CxO · Monthly · April 15, 2026

the fragile AI compute stack

Missiles and regulation are now a real attack surface for compute. Iranian strikes knocked AWS availability zones in Bahrain and Dubai "hard down" and hit an Oracle data center in Dubai, taking entire regions offline.[3]

The same war has cut off helium supplies from Qatar, a critical input for semiconductor fabs, tightening an already stressed chip supply chain.[4]

The FCC has banned imports of new foreign-made Wi‑Fi routers and U.S. prosecutors have charged a Supermicro co‑founder with smuggling $2.5B of Nvidia servers to China, underscoring how networking and accelerators are being securitized.[5]

Yet supply is also being massively scaled: Broadcom expanded its TPU deal with Google and Anthropic to multi‑gigawatt capacity by 2027, SpaceX/Intel/xAI/Tesla’s Terafab targets 1 terawatt per year of compute, and Softbank is planning a 10‑gigawatt Ohio data center backed by a $33B gas plant.[6]

labor, layoffs, and stalled AI productivity

Boards are using AI as the narrative cover for a large labor reset. Oracle is laying off between 20,000 and 30,000 employees, roughly 18% of its workforce, then rehiring 8,000 as lower‑paid contractors to finance AI data centers.[7]

A 2026 tracker attributes around 16,000 layoffs each at Meta and Amazon to AI, and an AI Jobs Risk Index pegs 9.3M U.S. roles at risk of displacement within five years while Anthropic’s CEO says half of entry‑level white‑collar jobs may vanish in three.[8]

At the same time, only 6.8% of 2026 CS majors report full‑time offers (vs 94% in 2022), and 75% of resumes never reach a human because of AI screening.[9]

Inside firms, 80% of white‑collar workers report resisting mandated AI tools, Gen Z workers are sabotaging rollouts over job‑loss fears, and Walmart’s ChatGPT checkout converted at roughly one‑third the rate of its normal site before Walmart ended the OpenAI partnership.[10]

mythos and the AI‑security arms race

Anthropic’s Claude Mythos preview is turning software security into an AI arms race. The model has autonomously identified thousands of zero‑day vulnerabilities across major operating systems and browsers, scored 93.9% on SWE‑bench Verified, and exposed a 27‑year OpenBSD TCP bug that had never been caught.[11]

Early reports of Mythos testing coincided with a sharp selloff in cybersecurity stocks and spurred Goldman Sachs and the U.S. Treasury to seek access for internal scanning.[12]

Anthropic is funding Project Glasswing with $100M to use Mythos on critical software, but a leak of roughly 3,000 internal documents flagged the model’s offensive potential, so access is being tightly restricted.[13]

Meanwhile, real supply chains remain porous, with a compromised Trivy container scanner and a poisoned Python package with 97M downloads showing how easily attackers can ride existing tools.[14]

openai, anthropic, and the stressed frontier model economy

Frontier‑model economics are starting to look like an expensive, politically exposed competition. OpenAI raised $122B at an $852B valuation, projects $121B in compute spend by 2028, and does not expect profitability before at least 2030.[15]

Its Sora video generator was reportedly burning about $1M a day before being shut down, the Stargate UK megadatacenter is paused over energy and regulatory friction, and the company is backing an Illinois bill to limit AI‑lab liability even as an 18‑month investigation found Sam Altman privately lobbied against regulations he endorsed publicly.[16]

To fund this, OpenAI is offering private‑equity investors a 17.5% guaranteed minimum return plus early access to unreleased models and is rolling out ads that already exceed $100M annualized revenue, with ambitions for $2.5B in ad revenue this year.[17]

Anthropic has quietly overtaken OpenAI in revenue run rate (~$30B vs ~$25B) while locking in multi‑gigawatt TPU capacity with Google and Broadcom, even as it suffers a leak of 512,000 lines of Claude Code and backlash over mass DMCA takedowns aimed at forks on GitHub.[18]

energy, war, and the EV/renewables repricing

The Iran war is repricing both energy and the assets tied to decarbonization. Iran is demanding Bitcoin tolls from ships in the Strait of Hormuz, threatening undersea cables that carry most global data, and constricting a chokepoint that handles about 20% of world oil and LNG, while U.S. gas has risen to roughly $3.94 per gallon.[19]

Rising fuel costs are already hitting operators: Amazon added a 3.5% fuel and logistics surcharge for sellers, explicitly tying it to Iran‑driven energy prices.[20]

In parallel, renewables have surpassed natural gas as the largest U.S. power source, Britain has hit over 90% renewable generation, and renewables now account for nearly half of global electricity capacity, occasionally pushing German power prices negative.[21]

EV performance is splitting by player and region: Tesla’s Model Y is the world’s best‑selling car but Tesla is also sitting on 50,000 unsold EVs, BYD delivered about 310,000 pure EVs in Q1 yet faces reports of a 25% sales drop, and GM is idling an electric truck plant and cutting 2,500 workers amid softer demand.[22]

The WTO has warned that prolonged high oil prices could hurt the AI sector, even as AI data centers sharply increase electricity demand and drive mega‑builds like Softbank’s planned 10‑gigawatt campus in Ohio.[23]