AI News Digest — 26 April 2026

AnalysisApoorva Mehta, who co-founded Instacart, launched a hedge fund called Abundance with $100 million in seed equity that aims to let AI agents run the entire fund. Thousands of bots scrape the internet for trade ideas, conduct research, choose long and short stock bets, size positions, and execute trades. A 10-person team of quants and engineers maintains the models. Mehta said he started the firm after OpenAI's o3 model showed reasoning advanced enough to make capital allocation decisions. Citadel founder Ken Griffin said in late 2025 that generative AI is not yet beating the market for hedge funds. Mehta claims Abundance has outperformed indexes but will not say which. The fund trades its own capital for now. The interesting part is that a founder with a successful exit just bet $100 million that fundamental traders are the next role to go.

AnalysisA US judge in Oakland on Friday dismissed Elon Musk's fraud and constructive fraud claims against OpenAI, Sam Altman, and Greg Brockman, at Musk's own request. That leaves two claims of the original 26: breach of charitable trust and unjust enrichment. Jury selection begins Monday. Musk's argument is that OpenAI, founded as a nonprofit to develop AI for humanity, betrayed that mission when it built a for-profit arm in 2019 and took billions from Microsoft. He is seeking around $150 billion in damages, with proceeds going to OpenAI's charitable arm rather than to him personally. The trial lands as OpenAI explores an IPO that could value it at $1 trillion and as Musk's xAI competes directly. Stripping the fraud claims sharpens the case. The jury now decides whether the most-watched AI company broke the deal it was founded on.

AnalysisVast Data, a New York firm that builds storage software for AI training clusters, raised $1 billion at a $30 billion valuation on Wednesday, April 22, more than tripling its 2023 mark. Drive Capital and Access Industries led, with Nvidia, Fidelity, and NEA joining. Vast says its platform now feeds clusters running on millions of GPUs and lists CoreWeave, Mistral AI, Cursor, and the US Air Force as customers. Cumulative bookings have crossed $4 billion. The valuation is a bet on a quieter problem behind the model wars: GPUs sit idle when storage cannot stream data fast enough. McKinsey estimates 25% of the projected $5.2 trillion in AI infrastructure spend through 2030 goes to power, cooling, and the supporting data layer. Nvidia gets the magazine covers. The companies feeding the chips so they do not stall are catching up on revenue.

AnalysisTrendForce cut its 2026 server shipment growth forecast from 20% to 13% on Wednesday, blaming long lead times for the power management chips and motherboard management chips that AI servers consume in volume. uPI Semi, which designs power chips, expects shortages to last through 2026 because foundries are prioritizing higher-margin AI orders. On Sunday, Digitimes reported optical module suppliers in Taiwan are riding a stock surge as data centers hit interconnect limits, the wires moving data between GPUs. None of this is theoretical. Goldman Sachs projects data center power demand will rise 165% by 2030, requiring $720 billion in grid upgrades. The five largest US cloud companies have committed roughly $660 billion in 2026 capex (long-term spending on chips, buildings, and fiber), much of which sits idle until utilities can deliver power. The 2026 ceiling on AI is electricity, copper, and glass.