"For now, internet innovation has stalled"
IPv4 network growth is over; contraction expected
It looks very much as if the internet has reached a tipping point. IPv4 address prices have started to drop, and the number of IPv4 routes is no longer rising. What's more, the efficiency of IPv4 use has stopped increasing and started to decline. And some access providers are finding that NAT-based expansion of their infrastructures has ceased to be cost-effective. The evidence comes from the annual surveys of IP address use and BGP routes published by Geoff Huston, Chief Scientist at APNIC. According to Huston, we may very well have passed peak IPv4 last year, with contraction of the IPv4 network likely in the years ahead.
The worrying aspect of Huston's latest surveys is his conclusion that, although the IPv4 network still functions, it is holding back innovation, openness and diversification on the internet. The market has concentrated around a small number of very big players, who are inherently risk-averse, conservative and controlling. As a result, there is a danger that the existing internet will disintegrate.
Huston doesn't expect innovation and enterprise on the internet infrastructure to recover momentum until the IPv4-to-IPv6 transition is complete. However, sluggish adoption of IPv6 may mean that the Netherlands emerges from that transition well behind many other countries.
Falling prices
Last autumn, we published a blog about reaching peak IPv4. That piece was prompted by a downturn in the price of traded address blocks. After years of steady increase, prices had plateaued in the period 2021 to 2022, then fallen sharply in 2023. As the following graph shows, the downward trend has continued over the last 6 months.
Figure 1: IPv4 address prices. [Source: IPv4.Global]
Big cloud operators
There are various possible explanations for the downturn in market prices: In recent years, the big cloud operators have bought up huge numbers of IPv4 addresses, so that they can continue offering their customers IPv4 portals for their online services.
Since then, the scramble to buy up addresses has abated, and that may be why prices have come down. However, while the prices paid for larger address blocks (/16 and above) have certainly fallen, prices for smaller blocks have fallen much more sharply. So it doesn't appear that the overall downturn can be attributed to a shift in the trading pattern, away from large deals towards the buying and selling of ever smaller address blocks.
Figure 2: Prices of IPv4 addresses (address blocks /16 and larger). [Source: IPv4.Global]
Figure 3: Prices of IPv4 addresses (address blocks /17 and smaller). [Source: IPv4.Global]
Another possible explanation is that address price trends simply reflect economic trends. So prices rose in 2021 when the economy fired up again after the coronavirus crisis, then fell when the economic climate changed. [1]
Peak IPv4
What we're seeing is a structural decline in the prices paid for IPv4 addresses across the entire spectrum of block sizes. What's more, a major downturn had been predicted: as the adoption of IPv6 progresses, there will inevitably come a point where the demand for IPv4 addresses goes into decline and prices follow suit.
According to Huston, that's exactly what has happened in the last few years. In 2021, IPv4 address block holders delayed selling in the expectation that prices would rise even further, thus squeezing the supply. Conversely, the fear of further price inflation made network operators keen to buy sooner rather than later, boosting demand.
However, use of the latest generation of transition mechanisms (464XLAT, dual-stack with Happy Eyeballs and (CG)NAT) is (slowly) driving a shift in the traffic balance, from IPv4 to IPv6. That has reduced demand for IPv4, resulting in lower prices for IPv4 addresses.
If we have indeed passed peak IPv4, the value of IPv4 addresses will only fall further from here on. That will induce major holders to put their unused and newly released IPv4 addresses on the market as soon as they can, increasing supply and amplifying the downward price trend.
Young address blocks and fragmentation
In his analyses, Huston has also considered the registration ages of the traded address blocks. In the years of strong price growth, it was mainly small, young address blocks that were being traded en masse. A newly obtained block (of which no more are available) couldn't be sold on for at least 2 years. However, the trade in (large) legacy blocks has dwindled over the years, implying that the reuse of such addresses has never really taken off.
Huston also flags up the increasing fragmentation of IPv4 address blocks: 25 per cent of transactions now involve sub-blocks (typically half blocks). However, such subdivision does not appear to have major implications for the address space or BGP routing tables. In 2023, less than 5 per cent of the total number of registered allocations were involved.
Figure 4: The cumulative registration age spread of traded address blocks over time. [Source: Geoff Huston]
Those transaction statistics reinforce the impression that we have now passed peak IPv4, and that a collapse of IPv4 address prices is yet to come. Huston believes that we have reached a point where both the IPv4 network and the IPv6 network have sufficient infrastructural capacity to meet demand.
Transition to IPv6
IPv6 users now account for 45 per cent of Google visitors. What's more, that percentage has for some time been increasing by about 5 percentage points a year. If it continues to do so, IPv6 will achieve a 60 per cent share of Google visits quite soon.
Figure 5: Global client-side IPv6 use. [Source: Google]
IPv6-only with IPv4aaS
Interestingly, RFC 9386 (published just about a year ago) says that access providers should switch to IPv6-only when IPv6 traffic accounts for 50 to 60 per cent of total traffic (not quite the same thing as visitor numbers). We recently published an article exploring the technical implications of reaching that tipping point: Internet access providers, most of whom run dual-stack set-ups with (CG)NAT and DS-Lite, will need to migrate to IPv6-only architectures. Access to IPv4-only servers will then be realised by means of IPv4aaS, based on the latest generation of transition mechanisms: NAT64, DNS64 and 464XLAT. Many mobile service providers have already gone over to 464XLAT because of their large customer bases.
Not only does an IPv6-with-IPv4 architecture reduce the number of client-side IPv4 addresses needed, but IPv6-only also implies fewer server-side IPv4 addresses. Access to IPv6-only servers for IPv4-only clients is realised using (reverse) proxies. Combining the proxies with Server Name Indication (SNI) means that, on this side too, the number of IPv4 addresses required comes down in step with the decline in inbound IPv4 traffic.
Tipping point
According to Huston, the trends discernible in BGP routing tables confirm that a tipping point has been reached. The following graph shows the change over time in the number of entries in the IPv4 routing table in the internet's default-free zone. A clear turning point can be observed in about 2021.
Figure 6: Change over time in the number of entries in the IPv4 routing table. [Source: Geoff Huston]
Decline
Various other, more specific graphs show a similar tipping point or even a decline. For example, if one disregards a number of major additions by the US Department of Defense in 2021, the total volume of the accessible IPv4 address space has been declining slightly since 2018. The number of unadvertised addresses has also been falling, and more recently the total number of assigned addresses has been coming down as well.
Figure 7: The total volume of the accessible IPv4 address space has been declining since 2018. [Source: Geoff Huston]
Figure 8: The total volume of the assigned but unadvertised IPv4 address space has been declining for a decade (before increasing slightly in 2023). [Source: Geoff Huston]
Figure 9: Very little development has taken place in the IPv4 address space over the last decade, due to exhaustion of the RIRs' address stocks. [Source: Geoff Huston]
Impact on efficiency
One of Huston's most notable and significant observations is that the exhaustion of the IPv4 address stocks held by the RIRs isn't reflected either in more efficient use the address space, or in the number of entries in the IPv4 routing table – despite the fact that the assignment of IPv4 addresses has been (strictly) rationed since 2011. As a result, the huge growth in mobile connections 10 years ago has (necessarily) been accommodated almost entirely by using CGNAT.
Figure 10: Number of IPv4 addresses assigned by the RIRs. [Source: Geoff Huston]
Impact on topology
The following graph shows that the fragmentation of IPv4 prefixes and the lively trade in increasingly small address blocks (in which the Netherlands plays a big role) has not resulted in degradation of the topology of internet. For example, the average number of steps in a route announcement (the AS path length, a measure of interconnectivity) has been stable for some years.
Figure 11: The average number of steps in a route announcement (the AS path length) has been stable for some years. [Source: Geoff Huston]
Saturation
Other developments that Huston interprets as signs of saturation are that growth in the number of autonomous systems (ASes) has slackened and that the number of transit networks has remained unchanged for some time. He suggests a number of reasons why growth tailed off: saturation of the internet market (innovation), more efficient IPv4 use by content/service providers (SNI) and end users (NAT), consolidation in the service and access provider markets (464XLAT), use of content delivery networks (CDNs), and of course the steady rise of IPv6.
According to Huston, the widespread use of CDNs is particularly significant in relation to the easing of IPv4 network growth. CDN use has also shifted a large volume of traffic from transit networks to last-mile access networks.
Where the number of ASes is concerned, we would additionally highlight the fact that, at the RIR for our region, RIPE Labs previously predicted that the number of LIR accounts/members (and therefore the number of ASes), which was then rising quickly, would fall once membership ceased to give entitlement to an IPv4 address block.
First contraction of the IPv4 network
All things considered, Huston thinks that we may very well have passed peak IPv4 last year, and that contraction of the IPv4 network is now in prospect. That could begin as soon as 2026, although projections made with outliers removed from the statistics suggest that it might be a few years further away. However, regardless of when it starts, negative growth is inevitable in the years ahead. And, once the tide turns, Huston expects the market for IPv4 addresses to collapse very quickly.
Figure 12: A quadratic polynomial equation provides the best model of the IPv4 routing table; it predicts a switch to contraction in the coming years. [Source: Geoff Huston]
Risk aversion, conservatism and a desire for control
The worrying aspect of Huston's latest surveys is his ultimate conclusion that, although the IPv4 network still functions, it is holding back innovation, openness and diversification on the internet. Large scale use of NAT means that now only client-server connections initiated by the client over TCP and UDP are possible. Applications that aren't consistent with that architecture – such as multi-user games and real-time peer-to-peer applications (internet/video telephony) – don't (or no longer) work properly. Although most end users aren't inconvenienced, the self-hosting of servers is now possible only in exceptional cases where an access provider assigns a true (static) IPv4 address to each customer [e.g. Freedom Internet].
As a result, the market has concentrated around a small number of very big infrastructure and content providers, who are inherently risk-averse, conservative and controlling. What we get in return for all those concessions is an infrastructure with roughly 3 billion addresses, to which about 30 billion devices are connected. We nevertheless run the risk that the existing internet will disintegrate into a number of separate parts, or 'service cones', each with its own private address space, formed around CDN points-of-presence (PoPs).
What's more, it's only a matter of time before the first IPv6-only services are launched, probably in regions such as Asia or South America. Servers on those networks will be unreachable for IPv4-only clients.
Innovation and enterprise have stalled
All things considered, it seems that innovation and enterprise on the existing internet infrastructure have stalled. And Huston doesn't expect them to recover momentum until the IPv4-to-IPv6 transition is complete.
His conclusion ties in with the concerns that we ourselves have previously expressed regarding the Netherlands' sluggish adoption of IPv6, which we believe is damaging the Dutch economy. It makes the Netherlands less attractive for initiatives with the Internet of Things (IoT) and tarnishes the country's image as a business-friendly innovation centre. The situation is made all the more worrisome by the fact that the Netherlands is continuing to lose ground. If we have indeed passed peak IPv4, the Netherlands looks likely to emerge from the IPv4-to-IPv6 transition well behind many other countries.