Equinix Metal

Equinix Metal no longer exists as a purchasable service. Equinix stopped selling the on-demand bare metal platform, ran it in maintenance mode, and sunset it entirely on 30 June 2026 — any servers still provisioned after that date were permanently deleted, and access to instances ended at 11:59 PM PDT on the final day. If our bare metal cloud category is a shopping list, this entry is the receipt from a shop that closed. We keep the review up for people migrating off it, for portfolio history, and because the shutdown is itself the most useful thing this product can now teach a buyer.

What Equinix Metal actually was

The product began life as Packet, founded in New York City in 2014 by Zachary Smith and Jacob Smith, one of the first platforms to sell single-tenant bare metal servers with cloud-style automation: API-first provisioning in minutes, hourly billing, Terraform and CLI tooling, custom hardware flavours including ARM, and a reputation for engineering quality well above its size. Equinix acquired Packet in a deal reported at $335 million, completed on 3 March 2020, and rebranded it Equinix Metal — the pitch being bare metal deployed inside Equinix's carrier-neutral footprint, with private interconnects to the major clouds and to the rest of Platform Equinix.

At its height the offering was genuinely good infrastructure: x86 and ARM servers across global metros, on-demand provisioning measured in minutes (the marketing claimed 60 seconds), private VLANs and VRFs, dedicated ports, Equinix Fabric virtual connections, block and object storage options, IPv4 and IPv6 addressing, and a developer surface — API, CLI, Terraform provider, Kubernetes integrations — that many larger competitors envied. Third-party trackers near the end listed bare metal starting around $535 per month, with on-demand usage billed in blocks of at least one hour and reserved hardware carrying a 5–50% discount depending on term length. There was no free tier; proof-of-concept credits were a sales conversation.

Pricing: what you really pay

Nothing — you cannot buy it. The commercial sale of Equinix Metal stopped before the shutdown announcement became public knowledge, and the final billing runs completed in mid-2026. What remains is a cautionary tale about price structure: Metal was priced at the premium end of bare metal, and the audience it needed — price-sensitive developers and platform teams choosing between $535-per-month metal and cheap dedicated servers elsewhere — kept migrating to lower-cost options while the hyperscalers absorbed the enterprise automation spend. Reporting on the wind-down cited exactly that dynamic: competition from established hyperscalers made the market hard to hold.

For buyers reading this while planning a migration, the pricing lesson transfers. Premium bare metal only survives where the price is justified by interconnection, compliance or hardware choice; when those are absent, the market moves to the cheapest credible automation. Compare the total: server rate, bandwidth and egress, IP charges, and the cost of the interconnects you were actually using. Metal's own end-of-life notice warns that IP addresses, VLANs, VRFs, Fabric virtual connections and dedicated ports continue to bill if not deleted — infrastructure line items outliving the servers they served is not a metaphor, it was written into the shutdown FAQ.

Performance & infrastructure

We never benchmarked Metal in our benchmarks programme, so we have no hands-on numbers to report — and there is now no way to produce any. What can be said from documentation and the product's public record: it was single-tenant hardware with cloud-style automation, so performance was the hardware's own (EPYC and Xeon classes, NVMe storage, up to 64-core machines in the larger plans), not a virtualised approximation. The infrastructure envelope was Equinix's — carrier-neutral facilities, N+1 power, dense peering — which is why the interconnection story was the main selling point: private links to AWS, Azure and Google Cloud via Equinix Fabric, and cross-connects into the same buildings as networks and content platforms.

The operational endgame was handled with more care than many platform shutdowns get. Equinix committed to performance, security and stability updates through the sunset period rather than freezing the product; published a dated end-of-life timeline (final day of service 30 June 2026, resources removed 1 July 2026, owners downgraded to billing-only roles, collaborators removed, final invoices issued for June usage); ran a migration survey and account-team support for transitions into Equinix colocation, managed services or third-party providers; and gave reserved-server customers a defined month-to-month path to cancellation with seven days' notice to stop renewal billing. That is a professionally run shutdown. The status page and support mailbox were kept alive through the transition, which is the minimum a platform owes its customers when it switches the lights off. It is still a shutdown: workloads moved or died, and anyone who ignored the timeline lost data.

Where it falls short

It is discontinued, which supersedes every other critique — but the reasons matter for anyone choosing a platform today. First, vendor concentration risk: a product living inside a much larger company is subject to portfolio decisions that have nothing to do with its quality. Metal's engineering reputation never really mattered against Equinix's strategic arithmetic. If you build on a single-vendor platform that is not that vendor's core business, you own that risk; this shutdown is the case study to cite in your own architecture reviews.

Second, price-to-market mismatch. At roughly $535 per month for entry bare metal, Metal asked premium rates for commodity hardware plus automation. The automation was excellent; the market increasingly expected it for free or near-free. Third, positioning drift: rebranded into the Equinix fold, the product sat awkwardly between colocation, Fabric and the hyperscalers it connected to, and the ideal customer — someone who wanted Equinix interconnection but not a colocation contract — was precisely the customer most courted by alternatives. Finally, for anyone still finding stale pages promising 25 global metros and 60-second provisioning: those claims described a service that has been switched off. Marketing outlived the product; we would rather our database did not.

Who should pick this

Nobody. The honest answer for this section, and we are not going to dress it up. Former Metal customers who have not yet migrated should treat every remaining resource — instances, IPs, VLANs, VRFs, Fabric connections, dedicated ports — as cost and risk to be closed out, and should have received their final invoices by now. If you are mid-migration and want the closest modern equivalents, the shortlist in our bare metal cloud category and our benchmarks are the right places to start: look for providers with published pricing, API-driven provisioning and a business model where bare metal is the core product rather than a side line. If what you actually valued was Equinix interconnection, the company's own answer is colocation or managed services — see our Equinix Colocation entry for what that looks like.

Migration specifics worth carrying across: audit what you actually consumed beyond the server rate — reserved capacity terms, IP allocations, VLANs, Fabric virtual connections and dedicated ports were the line items that complicated exits — and price the replacement on the same basis. Keep the provisioning code, swap the provider layer, and treat the forced move as the moment to standardise on a provider whose core business is bare metal. The broader buyer's lesson, which is why this page stays: before standardising on a platform, check whether the product is the vendor's main event, read the exit path in the contract, and keep infrastructure definitions in code you control so a sunset is an inconvenience instead of an outage.

Key takeaways

• Equinix Metal was sunset on 30 June 2026 after Equinix stopped commercial sales; un-migrated servers were permanently deleted and access ended at 11:59 PM PDT on the final day.
• Lineage: Packet, founded 2014 in New York by Zachary and Jacob Smith; acquired by Equinix for a reported $335 million, completed 3 March 2020; rebranded Equinix Metal.
• The product itself was strong — API-first bare metal, hourly on-demand billing (minimum one-hour block), reservations at a 5–50% discount, Terraform/CLI tooling, ARM and x86, Equinix Fabric interconnects.
• Near the end, trackers listed bare metal from about $535 per month — premium pricing in a market drifting to cheaper automation and hyperscaler dominance.
• Shutdown mechanics were professional (dated timeline, migration support, maintenance until the end) but absolute: IPs, VLANs and Fabric connections kept billing if not deleted.
• Buyer's lesson: platform risk is vendor-priority risk; keep infrastructure in code you own and know the exit path before you commit.

Sources

• https://docs.equinix.com/metal/
• https://deploycue.com/providers/equinix-metal
• https://web.archive.org/web/2025id_/https://deploy.equinix.com/metal/pricing/
• https://newsroom.equinix.com/2020-03-03-Equinix-Completes-Acquisition-of-Bare-Metal-Leader-Packet
• https://news.crunchbase.com/startups/equinix-closes-355m-deal-for-new-yorks-packet
• https://datacenterdynamics.com/en/news/equinix-to-kill-off-metal-by-june-2026
• https://finance.yahoo.com/news/equinix-closing-bare-metal-iaas-160000028.html
• https://tracxn.com/d/companies/packet/__ZDKQmtGfpIlFk8-a4RPNNZ0Pl7_ki1IkK4T9EipWQQ