CxO · Weekly · June 5, 2026

frontier labs, index rules, and the reflexive AI bubble

Anthropic has confidentially filed an S‑1 with the SEC.[1] The company is reportedly valued at $965B after a giant primary round, overtaking OpenAI as the most valuable AI startup.[2]

Commentary puts Anthropic’s annual revenue around $47B, putting a five‑year‑old lab into hyperscaler territory on top line.[3] NASDAQ now allows newly listed firms to join the NASDAQ 100 after just 15 trading days, raising concerns that passive funds will be forced into names like Anthropic and SpaceX at IPO-inflated prices.[4]

Morningstar values SpaceX at $780B, while the company is reportedly aiming for an IPO valuation of at least $1.8T even as critics describe its AI arm xAI as heavily indebted and unprofitable.[5]

ai infra capex, chips, and the data center backlash

Alphabet plans to raise about $80B in equity to expand AI infrastructure and compute, with Berkshire Hathaway alone reportedly committing $10B.[6] SoftBank briefly became Japan’s largest company by market cap on AI enthusiasm, and now plans to invest up to €75B in French data centers despite mounting concern over its liquidity and debt load.[7]

At the component level, 32GB of DDR5 RAM now costs about $375 amid AI-linked shortages, and Nvidia’s H100 GPU faces its own production bottlenecks as TSMC warns chip supply will trail AI demand for years.[8]

South Korea’s chip exports are surging on AI demand, while more money is now being spent building data centers than the U.S. government spends on transportation.[9]

At the same time, Ohio has paused data center tax breaks, Seattle is moving to ban new facilities, and the first U.S. city has already enacted an outright data center ban as public sentiment turns sharply anti‑DC.[10]

ai unit economics and runaway bills

Uber exhausted its 2026 AI budget in only four months.[11] It has now capped AI coding tools at $1,500 per engineer per month while reporting that 95% of its engineers already use AI.[12]

A separate case saw a mystery company rack up a $500M Claude bill in one month after failing to enforce usage limits on its licenses.[13] Microsoft’s own data suggests that for some workloads, AI services are now more expensive than hiring human workers, a point echoed by developer anger over GitHub Copilot’s new usage-based pricing.[14]

Bain’s survey work indicates that around 95% of corporate AI projects to date have produced no measurable return, even as companies complain about steep recent price hikes for AI tasks and flagship models.[15]

regulation, ownership fights, and liability

Florida’s Attorney General has sued OpenAI and Sam Altman, alleging negligence about AI risks to human life and deceptive marketing around safety, making Altman personally a test case for AI liability.[16]

Bernie Sanders has introduced a bill that would give the public a 50% ownership stake in major American AI companies via an AI sovereign wealth fund, explicitly arguing that AI should be owned by the people rather than billionaires.[17]

Commenters liken the proposal to nationalization and question whether taxpayers should underwrite currently unprofitable firms with sky‑high valuations.[18]

Regulators are also tightening at the edges: California’s Senate passed what it calls the first ban on AI chatbot toys over safety concerns, while its Assembly voted 68–1 to exempt open‑source Linux from a broad age‑verification law, signaling how messy AI-adjacent rules will be.[19]

In Europe, Google faces a planned high triple‑digit million euro fine and UK publishers can now block their content from appearing in its AI search summaries, directly attacking the data flows feeding generative products.[20]

ai security: agents as attackers and a fraying supply chain

Security researchers have documented the first confirmed LLM-agent cyberattack, where an AI agent autonomously exploited a vulnerability, hacked a server, and exfiltrated an AWS credential store in under an hour.[21]

New work on attacks like MemPoison shows adversaries can poison memory mechanisms in LLM agents to bypass selective defenses, expanding the offensive toolkit against AI-driven systems.[22]

Hackers have hijacked high-profile Instagram accounts by tricking Meta’s AI support chatbot into granting access, turning automated support into an attack surface.[23]

On the supply-chain side, Red Hat and IBM disclosed a campaign of malicious npm packages in their cloud services dubbed “Miasma: The Spreading Blight,” while Oracle is contending with active exploits of a two‑year‑old flaw.[24]

Google has also patched an actively exploited Android zero‑day among more than 100 other fixes, underscoring how AI, mobile, and cloud vulns are converging.[25]