GCP Pricing Changes in 2026: Every Update That Affects Your Bill

The single most consequential Google Cloud pricing event of 2026 is a networking egress increase that took effect May 1, 2026: list prices for data transfer out through CDN Interconnect, Direct Peering, and Carrier Peering roughly doubled in North America (from $0.04/GB to $0.08/GB), with smaller but still substantial increases in Europe ($0.05 → $0.08/GB) and Asia ($0.06 → $0.085/GB) (Google Cloud, CDN Interconnect and Peering pricing update). Google cited continued investment in global network capacity as the rationale and did not raise Dedicated Interconnect or Partner Interconnect pricing in the same change.

On the same date, Google increased pricing for A3 Ultra GPU VMs in Europe and Asia — Google's public announcement page for the change did not publish the exact percentage or new rate at the time of writing, so teams running A3 Ultra workloads outside North America should check the GPU pricing page directly for their region rather than relying on a cached quote. For a side-by-side view of where GCP undercuts or loses to AWS and Azure, see our AWS vs GCP pricing comparison and Azure vs GCP pricing comparison.

The other structural 2026 change is on the commitment side: spend-based Committed Use Discounts began migrating to a new billing data model on January 21, 2026. Google states the change does not increase total cost, but it does change how CUD savings appear in billing exports and BigQuery data — usage now bills directly at a discounted "consumption model" rate instead of list price offset by separate credits (Google Cloud, spend-based CUDs). Teams with automated billing pipelines or finance dashboards built on the old credit schema need to update them, or savings will appear to vanish even though the underlying discount is unchanged.

Three confirmed changes have landed through 2026 so far. Check back or subscribe for updates as Google announces more — Google Cloud typically notifies affected billing accounts by email ahead of pricing changes, but (like AWS and Azure) does not surface a persistent in-console warning for every affected resource.

Q1 2026 (January – March)

Q2 2026 (April – June)

Networking is where the two biggest 2026 GCP stories intersect: a price increase on one egress path, and a long-standing waiver on another. The May 1, 2026 increase applies specifically to CDN Interconnect, Direct Peering, and Carrier Peering — not to standard internet egress, and not to Dedicated Interconnect or Partner Interconnect (Google Cloud). If your team does not use direct or carrier peering, this specific change does not touch your bill — but it is worth confirming that in Cloud Billing rather than assuming it.

Separately, and unrelated to the May 2026 increase, Google has offered free data transfer for customers fully exiting Google Cloud since January 2024 — covering services including BigQuery, Cloud Bigtable, Cloud SQL, Cloud Storage, Datastore, Filestore, Spanner, and Persistent Disk. The waiver requires Google Cloud Support review and approval, and customers then have a limited window (60 days) to complete the exit. It applies only to the one-time bulk transfer when leaving — it does not reduce the ongoing operational egress a running workload generates day to day. This policy is unchanged in 2026 and predates the May 1 peering price increase by over two years.

For a deeper breakdown of how egress pricing compares across the big three providers, see our cloud egress costs guide.

On the GPU side, the May 1, 2026 A3 Ultra price increase in Europe and Asia is the clearest 2026 compute change — but Google's public pricing-change page does not state the magnitude, so the only reliable figure is whatever the GPU pricing page shows for your region today. Treat any third-party estimate of the exact percentage with caution until Google publishes one directly.

On the commitment side, the January 21, 2026 spend-based CUD migration is a reporting change rather than a rate change, but it has real operational weight for FinOps teams. The old model billed at list price and applied credits; the new model bills directly at a discounted rate tied to a "consumption model" (for example, "Compute Flexible CUD — 1 Year") that Google attaches to each eligible SKU. Google also notes the change expands the scope of usage eligible for some spend-based CUD discounts, meaning a larger share of spend may now qualify (Google Cloud Docs). Resource-based CUD sharing also defaults to enabled for any Cloud Billing account created on or after June 16, 2026, making it easier for multi-project organizations to pool a single commitment automatically rather than configuring sharing manually. For a broader breakdown of commitment models across providers, see our reserved vs spot vs on-demand comparison.

Practically, teams with BigQuery billing-export dashboards or automated CUD-purchasing scripts built against the pre-2026 schema should treat this as a required update, not an optional one — a dashboard that suddenly shows "zero CUD credit" after January 21, 2026 is very likely a schema mismatch, not a lost discount.

Gemini API pricing in 2026 is structured around a fast-moving Flash line and a higher-cost Pro line, with promotional introductory rates on the newest Flash models. Gemini 3.6 Flash and Gemini 3.7 Flash are priced at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, stepping up to $1.50/$7.50 per million tokens afterward — a published future increase, not a hypothetical one (Gemini API Pricing). Gemini 3.5 Flash holds steady at $1.50/$9.00 per million tokens, and Gemini 2.5 Flash-Lite remains the cheapest general-purpose option at $0.10/$0.40 per million tokens.

On the Pro tier, Gemini 2.5 Pro bills $1.25/$10.00 per million tokens for prompts up to 200K tokens, doubling to roughly $2.50/$15.00 for the portion of a prompt above that threshold — a detail that matters for long-context RAG or document-analysis workloads that regularly cross the 200K boundary without realizing it changes the effective rate.

Vertex AI wraps the same underlying Gemini models with enterprise deployment options, including a Priority tier for tighter latency SLAs and dedicated capacity at a premium over Standard tier pricing. For teams evaluating whether to route through the Gemini Developer API directly or through Vertex AI, the model-level token rates are the same starting point — the difference is deployment guarantees and enterprise billing integration, not the underlying per-token cost. For a broader look at inference pricing across providers, see our LLM inference cost per million tokens breakdown.

Batch processing continues to offer a 50% discount on standard token rates for workloads that can tolerate asynchronous turnaround, and context caching is available at a fraction of standard input pricing for repeated long-context prompts (such as system instructions or RAG context windows), plus a per-hour storage charge for cached content. Teams with high-repetition prompt patterns should model caching against their actual cache-hit rate before assuming it is a guaranteed discount — the storage charge applies whether or not the cache is reused.

GKE's cluster management fee remains $0.10 per hour per Standard or Premium cluster — roughly $73/month — billed by the second once any included credit is exhausted (GKE Pricing). Every Cloud Billing account receives a $74.40 monthly credit, which is enough to fully cover the management fee for one small eligible cluster — effectively one free cluster per billing account, not per project.

As with AWS EKS's flat per-cluster fee and AKS Standard/Premium's management fee, GKE's cluster fee rewards consolidation. Teams running separate clusters per environment (dev, staging, prod) only receive the $74.40 credit once per billing account in total, not once per cluster — so a three-cluster setup pays the management fee in full on two of the three clusters. Using namespaces and RBAC to isolate environments within fewer clusters avoids stacking that fee for workloads that do not need dedicated control planes. For a full three-way comparison of managed Kubernetes pricing, see our EKS vs AKS vs GKE pricing guide.

Autopilot mode bills per-pod resource requests rather than per-node capacity, which removes node-level bin-packing waste but means every CPU, memory, or ephemeral storage request directly drives cost — over-requesting resources in a pod spec has an immediate and visible bill impact under Autopilot in a way it does not under Standard mode's node-based billing.