it short, AWS has more ips compared to Germany!

While writing this I was parellely researching on AWS, GCP and AZure, the big 3s of cloud computing. The services, I use as proxy alternative or my own datacenter ips.[Thanks to serverless compute]

Coming back to AWS, they themselves publicly claims 188.7 million IPv4 addresses (5.09% of all routable IPv4), with a further 55.5 million routed via BGP but not published. If you are keen on knowing more about it, I wrote a script to analyse, filter and extract info from open AWS ip data and the report is here: How Many IP Addresses Does AWS Have? . Have a look at full breakdown after this one.

Did I mention AWS has more ips than Germany? Have a look at other countries as well and see for yourself how many countries are actually having lesser ips than AWS.

Top Countries With IPv4 Share :

(Sources: Wikipedia list of countries by IPv4 allocation; RIPE NCC understanding IP addressing report.

I have some calcullations on how Per-1,000 figures are calculate in my How to Verify section below)

Now that we read that table. Let me present some comparisons.

1. The US has over 36% of all publicly routable IPv4
   And, it accounts only to 4% of the world's population.

2. India has a 1.5 billion people and has lesser IPs compared to AWS.

On that note, lets see how concentration vs scarcity play out and why proxy services makes sense in enabling a person living in India getting access to US IPs

IP Addresses Per Person: Concentration and Scarcity

Globally, there are 3.72 billion public and routable IPv4 addresses
And globally there are more than 8 billion people

This math works out at roughly 465 ips per 1,000 people.

But this global average hides the stuctural details and significant difference in how internet access works across countries.

Consider two people: one living in Mumbai, India, and another in Minneapolis, US. Both use a laptop, a home internet connection and want to access the same website.

The person in Minneapolis is more likely to have a dedicated public IPv4 address assigned by their ISP - Give their 4000 odd ips per 1000 people ratio. And their household's internet traffic leaves through that address.

And now the person in Mumbai might be behind Carrier-Grade NAT (CGNAT). Their ISP could have hundreds of customers sharing a much smaller pool of public IPv4 addresses.

On paper both people can browse the internet normally and might acceess that same targeted website normally. They might not even notice the difference unless something like this happens.

Imagine this scenario, the american and indian started scraping product prices from an e-commerce website for personal use and not so aggressive.

The Minneapolis user's requests might come from a public IP shared with relatively few customers. The Mumbai user's requests might come from an IP shared by hundreds of households.

Still both might get proper 200s and sucesses

But imaging this now, another customer sharing that Mumbai exit IP is running an aggressive scraper and gets flagged by the website.

The IP itself can now have a poor reputation. Our first Mumbai user who is normally browsing the website with a perfectly legitimate request, he may be blocked or in most cases carry the weight of that aggressive scrapers reputation on him.

This is the structural fact behind why residential proxy pools "in India" or "in Africa" are thin and high-abuse-velocity. It's not a proxy provider quality issue.

NAT: The Fix That Became the Foundation

Network Address Translation was proposed in 1994 as a temporary workaround to extend IPv4's life, a bridge to hold things together while IPv6 rolled out. That temporary fix became the permanent architecture of consumer internet, and it never left.

How NAT Works

Your home router has one public IP address and it might connect to 5 digital devices phone, laptop, TV, tablet, speaker and even your robot vaccum. Each have their own private 192.168.x.x address.

However, when the traffic leaves your home, the router rewrites the source address to its public IP. When traffic returns, it rewrites the destination back to the right private device.

From the internet's perspective, your household is one IP. From inside, it's five independent devices.

Thats NAT in simple words.

CGNAT, used by ISP providers takes this tech a step further.

Carrier-grade NAT (CGNAT) Like we talked before, in countries like India, where 30-40 million public IPs are shared among a billion and half people and companies, ISP uses similar NAT tech. They place entire neighborhoods behind a single public IP. And then there is your router doing the second NAT layer for your home network.

So in short, hundreds of users share one exit address.

CGNAT is documented in RFC 6598 (the 100.64.0.0/10 block is reserved for it). It's standard practice in regions with IPv4 scarcity.

And This Matters for Proxy Pools

As proxy providers and ips researcher, I want to look this from a proxy use case perspective. If me as a serp scraping enthusiast, use a "residential IP" in a proxy pool, lets say monitoring a ranking for google.co.in, I may be using a residential proxy IP that might be a CGNAT exit node serving hundreds of real households.

That exit node has an abuse ceiling far lower than a datacenter IP used by a single operator. And there comes the proxy hack. In some countries, data center IPs probably are better than residential IPs, if your use case involves IP blocks and pool burn rates.

So, when a proxy provider says "We have 5 million residential IPs," as a buyer, the first question, we should ask is

• how many of those are unique public exit addresses

vs

• how many of them are private addresses that NAT to shared exits.

So, far I have gone through 6 provider websites in and out and no proxy provider marketing page quantifies or discusses this CGNAT problem that actually defines the proxy quality outside US market.2

IPv6: 340 Undecillion and Counting - The Solution.

IPv6 addresses are designed to handle scarcity problem and growing need of IPs by switching from 32-bit to 128-bit addressing.

The math: 2^128 = 340,282,366,920,938,463,463,374,607,431,768,211,456 addresses. Thats roughly 340 undecillion.

• To put this 340.28 undecillion ip addresses in perspective,

• it is like 340.28 trillion trillion addresses - Trillion trillions is not a mistake here.

• 340.28 billion billion billion addresses

• 79.3 quintillion times the entire 4.29 billion IPv4 address space.

• 91.5 quintillion times the 3.72 billion publicly routable IPv4 addresses.

To make that number tangible: there are approximately 7.5 × 10^18 grains of sand on all of Earth's beaches. IPv6 provides enough addresses to give every grain of sand roughly 45 quintillion addresses each.

This i think how they probaby felt when they had billions of ipv4s. Lets hope we the ipv6 goal of trying to make scarcity structurally impossible in future works out.

IPv6 Allocation Architecture

The architecture works with a small difference. In addition to the large pools it gives the organizations the ability to divide them to seperate networks.

The usual practice is giving an organization a /48 block

But what is a that /48 block? -

• The first 48 bits identify the organization's allocated network block.

• And the organization has another 16 bits it can use to create internal subnets.

• Following that, every subnet has 64 bits available for individual interface addresses.

In short a single /48 allocation can allow an organization have

• 65,536 separate subnets (2^16).

• 18.4 quintillion addresses per subnet (2^64).

• 1.2 septillion addresses in total (2^80).

In simple terms one organization with /48 allocation can have roughly 281 trillion times more addresses than the entire IPv4 address space (4.29 billion).

That sounds like the problem is solved anywyas. Then

Why IPv4 Persists Anyway

IPv6 was already standardized in 1998 and was ready to use.

But, as of 2026, only 40–45% of internet traffic uses IPv6

Google says approximately 45% among its users are on IPv6. The remaining 55% is still IPv4, reached through dual-stack deployments.

If you are wondering why, the reasons are purely operational, not technical.

• Network equipments are decades old.

• CGNAT was cheaper. So, ISP infrastructure did not get the fast paced upgrades it needed

• Enterprise networks where IPv4 → IPv6 migration costs more than the pain of living behind NAT.

Surprising facts on IPv4 vs IPv6:

1. Even wit practically endless supply of IPv6, IPv4 addresses still trade at $30–60 each.

2. Proxy providers measure pool quality in IPv4.

3. Blocklists are built on IPv4.

4. Anti-bot systems score IPv4 ranges.

In short, IPv6 is the future, a future that has been a future every since 1988.

What about the counts on Blacklisted IPs, Bot Traffic, and Pool Quality

Since pool quality is what sent me down this road to research on number of ips, I went around to find on how many of that 3.7 billion ips are uselsess because of blacklists and blockades.

What drives IPs out of usable circulation:

And my assumption was correct. Some of ips in this already scrace IP pool are being blacklisted or blocked.

• Botnet history: An IP that in past ran some spam botnet will obviously appear in some Spamhaus DROP, SORBS, or similar blocklists. The ips blocked here are useless. Spamhaus's DROP list usually covers blocks that are completly hijacked and are consistently abused. These are not used for routing.

• Platform-specific blocks: Google, Meta, Cloudflare, and Microsoft maintain internal reputation lists that are not published. An IP flagged by Google's reCAPTCHA infrastructure may pass basic connectivity tests but fail the moment you send a search query or submit a form. I experience this on daily basis.

• ASN-level blocks: Some hosting provider ASNs are blocked wholesale by certain platforms. If your residential proxy pool routes through an ASN that's been flagged (which happens when a datacenter buys IPs and re-marks them as residential), the whole block is degraded.

• CGNAT velocity abuse: Shared CGNAT exit nodes accumulate abuse flags from thousands of users. The IP itself may have never been in a botnet, but its collective usage pattern triggers rate limits.

Cloudflare's 2025 Year in Review reported that roughly 25–30% of internet traffic is bot traffic. The fraction originating from residential IPs (versus datacenters) has been increasing as bot operators shift to residential proxy networks to evade datacenter-range blocklists. That shift is precisely what degrades residential pool quality over time.

Numbers Blown Out Of Proportion or Needs Extra Context

"IPv4 is exhausted"

"IPv4 ran out in 2011" stated as a simple fact across networking coverage and Wikipedia leads. This was a click bait and news headline even in 2026. Sure it sounds scary and thats why it gets that attention. But how real is that IPv4 depleted completely is a reality.

IANA's central pool ran out February 3, 2011. Each RIR's free pool hit zero over the following decade.

• But addresses never stopped moving the secondary market,

• returned address reuse, and RIR waiting lists mean IPv4 addresses are still being allocated. ARIN currently allocates from a waiting list; returned addresses queue for the next applicant.

If you ask me, I would write it as IPv4 addresses are economically scarce, not physically absent. The era of free IPv4s is ended.

What Data Cannot Tell Us or What I could Not Find Directly Related to IPv4 address counts

1. How many publicly routable IPv4 addresses are currently active in routing tables
   - IANA allocation data shows addresses that have been assigned, not addresses currently being advertised in BGP.
   - CAIDA and RouteViews maintain BGP routing table snapshots, but I couldn't find a single consolidated "currently active IPv4 addresses" figure.
   -A substantial fraction of allocated space is likely dark — assigned but not routed.

2. What percentage of residential proxy pool IPs are pre-flagged by major platforms
   I have been trying to break this news for quite some time both with scripts and raw data hunting. But I could not.
   Google, Meta, Cloudflare, and Microsoft maintain internal blocklists that are not public. No study I found produces a reliable estimate of what fraction of "legitimate residential" proxy IPs are flagged across all major platforms. I genuinely think this is the number that can actually break all the false priomises of proxy provider marketing campaings full of ip pool numbers. Unfortunately I could not find out outright number. But I am not giving up, yet.

3. Bot traffic share broken down by IP type
   Cloudflare and others publish aggregate bot traffic percentages (~25–30%), but no breakdown separates datacenter ASN traffic from residential ISP ASN traffic from CGNAT exit node traffic. The rate by address type matters enormously for understanding proxy pool degradation over time. I really think this is only a matter of time, Cloudflare realizes this as a very good data angle and publish these numbers. With 20-30% of websites routed through it, it can clearly make web a better place with this kind of studies.

4. The actual secondary market size
   ARIN, RIPE, and APNIC all publish transfer logs, and several brokers (IPv4.Global, Hilco Streambank) operate in this space. But total annual market value and transaction volume aren't consolidated into a single public figure.
   Estimates of $1–2 billion/year appear in trade press without clear methodology.

How to Verify and Calculate some numbers in here

I did some math calculations in this article and here are some clarifications around those numbers.

Total IPv4 address space: 4,294,967,2962^32 = 4,294,967,296 Not a measurement — this is the mathematical consequence of 32-bit addressing, verifiable by anyone with a calculator.

RFC 1918 private total: 17,891,32816,777,216 (10.0.0.0/8) + 1,048,576 (172.16.0.0/12) + 65,536 (192.168.0.0/16) = 17,891,328 Source: RFC 1918 (Rekhter et al., 1996).

Publicly routable estimate: ~3.72 billion4,294,967,296 − 17,891,328 − 268,435,456 (multicast 224/4) − 268,435,456 (reserved 240/4) − 16,777,216 (loopback 127/8) − 4,194,304 (CGNAT 100.64/10) − ~700,000 (documentation + link-local + other) ≈ 3,718,533,536 Source: IANA Special-Purpose Address Registry (current).

US IPv4 concentration: ~36%1,540,000,000 ÷ 4,294,967,296 × 100 = 35.8% Source for 1.54B: Wikipedia list of countries by IPv4 allocation; RIPE NCC understanding IP addressing report.

India IPs per 1,000 people: ~3245,000,000 ÷ 1,400,000,000 × 1,000 = 32 Source for 45M: Wikipedia list of countries by IPv4 allocation. Source for 1.4B: UN World Population Prospects 2024.

US IPs per 1,000 people: ~4,5971,540,000,000 ÷ 335,000,000 × 1,000 = 4,597 Source for 1.54B: same as above. Source for 335M: US Census Bureau 2024 estimate.

IPv6 total: 340,282,366,920,938,463,463,374,607,431,768,211,4562^128 = 3.4028 × 10^38 Not a measurement — mathematical consequence of 128-bit addressing.

Grains of sand vs. IPv6 comparisonIPv6 (3.4 × 10^38) ÷ grains of sand on Earth (7.5 × 10^18) = 4.5 × 10^19 IPv6 addresses per grain of sand Source for grains of sand: Howard C. McAllister, University of Hawaii (commonly cited in the literature as ~7.5 × 10^18).

Missing Data / Research Gaps

None of articles are every done deals, I always come back to them and revamp them from time to time. But for now, to call this a done deal, here are the figures I wanted and couldn't source from the initial URL set. If you can point me to a primary source for any of these, I want it.

1. Exact IPv4 addresses currently in BGP routing tables CAIDA's AS Rank project and RouteViews both maintain routing table data. I need a current snapshot of unique IPv4 prefixes being actively routed. If you work with BGP data, the CAIDA dataset is the right starting point. We've answered this for one specific major holder: AWS announces 244,140,288 IPv4 addresses across AS16509, AS14618, and AS8987 — 55.5 million more than appear in their published ip-ranges.json. See How Many IP Addresses Does AWS Have? for the methodology, which applies directly to any other large ASN holder.

2. Blacklist coverage by major platform, by address type Google's Safe Browsing, Meta's fraud detection, Cloudflare's threat intelligence, and Microsoft's Defender all maintain internal IP reputation lists. None publish coverage figures. I want: what fraction of residential ISP IPs vs. datacenter IPs vs. CGNAT exit nodes are pre-flagged. This is the number that makes proxy pool quality claims verifiable.

3. Bot traffic % by IP type Cloudflare, Imperva, and Akamai publish annual bot traffic reports (~25–30% of traffic is bot). None break this down by address type. The question that matters: of residential proxy IPs in circulation, what fraction is currently generating bot traffic?

4. DoD-held IPv4 blocks: allocated vs. actively routed The US Department of Defense holds multiple /8 blocks. IANA records who holds each block; RouteViews records what's being announced in BGP. The gap between the two is "dark space." I want the current DoD dark space figure. We've demonstrated this methodology on AWS: 188.7M published vs 244.1M BGP-routed = 55.5M dark space (22.7%). The same RIPE Stat pipeline applies to any large ASN — see How Many IP Addresses Does AWS Have? for the working implementation.

5. ARIN/RIPE/APNIC secondary market volume All three publish transfer logs (transfers between organizations, not direct sales). I need someone who's aggregated these into annual transaction count + estimated dollar volume. The IPv4.Global and Hilco Streambank broker data would help cross-check.