AT&T Adds 400G Wavelength to a Dozen Metros, Tightens AWS Interconnect Grip

AT&T's 400G Expansion Changes Metro Wavelength Economics

AT&T Business expanded 400G wavelength service to more than a dozen metro locations with edgeless handoff capabilities, moving dense metro connectivity from premium to mainstream procurement. The upgrade matters for enterprises building higher-capacity metro-to-metro and cloud-adjacent links because it reduces the number of circuits needed and simplifies port management. The trade-off: committed spend thresholds rise, making 400G harder to justify for smaller deployments.

The move puts pressure on Verizon Business, Lumen, Zayo, and Colt, which still compete on dense metro wavelength and cloud adjacency. Buyers planning AI, backup, replication, and multi-cloud traffic can now procure 400G as a standard option rather than a special build, which changes both circuit design timelines and vendor negotiation leverage.

AWS Adds AT&T to Interconnect, Bypasses Aggregators

AWS integrated AT&T into its Interconnect service, giving AT&T a direct path into AWS connectivity products and reducing the role of intermediary providers. The integration is a service tie-in, not a vague partnership — AT&T's network now extends into AWS Interconnect, which shifts the competitive field against Equinix Fabric, Megaport, and Cologix.

For buyers, this changes circuit design, procurement timing, and resiliency planning. AT&T becomes a more direct option for AWS adjacency, potentially lowering operational complexity by removing aggregator layers. Enterprises using AWS should reassess whether their current cloud connectivity architecture still makes sense or whether a direct AT&T path reduces both cost and latency.

DriveNets Targets AI Fabric with 1.6T Platform

DriveNets unveiled a 1.6T AI fabric platform using Broadcom silicon, designed to connect thousands of XPUs and reduce latency for AI workloads. The platform challenges AI networking stacks from NVIDIA, Arista, Juniper, Cisco, and Broadcom-based white-box ecosystems, especially where fabric scale and latency are the buying criteria.

Buyers building AI clusters should expect more procurement emphasis on interconnect speed, oversubscription ratios, and east-west congestion, not just compute pricing. If DriveNets' claims hold, spend could shift from traditional chassis-based networking toward fabric architectures optimized for massive GPU and accelerator clusters. The risk: fabric maturity and operational tooling lag behind incumbent platforms, which can increase integration and troubleshooting costs in the first 12 months.

Nvidia-Backed Upscale Raises $190 Million, Intensifies Switch Silicon Competition

Nvidia-backed AI networking startup Upscale raised $190 million, signaling more vendor churn and faster product cycles in AI interconnect. The funding intensifies pressure on incumbent switch silicon and AI fabric vendors, including Broadcom, Cisco, Arista, and Marvell, because Nvidia continues to shape the vendor ecosystem around AI networking even when it is not selling the entire stack itself.

For enterprise buyers, this creates both better performance options and a short-term risk of fragmentation or lock-in around emerging ecosystems. Procurement teams should track which hyperscaler and cloud provider partnerships Upscale announces next — those will determine whether the platform becomes a viable alternative to Broadcom Tomahawk or a niche play with limited interoperability.

Subsea Buildouts Add Route Diversity in Asia and Australia

Lightstorm launched an India–Malaysia–Singapore subsea cable system with Microsoft, Singtel, and Tata Communications, adding route diversity and potentially lower latency between India and Singapore. For multinational buyers operating across Asia, the practical effect is more cloud connectivity options and reduced single-path risk in a region where subsea resilience matters for disaster recovery and application performance. The system competes directly with PCCW Global, Telstra, SubCom-supported systems, and other hyperscaler-backed cable ecosystems.

SubCo's SMAP subsea system reached ready-for-service status, connecting four Australian cities, with a Tasmania extension planned for 2027. Enterprises with Australian operations gain another path for intercity diversity and disaster recovery planning, which can influence procurement of protected connectivity and lower outage concentration risk. The system increases pressure on Telstra, Optus, Vocus, and other Australian metro and intercapital backhaul providers.

What to Watch

Google is planning a connectivity hub in Eastern Cape, South Africa — the first of four hubs in Africa — which signals continued hyperscaler investment in regional interconnect infrastructure. Buyers with African footprints should watch this because new hyperscaler hubs can reduce latency and transit dependence while changing where enterprises place regional workloads and internet exchange budgets.

The broader pattern: AI interconnect, cloud adjacency, and subsea diversity are moving faster than enterprise procurement cycles. Buyers who locked in three-year wavelength or cloud connectivity contracts 18 months ago should revisit their architectures now, before the next budget cycle makes mid-contract changes harder to justify.