CxO · Weekly · March 23, 2026

the hard choke points: chips, euv, and war

Supermicro’s co‑founder was arrested for allegedly smuggling $2.5B of Nvidia GPUs to China.[3] The news triggered a 28% drop in Supermicro’s stock.[4]

TSMC still manufactures 90% of the world’s most advanced logic chips.[5] Taiwan imports 97% of its energy, tying leading‑edge AI compute to a highly exposed grid.[6]

Nvidia’s Blackwell B200 GPU doubles tensor compute over H100, while its Vera Rubin platform is advertised at 10× inference efficiency and 4× training performance versus Blackwell.[7]

Nvidia is projecting $1T of AI‑chip demand by 2027 and says it already has a $1T order book.[8] On the lithography side, Belgium’s imec bought ASML’s EXE:5200 High‑NA EUV tool, targeting qualification by Q4 2026 for sub‑2 nm chips.[9]

The Iran war has taken out 17% of Qatar’s LNG capacity for up to five years and sparked sharp oil price jumps, feeding concerns about helium shortages for chipmaking and warnings that sustained high oil could undermine the AI boom.[10]

frontier labs, governments, and lawsuits

Anthropic is fighting on multiple legal fronts, including a BMG Rights Management suit over alleged use of copyrighted song lyrics in model training and separate action tied to the OpenCode project.[11]

Nearly 150 retired federal and state judges filed an amicus brief supporting Anthropic in its lawsuit against the Pentagon, and Microsoft plus retired military chiefs have publicly backed the company.[12]

Despite that support, the Pentagon is reportedly developing alternatives to Anthropic’s technology over security concerns.[13] Anthropic’s CEO is on record saying there is up to a one‑in‑four chance that AI could cause an existential catastrophe within three years.[14]

At the same time, the Pentagon plans to adopt Palantir AI as a core system across all U.S. military branches after using it to detect and prioritize more than 1,000 targets in the first 24 hours of a military operation.[15]

The U.S. military is preparing to let AI companies train on classified data, and DHS is contracting AI vendors to surveil Americans, while Google leans further into national security contracts.[16]

OpenAI is simultaneously scaling up as a quasi‑infrastructure provider, planning to double its workforce to 8,000 by 2026, integrate ChatGPT, Codex, and Atlas into a single ‘Super App,’ and build a fully automated AI researcher by 2026.[17]

That expansion comes alongside commercial and legal steps like adding ads to ChatGPT in the U.S., being sued by Encyclopedia Britannica over alleged content ‘memorization,’ and facing backlash over a proposed adult mode.[18]

the compliance bubble pops

Delve, a YC‑backed startup that raised $32M, is accused of faking SOC2, ISO 27001, HIPAA, and GDPR compliance reports for 494 companies.[19]

Investigators say Delve pre‑generated auditor conclusions before reviewing evidence and reused the same template for 99.8% of its reports.[20] Commentary across the industry frames this as evidence of ‘compliance theater’ and ‘certification mills,’ where badges matter more than underlying security.[21]

The EU AI Act will require high‑risk AI systems to meet formal compliance obligations by August 2026, adding hard timelines on top of already strained processes.[22]

A U.S. federal judge has ruled that Workday’s AI‑based hiring tools can be sued for allegedly filtering out applicants over 40, putting algorithmic employment decisions squarely in legal crosshairs.[23]

Separately, the FBI has confirmed it buys Americans’ location data from brokers, and Essex Police halted use of facial recognition after finding racial bias, even as AI misidentification kept an innocent grandmother in jail for six months.[24]

These episodes sit against a broader consensus that AI systems often reflect and reinforce biases in their training data, particularly around race and performance evaluation.[25]

china’s full‑stack ev + ai play

BYD showrooms in Asia are reportedly packed with new EV buyers after the Iran‑driven oil shock, with the company marketing chargers that can refuel cars almost as fast as gas pumps.[26]

BYD claims five‑minute EV charging and has opened a supercharging station with its own solar panels and battery energy storage system, while also being integrated into Nvidia’s self‑driving stack.[27]

China’s EV charging infrastructure is already using higher‑voltage systems than the U.S., giving it faster charging and denser networks in many regions.[28]

Xiaomi has launched the SU7 EV with a claimed 902 km range and Lidar, and is building an in‑house end‑to‑end driving model on top.[29]

Xiaomi’s MiMo‑V2‑Pro AI model scores 49 on the Artificial Analysis Intelligence Index, offers a 1M‑token context window, is said to perform similarly to GPT‑5.2 and Opus 4.6 at lower cost, and is only available via Xiaomi’s own API.[30]

Alibaba, meanwhile, is cutting headcount by 34% by 2025 after a 66% net income drop, even as it commits to open‑sourcing new Qwen and Wan models, launches the Wukong agent platform, and admits its 470,000 AI chips may be inferior to competitors.[31]

Alibaba’s Qwen team has also lost its technical lead and two senior researchers, raising questions about execution at the same time that Chinese public opinion remains broadly optimistic about AI’s societal impact.[32]

the backlash: ai slop and automation regret

Nvidia’s DLSS 5, pitched as an AI breakthrough for game visual fidelity, is facing heavy backlash from gamers who call its generative imagery ‘AI slop’ and prefer traditional graphics techniques.[33]

Critics argue DLSS 5 relies too much on 2D frame data and produces visual hallucinations, while Nvidia CEO Jensen Huang says gamers are ‘completely wrong’ and defends the tech as misunderstood.[34]

YouTube is now explicitly asking users whether videos ‘feel like AI slop,’ institutionalizing a term that has spread across platforms to describe low‑quality AI output.[35]

Beyond content quality, 55% of companies that replaced employees with AI agents say they regret doing so, and current agents are widely described as 5% technology and 95% hype.[36]

At the same time, workers and the public are absorbing dire forecasts, from the Anthropic CEO predicting that 50% of entry‑level white‑collar jobs could vanish within three years to the ServiceNow CEO warning of 30% Gen Z unemployment due to AI.[37]

Labor resistance is already visible, with 2,400 Kaiser mental‑health professionals striking over AI issues and 1.5M users reportedly quitting GPT over ethical concerns.[38]

Polls show many Americans explicitly view AI as a ‘wealth inequality machine,’ linking these technology choices to broader distributional anxiety.[39]