7 min read · 13 sources

Enterprise IT is hitting two opposing walls at once: vendors demanding mandatory cloud shifts on one side, and the physical limits of network and software architecture pushing back on the other. At the 27th annual DSAG congress, European enterprises made it clear they are not going quietly into SAP’s public cloud corral.

At the same time, the reality of running high-scale modern infrastructure is exposing cracks in conventional tooling. Distributed version control is breaking under machine-generated commit volumes, live sports streaming is crushing core transit exchanges, and idle database connections are quietly doubling their query latency over basic TCP behavior.

Here is what is breaking across enterprise systems and the infrastructure supporting them.

Driven by autonomous coding loops, Anthropic’s commit volume surged from 2,500 daily merges in May to over 30,000 per day in September, breaking standard Git workflows.

SAP Customers Push Back on Mandatory Cloud Timelines

Source: cio.com ↗

At the annual DSAG congress, German-speaking enterprise users openly pushed back against SAP’s cloud-first product roadmap. Enterprise architects and IT leaders demanded that the vendor stop reserving core product innovations exclusively for its public cloud SaaS offerings.

The resistance comes down to cold deployment reality: according to DSAG’s early 2026 investment survey, 54% of member enterprises still run SAP ECC or SAP Business Suite. Mainstream maintenance for ECC is scheduled to terminate in late 2027. Migrating complex, highly customized on-prem ERP cores to S/4HANA Public Cloud represents massive capital expenditure with disputed ROI, leaving operations teams caught between vendor deadlines and functionally stable legacy systems that cannot simply be lifted and shifted overnight.

Big Tech Claims Europe’s Cloud Sovereignty Will Cost €115 Billion

Source: assets.theregister.com ↗

A study commissioned by a US tech industry lobby warns that the European Union’s aggressive push for digital and cloud sovereignty could cost the bloc €115 billion over the next ten years. The report models the financial impact of restrictive data localization policies, mandatory switching away from US hyperscalers, and compliance overhead driven by frameworks like DORA and GDPR.

While the figures stem from modeled industry scenarios rather than realized accounting, the underlying infrastructure tension is real. European enterprise architects face stringent regulatory demands to isolate operational data, while local sovereign alternatives often lack the sheer service breadth, global edge presence, and pricing discounts of AWS, Azure, and Google Cloud.

Champions League Matches Push DE-CIX Frankfurt to 21.47 Tbps Peak

Source: computerweekly.com ↗

Internet exchange operator DE-CIX logged historic peak traffic on October 13 driven by live-streamed UEFA Champions League matches, hitting an all-time record of 21.47 Tbps at its flagship Frankfurt exchange. Aggregate throughput across all DE-CIX nodes crossed 32 Tbps.

The Frankfurt node had just crossed the 20 Tbps threshold 25 days earlier, marking an unprecedented 7% leap in peak transit load in under a month. For edge engineers and CDN operators, this rate of peak streaming demand highlights accelerating capacity exhaustion on regional backbones and internet exchange point (IXP) interconnects during unbuffered live events.

Anthropic Drops Git for Scion After AI Merges Hit 30,000 a Day

Source: anthropic.com ↗

Autonomous AI loops have broken standard developer tooling at Anthropic. The company revealed it abandoned Git and built a custom VCS named Scion to support its engineering monorepo. Driven by automated Claude workflows writing and committing code around the clock, internal merge volume exploded from 2,500 daily commits in May 2026 to more than 30,000 per day by September.

Git’s distributed architecture choked under the throughput. Anthropic replaced it with Scion, a centralized system built on the assumption that clients are always online. Scion implements sparse checkouts, server-side commit synthesis, and horizontal scale across server nodes. The custom VCS reduced CI checkout times from several minutes down to seconds while cutting memory and CPU utilization across their developer infrastructure by 200x.

Fixing an Idle 32 ms Tail Latency Spike in ClickHouse

Source: clickhouse.com ↗

ClickHouse published a deep dive into an edge case where simple native TCP SELECT queries across high-latency connections doubled their expected round-trip time, jumping from ~32 ms to ~64 ms after an idle period.

The culprit was an interaction between the OS TCP stack and database protocol framing. During idle periods, the kernel resets the TCP congestion window (cwnd) down to its initial size. When a query ran, ClickHouse’s native protocol emitted query data followed by a stream of unbatched trailing metadata frames (ProfileEvents, Progress, Logs) and an EndOfStream packet. Even with TCP_NODELAY enabled, this cluster of small metadata packets instantly blew through the constricted cwnd. The connection had to pause and wait an entire extra RTT for an ACK before flushing the final bytes. ClickHouse eliminated the latency penalty by batching metadata packets at the application layer before handing them to the socket.

Idle Connection -> cwnd resets to initial window (e.g., 10 segments)
Query Response  -> Data + Unbatched Metadata (Progress, Logs, EndOfStream)
Result          -> Exceeds cwnd -> Stalls pipeline -> +1 RTT (32 ms penalty)
Fix             -> Application-layer frame batching -> Fits within initial cwnd

The Rise of Dedicated "DeployCos"

Source: tanayj.com ↗

Hyperscalers and AI labs are pouring billions into dedicated deployment service entities - dubbed DeployCos - to embed forward-deployed engineers (FDEs) directly inside traditional enterprises, as analyzed by Tanay Jaipuria.

The capital allocation is massive:

• Anthropic teamed with private equity to launch Ode with Anthropic with $1.5 billion in funding.
• OpenAI established a majority-owned $4 billion venture alongside TPG, fielding over 100 FDEs.
• AWS has earmarked $1 billion to spin up an in-house forward-deployed engineering group.

These entities aim to close the “capability overhang,” where off-the-shelf frontier models stall out during enterprise deployment due to complex internal data plumbing, strict permission boundaries, and brittle legacy integrations.

Firefox Ships an Enterprise-Managed Browser

Source: computerworld.com ↗

Mozilla launched Firefox Enterprise, pairing its independent Gecko rendering engine with a self-hostable centralized management console. The deployment targets security-conscious shops looking to escape the Chromium and WebKit duopoly.

The platform provides native data loss prevention (DLP), policy controls for enterprise AI tool access, local key management for encryption, and direct integration with SIEM platforms for audit events. By offering self-hosted policy infrastructure, Mozilla is specifically targeting European environments navigating DORA and GDPR constraints that discourage feeding endpoint browser telemetry into third-party cloud managers.

Enterprise Agent Frameworks Land at Workday, Shopify, and SolarWinds

Source: constellationr.com ↗

Agentic architecture is arriving across major operational platforms:

• Workday: At Workday Rising, the company established Sana Home as its primary user interface and introduced Agent Harness. The framework enables engineers to write context-aware HR and finance agents using TypeScript or Google’s Agent Development Kit (ADK), establishing a governed architecture for ambient agents to access enterprise records via consumption-based Flex Credits.
• Shopify: The platform opened access to its 1-billion-plus product catalog to Agentic Storefronts, letting tools like ChatGPT, Gemini, and Copilot parse inventories and stage carts through Shop Pay. The critical operational constraint: checkout automation remains strictly restricted, requiring mandatory, human-in-the-loop purchase validation.
• SolarWinds: Network ops tooling is getting its own agentic runtime with SolarWinds Avior. Designed for hybrid deployments, the agent parses telemetry to detect drift, correlate alerts, and execute runbook remediation across assistive, human-approved, or fully autonomous operating modes.

UEC 1.1 Drops Packet Overhead for GPU Fabrics

Source: networkworld.com ↗

The Ultra Ethernet Consortium released the UEC 1.1 specification, targeting the scale-up networking fabrics that tie AI accelerator clusters together.

The update tackles hardware-level bottlenecks:

• Unified Forwarding Header: Strips down the 66-byte framing overhead imposed by standard IPv4/UDP encapsulation for intra-rack traffic.
• Programmable Congestion Management (PCM): Allows custom telemetry and pacing algorithms to preempt tail latency spikes during heavy collective operations.
• In-Network Collectives (INC): Implements hardware-based aggregation directly on switches to accelerate distributed model reduction steps.

Silicon and API Upgrades: Cisco, Nvidia, and Google

Source: assets.theregister.com ↗

• 12.8 Tbps Optics: Cisco and Nvidia published a joint proposal for a new optical transceiver form factor designed to deliver up to 12.8 Tbps per pluggable module. The design targets high-density AI cluster fabrics to maximize inter-switch bandwidth while reigning in thermal and power envelopes.
• Gemini Long Decode: Google Cloud introduced Long Decode for Gemini 4 Argon. To prevent HTTP client timeouts during massive reasoning jobs that can span an entire 1M token context window (1,048,576 tokens), Long Decode segments inference passes into 64K token chunks connected by encrypted state continuation tokens via the GenerateContent API.