Aptum Kills Egress Fees as Modal Labs Raises $750M for Multi-Cloud GPU Access

Egress fees eliminated in new hybrid-cloud offer

Aptum launched Aptum Prizm on October 1, a hybrid-cloud management platform with no data egress charges. The service includes Shared Cloud priced per vCPU, RAM gigabyte, and storage gigabyte, and Dedicated Cloud using flat monthly host pricing. Bandwidth is included within network-throughput limits, and Prizm management is bundled at no additional cost.

For buyers, this directly attacks the hidden cost in multi-cloud: data transfer. AWS, Azure, and Google Cloud charge egress fees that penalize workload portability and cross-cloud replication. An application moving 10 TB monthly between clouds can incur $900 in transfer costs at standard hyperscaler rates before any compute charges. Aptum's model removes that penalty, but buyers must verify throughput caps, whether the economics apply to transfers between Aptum and hyperscalers, and SLA coverage for network performance.

The risk is replacing one lock-in with another. No egress fees lower the financial cost of exit, but operational dependence shifts to Aptum's control plane. Contract reviews should cover exit assistance, portability of infrastructure-as-code configurations, and pricing stability after initial terms expire. Rackspace, Kyndryl, and IBM offer competing managed hybrid-cloud services; Aptum's differentiation is eliminating a line item most competitors still charge.

$750M Series D funds multi-cloud GPU orchestration

Modal Labs raised a $750 million Series D on October 14 at a $15.75 billion post-money valuation, led by Accel. Total disclosed funding now stands at approximately $1.22 billion, up from a $4.65 billion valuation in May. Modal's platform scales from zero to thousands of containers or hundreds of GPUs in seconds, automatically selecting capacity across AWS, Google Cloud, and Oracle Cloud Infrastructure. It allows customers to apply existing hyperscaler committed spend through marketplace integrations.

The buyer case centers on GPU scarcity. Enterprises building AI platforms face weeks-long waits for hyperscaler GPU capacity and must manage separate operational workflows for each provider. Modal abstracts that complexity, offering burst capacity without negotiating with every cloud. The platform fee must be weighed against the cost of direct reservations and the time value of faster deployment.

The valuation tripled in five months, signaling investor confidence but raising supplier-concentration risk. Procurement teams should demand transparent GPU-hour pricing, accelerator and region availability, workload-placement rules, data-transfer charges between clouds, and portability of containers and model artifacts. Modal competes with CoreWeave, Lambda, Crusoe, RunPod, and hyperscaler-native services like SageMaker, Vertex AI, and Azure Machine Learning. Its differentiation is the abstraction layer; its risk is whether that abstraction becomes another choke point.

Anthropic commits $530B across 10+ clouds for GPU capacity

Anthropic signed data-center leasing agreements totaling more than 15 gigawatts across over 10 cloud providers, including AWS, Google, and neocloud provider IREN, with a reported aggregate value of approximately $530 billion. Its computing capacity is expected to rise from 1.4 GW in 2025 to 5 GW in 2026, 9.5 GW in 2027, and more than 15 GW in 2028. Infrastructure advisory firm Occam Edge estimates Anthropic spends about $200 million per month renting chips.

This is not a conventional SaaS announcement, but it validates multi-provider infrastructure as a strategy for securing GPUs and reducing dependence on a single supplier. For enterprise buyers, the implication is that long-duration, multi-gigawatt capacity commitments improve availability but create utilization risk if model demand, accelerator economics, or chip generations change. Flexible capacity contracts matter more than multi-cloud as a label.

89% multi-cloud adoption, 67% cite management overhead

Flexera's 2026 State of the Cloud report indicates 89% of enterprises use multiple cloud providers, while 67% cite management overhead as a primary concern. The global cloud infrastructure market is estimated at $247 billion in 2026, with hybrid and multi-cloud environments representing 72% of enterprise deployments. Organizations reportedly achieve average savings of 23% through multi-cloud optimization.

Multi-cloud is mainstream, and management cost is real. The 67% overhead figure argues for funding a central platform, FinOps capability, policy automation, and observability rather than treating additional clouds as free optionality. VMware Cloud Foundation, Red Hat OpenShift, Google Anthos, Azure Arc, and HashiCorp compete in this space.

What to do

Compare compute, storage, network transfer, committed-use discounts, platform fees, and operational labor across providers. Treat GPU access as a supply-chain issue: multi-cloud GPU brokers and neoclouds may offer faster access than hyperscalers, but contracts need capacity guarantees, accelerator substitution rules, and data-residency terms. Demand portability evidence, not portability claims. No-egress pricing and automated orchestration reduce two barriers, but both create new dependencies that require contract and operational scrutiny.