AI Reshapes Infrastructure and Identity

The technology industry is confronting infrastructure vulnerability at every level. Anthropic's new cybersecurity-focused AI model arrives the same day US agencies issue warnings about Iranian state actors targeting critical infrastructure. This timing reflects a broader inflection point where the abstraction layers that made software eat the world now demand reinforcement.

The infrastructure response takes multiple forms. Intel joining Musk's semiconductor manufacturing initiative signals recognition that chip supply chains remain a strategic weakness despite years of reshoring rhetoric. Meanwhile, AI-driven layoffs affecting nearly 80,000 workers in Q1 reveal how automation is now restructuring the employment infrastructure that sustained the industry's growth.

What connects these developments is a shift from expansion to consolidation. The 2010s prioritized speed and scale. The late 2020s prioritize resilience and control. AI models defend networks. Nation-states probe weaknesses. Companies rebuild domestic manufacturing. Workforces contract. Each story represents a different vector of the same underlying force: technology infrastructure that scaled without sufficient attention to durability now faces simultaneous pressure to harden, localize, and automate. The question is whether these responses address root vulnerabilities or simply add complexity to systems already straining under their own weight.

Deep Dive

AI Security Models Create New Competitive Dynamics

The limited release of Anthropic's Mythos model reveals how frontier AI capabilities are becoming too powerful to release publicly, forcing companies into a partnership-based distribution model that fundamentally alters competitive dynamics in the industry. This represents a shift from open competition to selective collaboration, with significant implications for startups and investors.

Anthropic chose 12 partner organizations including Amazon, Apple, Microsoft, and CrowdStrike to deploy Mythos for defensive cybersecurity work. The model allegedly discovered thousands of zero-day vulnerabilities, many one to two decades old, in both proprietary and open source software. But the company explicitly will not make Mythos generally available, citing concerns that the same capabilities used to find bugs could be weaponized to exploit them. Only 40 organizations total gain access beyond the core partners.

This creates a two-tier system where the largest tech companies gain early access to capabilities that remain out of reach for everyone else. For startups, this means the playing field is no longer level. If you are not among the chosen partners, you cannot access the same security tools as your larger competitors. For VCs, this suggests concentration of value in companies with existing relationships to frontier AI labs. The partnership model also raises questions about how future AI capabilities will be distributed. If models become too powerful for public release, access becomes a strategic asset controlled by a handful of labs, fundamentally changing how innovation flows through the industry.

AI Automation Triggers First Major Employment Contraction

Nearly 80,000 tech workers lost their jobs in Q1 2026, with nearly half of those layoffs attributed to AI implementation and workflow automation. This marks the first quarter where automation, not market correction or overhiring, drives the majority of job cuts. The employment infrastructure that sustained tech's growth is restructuring faster than most companies anticipated.

The scale matters less than the composition. Previous layoff waves stemmed from pandemic overhiring corrections or economic downturns. This wave stems from operational transformation. Companies are not cutting excess capacity. They are replacing human labor with AI agents and automated workflows. The US accounts for 76.7% of these cuts, suggesting American companies are moving faster on AI adoption than global competitors. This creates a productivity advantage but also a talent retention problem as experienced workers leave the industry entirely rather than accept reduced roles.

For founders, this changes hiring calculus. The conventional wisdom of scaling headcount to match revenue no longer holds if AI tools can handle increasing workload without proportional headcount growth. For VCs, this means adjusting unit economics assumptions in portfolio companies and questioning whether traditional SaaS metrics still apply when customer success and support teams shrink by 40% while serving 2x the customers. For tech workers, this signals that survival requires shifting from task execution to oversight and strategy roles that AI cannot yet handle. The question is whether the industry creates enough of those higher-level roles to absorb displaced workers or whether this represents a permanent reduction in tech employment relative to output.

Intel's Foundry Bet Shows Vertical Integration's Return

Intel joining the Terafab semiconductor project with SpaceX and Tesla solves a puzzle about how two companies with zero chip manufacturing experience planned to build a fabrication facility. The answer is they do not. Intel does. This partnership reveals the limits of the last decade's disaggregation model and signals a return to vertical integration for strategic technologies.

The implications extend beyond this specific deal. Intel gains two large anchor customers for its struggling foundry business at a time when it desperately needs revenue to justify the massive capital expenditures required to compete with TSMC. SpaceX and Tesla gain access to custom chip designs without the decade-long learning curve and $20 billion investment required to build fab capacity. But the broader market impact is that fabless chip design, the dominant model since the 1990s, may be reaching its limits for companies requiring cutting-edge performance or sovereign supply chains.

For VCs, this suggests opportunities in companies that can navigate the middle ground between fully integrated manufacturers and pure-play designers. For founders in hardware or AI infrastructure, this partnership demonstrates that controlling your supply chain increasingly matters more than optimizing for capital efficiency. The semiconductor industry spent 30 years disaggregating. The next decade may involve selective re-integration as geopolitical tensions and strategic importance override pure economic optimization. Companies that can secure manufacturing partnerships or build internal capacity gain a structural advantage over those relying on merchant foundries with allocation constraints.

Signal Shots

Anthropic Scales Compute as Revenue Hits $30B Run Rate: Anthropic expanded its compute partnership with Google and Broadcom to 3.5 gigawatts, coming online in 2027, as its annualized revenue jumped from $9 billion in late 2025 to $30 billion. The company now serves over 1,000 enterprise customers spending at least $1 million annually. This represents the largest compute commitment in AI history and signals that enterprise adoption is accelerating faster than infrastructure can scale. Watch whether other frontier labs can secure similar capacity or if compute access becomes a durable competitive moat favoring companies with hyperscaler partnerships.

Perplexity's Usage Pricing Drives 50% Monthly Growth: Perplexity's estimated annual recurring revenue surged past $450 million in March, jumping 50% in a single month following the launch of new agent tools and a shift to usage-based pricing. The pricing model change appears to better align revenue with actual value delivered than flat subscription fees. This rapid acceleration suggests the market for AI-powered search and research tools is expanding beyond early adopters into mainstream enterprise use. Watch whether competitors follow with similar pricing models or if this advantage proves temporary as capabilities commoditize.

Eclipse Raises $1.3B for Physical AI Ecosystem Strategy: VC firm Eclipse closed a $1.3 billion fund split between early-stage incubation and growth investments in physical AI across transportation, energy, infrastructure, and defense. The firm plans to build an ecosystem of portfolio companies that partner with each other to achieve scale faster. This represents a shift from passive capital allocation to active company building and cross-portfolio collaboration. Watch whether this ecosystem approach generates better returns than traditional VC models or if the coordination overhead outweighs the benefits.

CISA Faces $700M Budget Cut Amid Rising Threats: The Trump administration proposed cutting CISA's budget by $707 million for 2027, reducing the agency's operating budget to roughly $2 billion despite escalating cyberattacks on federal systems. The administration claims the cuts will refocus the agency on core infrastructure security rather than election security programs it labels as censorship. This reduction comes as the agency has already lost hundreds of employees and operates without a Senate-confirmed director. Watch whether Congress pushes back on these cuts as it did last year or if federal cybersecurity capacity continues degrading while threat activity increases.

Americans Lost $21B to Cybercrime in 2025: US cybercrime losses reached $21 billion in 2025, up 26% year-over-year, driven by investment scams accounting for $8.6 billion and cryptocurrency fraud exceeding $11 billion. The FBI received over 1 million complaints, with Americans over 60 losing $7.7 billion, up 37% from the prior year. AI-enabled scams using voice cloning and deepfakes accounted for $893 million in losses. Watch whether authentication technologies and user education can slow this growth or if losses continue accelerating as AI lowers the skill barrier for sophisticated fraud operations.

Google Launches Offline AI Dictation to Challenge Wispr: Google quietly released an offline-first dictation app using Gemma-based speech recognition models that automatically removes filler words and polishes text without cloud connectivity. The app competes with Wispr Flow and similar tools but leverages Google's edge AI capabilities for local processing. This signals a broader shift toward on-device AI models that prioritize privacy and reliability over cloud-dependent processing. Watch whether offline-first AI apps gain traction with enterprise users concerned about data exposure or if cloud-based models maintain dominance through superior accuracy.

Scanning the Wire

Chrome adds vertical tabs and Reading Mode to tame browser chaos: Google's latest update brings vertical tab management and a cleaner reading interface to help users navigate cluttered browsing sessions. (TechCrunch)

Uber shifts more infrastructure to Amazon's AI chips: The ride-sharing company is expanding its AWS contract to run core features on Amazon-designed silicon, moving workloads away from Oracle and Google infrastructure. (TechCrunch)

Modus raises $85M to deploy AI agents in accounting workflows: The startup, which acquires stakes in established accounting firms and layers in AI automation for audit work, closed a $5 million seed and $80 million Series A led by Lightspeed. (Axios)

TikTok commits €1B for second Finnish data center: The investment is part of a €12B European data sovereignty initiative aimed at isolating data for over 200 million European users from Chinese government access concerns. (Reuters)

Stack Overflow abandons redesign after user backlash: The developer Q&A site scrapped plans to shift toward a discussion-focused format, struggling to define its role as AI tools increasingly handle the technical queries that built its reputation. (The Register)

Amazon S3 Files lets AI agents treat object storage as a native file system: The new service mounts S3 buckets directly into agent environments with full file semantics, eliminating the sync pipelines and data duplication that previously broke multi-agent workflows. (VentureBeat)

Cities and states push back against data center expansion: Local governments are moving to restrict new facilities over concerns about electricity demand and environmental impact as AI infrastructure requirements accelerate. (WSJ)

OpenAI resets Codex usage limits to celebrate 3M weekly users: Sam Altman announced the company will raise rate limits every time it adds another million users, continuing resets until the developer tool reaches 10 million weekly active users. (Sam Altman)

Outlier

When States Sandbox AI Prescriptions, Security Firms Jailbreak Them: Utah became the first state to allow AI to prescribe medications through a regulatory sandbox program designed to encourage healthcare innovation. Security researchers at Mindgard immediately demonstrated they could manipulate the Doctronic chatbot into recommending dangerous drug combinations and inappropriate prescriptions. This reveals the core tension in AI regulation: sandboxes designed to accelerate deployment create exactly the permissive environments where adversarial testing should happen first, not after launch. The gap between "move fast" policy frameworks and "break things" security reality suggests we are building regulatory infrastructure that optimizes for speed over safety in domains where the stakes are highest. Watch whether other states pause medical AI approvals or double down on the sandbox model as the default path for bringing algorithmic decision-making into regulated industries.

The tech industry spent a decade teaching software to eat the world. Now it's learning that indigestion looks a lot like infrastructure collapse, budget cuts, and 80,000 people wondering if they trained their own replacement.