Happy Friday. The Akamai-Anthropic deal announced Wednesday is the kind of contract that reorders a company's identity overnight: $11.6 billion, seven years, and a potential expansion to $20 billion. It also signals how much dedicated infrastructure the AI model companies need, and how fast that demand is moving.
The other two stories today connect directly. AI-generated code is spreading across enterprise IT faster than the controls to manage it, and the security risk is compounding as dependency sprawl grows. One company, Fabrix.ai, shipped a product this month aimed squarely at that gap.
CLOUD & INFRASTRUCTURE
Akamai's $11.6B Anthropic Pact Is the Largest in Its 27-Year History
Akamai Technologies signed a seven-year, $11.6 billion take-or-pay contract with Anthropic on Sept. 24, committing to supply dedicated CPU-based cloud computing capacity and managed support services. The deal carries a $5.5 billion capital expenditure tag, more than six times Akamai's total capex for all of 2025. No revenue hits the books until the second half of 2027.
Watch the warrant structure. Akamai gave Anthropic an equity warrant on nonvoting convertible preferred stock, with roughly 2% vesting on the current commitment and additional vesting tied to each $3 billion Anthropic spends. CEO Tom Leighton called this a first for Akamai. It is a bet that aligning financial incentives keeps Anthropic from walking to a competitor when alternative infrastructure options multiply. For software companies evaluating cloud vendors, this deal is a reminder that the largest AI workloads are moving toward dedicated, contracted capacity rather than on-demand consumption.
SECURITY & PRIVACY
Your AI Coding Tools Are Adding Dependencies Faster Than You Can Audit Them
AI coding assistants are pulling in third-party packages at a rate that outpaces traditional security review. Each accepted suggestion can add one or more external libraries to a codebase. Multiply that across a team and a quarter of development, and the dependency count grows well beyond what periodic audits can catch. Attackers use typosquatting to place malicious packages with names close to legitimate ones, and AI tools do not verify package provenance before recommending them.
The practical fix is not complicated, but it requires deliberate policy. Restrict package pulls to a pre-approved internal registry. Enforce automated dependency scanning in CI/CD pipelines. Require human sign-off on any new dependency an AI tool introduces. The coverage gap most teams face is that their existing software composition analysis tools are reactive, flagging known vulnerabilities only after packages are already in the build. AI-accelerated development means that lag now carries more consequence than it did when engineers wrote every line by hand.
ENTERPRISE SOFTWARE
Fabrix.ai Targets the Governance Gap as Vibe Coding Spreads Beyond Engineering
Fabrix.ai shipped Governed VibeOps on Sept. 9, a platform that inspects AI-generated code and meters token spend across enterprise IT stacks. The timing follows data showing 65% of vibe-coded apps contain security issues and 58% carry critical vulnerabilities, per an Escape.tech scan of more than 1,400 apps. Gartner projects 40% of new enterprise software will be vibe-coded by 2028, and 91% of enterprises currently lack an AI governance framework, per McKinsey.
The detail that should concern software executives most is this: 63% of vibe coding users are not developers. Marketing and sales staff are committing code alongside engineers, often with no review process in place. Fabrix.ai's three small language models, the Argos family, are trained on a customer's own environment rather than on frontier models, which the company argues reduces cost and improves precision for operations-specific tasks. Whether the governance tooling market consolidates quickly or fragments across dozens of point solutions is the question worth watching as enterprises move from awareness of the problem to actual procurement.
Created by the robots at The SaaS Sentinel