The Espionage Economics

The pattern is unmistakable: nation-states are embedding themselves deeper into critical infrastructure, while the commercial incentives for exploitation keep declining. This isn’t just a cybersecurity problem anymore. It’s becoming a structural feature of how state power operates in the digital era, and it’s reshaping where American tech talent and capital flow in response.

Deep Dive

Chinese Spies Built Persistent Backdoors That Went Undetected for Years

The scale of what Brickstorm represents should alarm anyone tracking the long-term trajectory of state-sponsored cyber operations. Chinese espionage actors didn’t just breach networks and exfiltrate data. They embedded themselves as permanent fixtures, maintaining access for years across Linux, VMware, and Windows environments, specifically targeting the infrastructure that other organizations depend on.

What makes this particularly consequential is the dwell time and scope. In one incident CISA responded to, attackers gained access in April 2024 and maintained persistence until at least September. During that five-month window, they collected cryptographic keys, compromised domain controllers, and positioned themselves to reach downstream targets. Google’s Mandiant estimates dozens of US organizations remain compromised, with SaaS providers and edge device manufacturers being deliberately targeted as pivot points to reach their customers. CrowdStrike attributes the campaign to Warp Panda, a suspected Chinese group active since 2022 that’s also been breaking into Azure environments to access Microsoft 365 data.

The second-order effect here matters more than the headline: these aren’t quick smash-and-grab operations. They’re strategic placement of persistent infrastructure designed to enable long-term disruption and sabotage. CISA’s Nick Andersen explicitly called this out: “State-sponsored actors are not just infiltrating networks, they’re embedding themselves to enable long term access, disruption, and potential sabotage.” The Brickstorm malware is sophisticated enough to operate across hypervisor and guest VM layers, suggesting whoever built it understood enterprise infrastructure at a level that requires significant investment. This isn’t defensive espionage. It’s preparation for kinetic or economic disruption if geopolitical tensions escalate.

AI Agents Are Becoming Cheaper Weapons for Finding Exploitable Flaws

Anthropic’s SCONE-bench research landed like a bomb this week because it forced a conversation nobody in the AI safety space wanted to have: autonomous AI agents are now profitable attack vectors against real systems with real money at stake. The economics are suddenly favorable to attackers in ways they weren’t a year ago.

Claude Opus 4.5 and GPT-5 identified $4.6 million in exploitable smart contract vulnerabilities across Anthropic’s benchmark of 405 contracts. When the researchers ran GPT-5 against 2,849 recently deployed contracts with no publicly disclosed vulnerabilities, it found two zero-day flaws and generated profitable exploits. The per-run cost came to $1.22; the cost per vulnerable contract identified was $1,738; the average profit per exploit was $1,847. Net profit: $109 per vulnerability discovered.

That’s the critical number. Less than a year ago, researchers at UCL and University of Sydney estimated the cost of identifying a vulnerable smart contract at roughly $3,000. Anthropic's findings show that cost has dropped to $1,738, almost a 42 percent decline. This follows a pattern where “exploit revenue from stolen simulated funds roughly doubled every 1.3 months” over the past year. The same trajectory that made large language models go from research curiosities to production systems is now making autonomous exploitation tools economically viable. Anthropic’s counterargument that “AI can defend against the risks created by AI” rings hollow when the cost of attack keeps falling faster than the cost of defense.

Masayoshi Son is Building America’s New Industrial Base, With Trump

The real story buried in reports about Masayoshi Son’s “Trump Industrial Parks” isn’t the branding. It’s that Japan is quietly becoming the capital source for America’s AI infrastructure buildout, and the White House has essentially blessed it as part of geopolitical strategy. This is manufacturing policy by another name.

Son is negotiating with the White House to build large-scale AI infrastructure facilities on federal land, funded through Japan’s recently concluded trade deal with the Trump administration. The facilities would produce components for AI infrastructure, not finished consumer products. This matters because it signals a fundamental shift: the US can’t satisfy its own AI infrastructure demand with domestic capital and manufacturing capacity, so it’s outsourcing the buildout to a trusted ally rather than China. It’s a strategic concession dressed up in dealmaking language.

What Son understands that most Western tech executives miss is that infrastructure is power. He’s not building factories to sell chips back to consumers; he’s building factories that will lock American AI companies into dependency on specific hardware architectures and supply chain relationships. SoftBank gains leverage over how the American AI ecosystem develops. Japan gains a permanent seat at the table in how global AI infrastructure is governed. The US gets manufacturing jobs and reduced China exposure, but trades autonomy for security. This is the geopolitical calculus of the next decade: not who builds the most AI companies, but who controls the physical substrate those companies run on.

Signal Shots

Anthropic Signs $200M Snowflake Deal to Deploy AI Agents at Scale — Anthropic and Snowflake announced a $200 million partnership to integrate Claude into Snowflake’s governed data environments, enabling multi-step autonomous agents to work directly with customer data. This transforms Snowflake from a data warehouse into an AI execution layer, and it signals how the real value in enterprise AI shifts from model inference to data access and orchestration. Expect similar deals from every major cloud platform.

Moore Threads Soars 400% on Shanghai Debut After $1.1B IPO — China’s homegrown AI chipmaker raised $1.13 billion with a valuation around $7.6B, with retail orders oversubscribed 2,750x. This isn’t just a win for Chinese tech nationalism; it’s proof that the market believes China can build competitive AI hardware without waiting for NVIDIA. That belief, whether justified or not, changes capital allocation globally and accelerates China’s vertical integration play.

Apple’s AI Exodus Accelerates to OpenAI — Dozens of Apple staffers with expertise in audio, watch design, robotics, and hardware engineering have joined OpenAI in recent months, part of a broader wave of departures from Apple’s hardware division. This tells a story: hardware engineering talent sees more leverage building AI agents than shipping incremental product iterations. OpenAI is becoming a hardware company by stealth, pulling talent from the only company that’s genuinely world-class at shipping devices at scale.

Meta Faces EU Antitrust Probe Over WhatsApp AI Restrictions — The European Commission opened an investigation after WhatsApp blocked OpenAI’s ChatGPT and Microsoft’s Copilot from its platform, forcing millions of users to go elsewhere. This is Meta using platform control to slow rival AI adoption, which is exactly the playbook Europe wrote the DMA to prevent. Expect similar restrictions to face scrutiny across all gatekeeper platforms.

Server and PC Prices Jumping 15% as Memory Costs Spike — Hardware manufacturers are planning double-digit price increases as DRAM and NAND shortages bite, with major OEMs confirming the moves to The Register. The AI infrastructure boom is consuming memory like nothing before it, and the supply chain hasn’t caught up. This eventually forces a reckoning: either memory fabs expand capacity, or AI model deployments become even more expensive and capital-constrained.

Amazon’s Graviton5 CPU Hits 192 Cores, Becoming De Facto AWS Standard — Amazon unveiled its densest CPU yet, cramming 192 processor cores into a single socket, with Graviton now accounting for roughly half of all new CPUs added to AWS over the past three years. This is vertical integration at scale: AWS is making x86 increasingly optional, which means less revenue for Intel and AMD and more margin for Amazon. The real winner is customers who can use Gravitons without sacrifice; the real loser is anyone betting that Intel’s dominance is permanent.

Scanning the Wire

EU Opens Bidding for AI Gigafactories in Early 2026 — The European Commission wants to build large-scale AI facilities through public-private partnerships to compete with US infrastructure spending. WSJ
Waymo Under Federal Scrutiny for Illegally Passing School Buses — The NHTSA is investigating reports that Waymo self-driving vehicles passed school buses 19 times in Texas alone this year, including multiple violations caught on record. Reuters/Techmeme
Microsoft Quietly Patches Critical Windows .LNK Flaw After Years of Abuse — A critical flaw that allowed hidden malicious commands in Windows shortcut files was silently patched, ending a long-running espionage vector that state actors had weaponized extensively. The Register
ChatGPT Used by Alleged Violent Stalker Who Called Himself ‘God’s Assassin’ — A podcaster faces up to 70 years after allegedly using ChatGPT to escalate stalking behavior, with the DOJ claiming the AI model “hyped up” his violent delusions about being a divine instrument. Ars Technica
Micro1 Hits $100M Annualized Revenue While Fielding $2.5B Valuation Offers — The AI training platform that connects labs with data annotation experts crossed a major revenue milestone in just eight months, proving that AI infrastructure services, not just models, are where the real money flows. Forbes/Techmeme
Chicago Tribune Sues Perplexity Over Copyright Infringement — The newspaper alleges that Perplexity’s retrieval-augmented generation is systematically scraping copyrighted content without permission or compensation, marking the first major outlet to take legal action against the search startup. TechCrunch
Proxmox Delivers First Stable Release of Datacenter Manager — The open-source virtualization project released a production-ready multi-site VM management platform, creating a genuine VMware alternative for enterprises looking to break vendor lock-in. The Register
Apple’s Executive Exodus Accelerates With General Counsel and Policy Chief Departures — Following the exits of Jeff Williams, John Giannandrea, and Alan Dye, Apple’s general counsel Kate Adams and environmental VP Lisa Jackson are stepping down, signaling continued talent drain at the top. The Verge
Trump Admin May Deny H-1B Visas to Content Moderation Workers — The administration has ordered US consulates to consider rejecting H-1B applications from visa holders involved in content moderation or “censorship,” effectively blocking a key talent pipeline for tech platforms. The Verge
AI Chatbots Surprisingly Effective at Changing Voters’ Minds — New research shows AI chatbots can shift political views more effectively than traditional campaign advertising, raising questions about the 2024 and 2026 election cycles. Washington Post

Outlier

Twins Convicted of State Department Database Sabotage Hired Back by Government, Now Charged Again — Two contractors with prior convictions for similar crimes were cleared to work on sensitive government systems again, then allegedly deleted databases and used AI tools to cover their tracks before being arrested a second time. This isn’t a security story; it’s a cautionary tale about institutional amnesia. It suggests the government’s vetting process is either fundamentally broken or so compartmentalized that information doesn’t flow where it needs to. Either way, it hints at a much larger problem: how many people with expertise in critical infrastructure sabotage are cycling through government contracts precisely because nobody talks to each other?

We’ll see you tomorrow with more signals from the wire. The infrastructure wars are just getting started.

BLACK PINE INSIGHTS