H1 2026: The Six Months That Repriced Hosting

Every industry has six-month stretches that are just weather, and a few that turn out to be climate. The first half of 2026 was climate. Between January and June, nearly every layer of the hosting stack acquired a new price, a new owner, or a new regulator, and usually two of the three. Commodity memory and storage repriced, Europe’s biggest infrastructure providers followed, both major domain escalators restarted, and renewal spreads stretched past 5x. Meanwhile the customer-acquisition funnel that built the mass market kept measurably collapsing under AI search.

This publication spent the half-year building instruments instead of collecting adjectives: a renewal-multiplier index, audits of 736 companies’ security and AI posture, a registrar concentration index computed from registry data, a GPU rental index, a price archaeology of six years of Wayback snapshots, a locked cohort of 2,268 newborn domains. Read together, those instruments say something none of them says alone.

In H1 2026 the hosting industry stopped growing by customers and started growing by invoice, and it did so at every layer simultaneously, from the DRAM chip to the renewal notice. This is the synthesis: what actually happened, what it cost, who paid, and what the second half will decide.

Key facts, the half-year in six numbers (all from our verified reporting, January–July 2026)

• 5.3x: the highest measured renewal-to-intro multiplier in our June index (SiteGround StartUp, €2.99 to €15.99), with the mass-market center at 3.3–4.3x and Namecheap raising shared-hosting renewals by up to about 20 percent in May
• -604,000 and +19 percent: GoDaddy’s customer decline over two years against its ARPU growth ($203 to $242), the single cleanest statement of the industry’s new growth model
• Roughly +30 percent versus 0: cumulative US inflation since January 2020 versus the price change of Vultr’s and Linode’s entry instances; the renewal tier ran 33 to 100 percent over the same tape
• $10.97 and $11.00: the .com wholesale price from November 1 (announced April 23, into a record 176.1M-name .com/.net base) and the .org price since June 1, its first increase since 2016
• ~$850 billion: committed data-center leases across the major cloud providers that do not yet appear on balance sheets, up from ~$700 billion in March; Meta alone added $79 billion in a quarter
• 2 of 25: hosting providers offering a full executing stack for AI agents as of early July; ten ship an official MCP server, and the shared-hosting incumbents ship none

The Half-Year Ledger

The Invoice Layer: Where the Repricing Actually Lives

Start where the money is collected, because H1’s defining discovery, made twice by two different instruments, is that hosting’s inflation is real but selective. Our June Renewal Multiplier Index measured the spread between what recruits a customer and what keeps one: SiteGround at 5.3x, DreamHost at 3.8x, Hostinger at 3.3–4.3x on 48-month prepays, and, at the honest end, IONOS at 1.7–2.0x. Our July price archaeology then added the time axis, and the two tiers separated completely: where buyers compare, six years of zero nominal movement against roughly 30 percent inflation (Vultr’s IPv6-only entry at $2.50 since 2019, Linode $5.00, WP Engine cheaper than in 2020), and where buyers renew, +33 percent at GoDaddy, +53 at SiteGround, +100 at IONOS, with almost all of it landing after January 2023.

The mechanism is not a mystery; it is a strategy, and the filings admit it:

• GoDaddy spent a record $375 million on marketing in 2025 and finished with 604,000 fewer customers than two years earlier, while ARPU rose 19 percent
• Wix spent $514 million, lost 1 percent of its subscribers, and announced it would stop disclosing the count
• even DigitalOcean, the efficient outlier at 9 percent of revenue, redefined its metrics away from volume

The industry’s largest players are no longer priced for acquisition; they are priced for extraction, and every renewal-notice increase, attach-rate push and multiplier stretch our instruments recorded in H1 is that strategy executing. The counter-trend proves the trend: InterServer and HostMetro turned price locks into positioning precisely because, after Hetzner raised entry cloud lines by roughly a third (and dedicated-vCPU lines far more), OVHcloud put 43–49 percent on VPS and Netcup 24 percent, a promise not to reprice became a product. And the regulators arrived on schedule: New York’s renewal-consent law has been live since November, the Amazon ROSCA settlement ($2.5 billion) redefined the downside, and the renewal-jump model that built these margins now operates, for the first time, under adult supervision.

The Input Shock That Explains Half of It, and the Half It Doesn’t

Some of H1’s repricing was honest cost pass-through, and the honest part is easy to identify because it has a commodity ticker behind it. DRAM contract prices jumped 90–95 percent in the first quarter with another 58–63 percent projected for the second, leaving the benchmark DDR5 chip at roughly four times its September 2025 price; NAND projections ran 70–75 percent in a quarter; retail hard-drive prices rose roughly 50 percent in five months with Western Digital’s data-center capacity sold out through 2026. Our backup-economics analysis traced the pass-through in real time: Wasabi to $7.99/TB (July 1), Backblaze to $6.95 (May 1), Hetzner object storage +30 percent, all while NIS2 and DORA quietly turned data retention from a choice into a floor, so the repriced storage cannot even be declined. The RAM shock similarly underwrote the European VPS increases and will keep pressuring the segment through H2.

But the input story only explains the layers that held their margins; it cannot explain the layers that expanded them. The domain stack has no DRAM exposure, and it delivered the half-year’s purest pricing power: Verisign announced its 7 percent increase on April 23 while reporting the strongest new registrations since 2021 (11.5 million .com and .net names in Q1, a combined 176.1-million-name base, guidance raised), and .org’s June 1 jump to $11.00 made a nonprofit’s TLD the most expensive of the big three legacy extensions at wholesale. Our registry league table, built from eight registries’ published accounts, established what the service underneath actually costs: about one dollar per domain-year, triangulated three ways:

• DENIC’s €1.01 cooperative price
• PIR’s $0.95 backend contract
• Verisign’s own ~$1.14 cost of revenues

Everything between that dollar and the invoice is allocation, and in H1 2026 the allocators all moved in the same direction at once.

The Demand Side Broke First, Which Explains the Rest

Why did the whole industry pivot to the installed base in the same half-year? Because the machine that supplied new customers measurably broke, and H1 was when the breakage stopped being deniable. Google users click a result on 8 percent of AI-summary searches versus 15 percent without, per Pew; search referrals to the review publishers that feed hosting’s affiliate funnel fell 22 to 60 percent in two years depending on size; Future plc’s affiliate revenue dropped 24 percent in a half-year with Tom’s Guide’s audience down 36 percent. The affiliate channel, which pays $65 to $500 bounties per acquired customer, is dismantling in public. Free tiers, the other great top-of-funnel, were quietly repriced across the developer platforms, with Netlify doubling credit burn rates in April.

Meanwhile the aftermarket data added a twist nobody predicted cleanly: AI search killed the traffic value of domains (parking is dead, Sedo and Team Internet’s monetization businesses both up for sale) while inflating their brand value to records (AI.com at $70 million, reported sales volume +32 percent), a bifurcation that is the demand-side story of the decade compressed into one asset class. Whether the new demand that AI supposedly brings is durable is now, thanks to our July cohort study, an empirical question with a scheduled answer: we locked 2,268 registry-verified newborn .com domains against Verisign’s disclosed mid-40s first-time renewal rate, predicted 40–48 percent survival, and will report in October 2027. H1’s honest summary of the demand side: the funnel that made volume growth cheap is gone, the AI-era funnel is unproven, and every incumbent priced accordingly.

The Capital Structure Underneath, and the Machines Above

None of this happened in a financing vacuum. The half-year’s thirty-deal consolidation wave, mapped in June, ran on private equity (roughly 84 percent of data-center deal value), and the ownership layer’s debts explain the pricing layer’s behavior with uncomfortable precision: Newfold, carrying ~$3.5 billion at about six times earnings and cut to Caa3, is the same company whose brands sit at the bottom of our Core Web Vitals league (Bluehost 39.8 percent, HostGator 36.2 percent of customer sites passing) and whose renewal prices our index found served as unreadable “$x.xx” templates; WebPros, the cPanel-Plesk-WHMCS toolbooth, was reported raising ~$1 billion partly to fund a dividend weeks before the September licensing window.

Above the traditional industry, the AI buildout added a stranger balance sheet: ~$850 billion in committed leases that do not appear as debt, GPU depreciation schedules disagreeing by 50 percent about how long the underlying asset earns, cyber insurance getting cheaper (the first US premium decline in the market’s history) while claims rose 40 percent, and, in the half-year’s closing weeks, the surplus market arriving: SpaceX leasing out Colossus 1 for $1.25 billion a month, Google buying explicitly cancellable bridge capacity, and Meta, on July 1, reported to be building a resale business for its spare superintelligence infrastructure. Our GPU Rental Index baseline (a 6.2x spread on the same H100-hour) is the “before” photograph of that market.

And while the capital repriced, the customer changed species: WooCommerce made 4.5 million stores agent-addressable in June, Cloudflare’s 402 economy kept building its toll infrastructure, the CMA ordered the one uncharged crawler to accept an off switch, and our agent-readiness census found exactly two providers ready to let the new customer operate the product. The AI Act’s August 2 transparency obligations, surviving June’s omnibus, will bill the European sellers of that readiness first.

What H1 Settled, and What H2 Decides

Three things are settled:

• The growth model flipped. Every instrument we ran, from the multiplier index to the CAC filings to the archaeology, found the same pivot from volume to yield, and no H2 event reverses it.
• The cost base stepped up. RAM, storage, and labor repriced; the honest tier absorbed it and the captive tier passed it through with margin, and the November .com increase plus the AI Act’s first invoices are already scheduled additions.
• The machine layer is real. Agent-addressable commerce, priced crawling, and MCP surfaces all shipped in production during the half, whatever one thinks of the revenue attached.

What H2 decides is the durability of each:

• whether the renewal-extraction model survives contact with New York-style disclosure and 5x multipliers without a churn event
• whether the AI registration wave renews or evaporates (our cohort will say)
• whether the GPU spot market’s firm prices survive the surplus sellers (our October index edition will say)
• whether the frozen cloud tier finally thaws under the RAM shock, which would be the first industry-wide list-price inflation in seven years

The calendar is dense and mostly already written: August 2 (AI Act), August 15 intentions for the EU’s data-center labels, September 11 (CRA reporting), September’s cPanel licensing window, October’s Q3 prints, November 1 (Verisign, and our Renewal Markup Index baseline against it), December 2 (content-marking). This publication’s promise for H2 is the same one that organized H1: fewer adjectives, more instruments, and every number re-checkable at its source. The half-year repriced hosting. The next one prices in whether anyone churns.

How This Synthesis Was Built

This article synthesizes roughly forty pieces of our own H1 2026 business reporting; every figure was verified at its primary source when originally published, and the underlying articles name those sources individually. They draw on SEC filings, registry and registrar accounts, regulators’ texts, public industry datasets (NAIC, TrendForce as reported in trade press, Pew Research, HTTP Archive, BLS), and our own datasets published alongside the articles, from the 736-company audits to the renewal cohort. Where a figure was reported rather than filed, such as the WebPros loan or the Anthropic-SpaceX lease total, it carried that label originally and carries it here. Ledger dates are the dates of the underlying events. The interpretive frame, growth by invoice rather than by customer, is ours; the numbers it rests on are not.