Sponsored by
Our Founder Was Early to AI Winners. Now He’s Tracking What Comes Next.
Over the last 18 months, MavSource’s founder tracked and invested around major AI-related names like Micron +100%, Nvidia +74%, Sandisk +130%, Western Digital +74%, TSM +22%, Broadcom +27%, Okta +35% and Lam Research +39% — an average return of approximately +63% across the group.
Now, he’s bringing the sources and ideas behind those insights to you in one daily email digest. MavSource aggregates the most important AI updates from newsletters, podcasts, company news, AI labs, funding rounds, and more — then summarizes what matters in a simple 5-minute brief.
Past performance does not indicate future results. Informational only; not investment advice.
AI Box Daily
Thursday, July 2, 2026 · By Jaeden Schafer
In today's issue:
OpenAI offers US a 5% stake as Sanders pushes for a 50% tax
Cursor's $60B SpaceX deal puts its OpenAI and Anthropic ties in doubt
Anthropic in talks with Samsung on a custom AI chip
Meta puts smart glasses features behind a Meta One Premium subscription
Microsoft commits $2.5B to Frontier, a new AI deployment company
OpenAI offers US a 5% stake as Sanders pushes for a 50% tax
OpenAI is proposing to give the US government a 5% equity stake, modeled on Alaska's oil-wealth dividend fund, as part of early talks with the Trump administration. The offer sits squarely between public pressure for AI regulation and Bernie Sanders' competing plan for a 50% one-time tax on leading AI firms.
THE DETAILS:
Sanders' proposal would raise an estimated $7 trillion through a one-time 50% stock tax, with proceeds for direct payments and public programs like healthcare and housing.
The Trump administration has separately approached Google and Meta about similar equity stakes, though neither company has publicly signaled agreement.
70% of Americans oppose AI data centers in their area, and 50% report more concern than excitement about AI, per recent polling cited in the article.
WHY IT MATTERS: This is a play for political cover dressed as equity. OpenAI is betting a small, passive stake—framed around shared prosperity—will defuse Sanders' push for structural control and heavy taxation. The real negotiation isn't about the 5% number; it's whether the next administration lets AI companies retain autonomy in exchange for public ownership optics or demands genuine seats at the table. Watch whether Google and Meta fold or hold.
Cursor's $60B SpaceX deal puts its OpenAI and Anthropic ties in doubt
SpaceX is acquiring Cursor, the $60 billion AI coding platform, forcing OpenAI and Anthropic—Cursor's two largest model suppliers—to decide whether to keep selling through a competitor-owned product. The deal closes later this year and puts Cursor's multi-model strategy in tension with its new owner's rival AI ambitions.
THE DETAILS:
Cursor is already training its next model on SpaceX infrastructure with 10–20x more compute than previously available, signaling Musk's plan to build native capabilities.
Anthropic cut off Windsurf's Claude access when OpenAI acquired it last year; the precedent suggests a similar move is possible, though Anthropic's new multi-billion-dollar compute deal with SpaceX complicates that calculus.
OpenAI's startup fund, an early Cursor investor, will convert its stake into SpaceX stock through the acquisition, giving OpenAI a direct financial stake in Cursor's success despite Codex competing with it inside the platform.
WHY IT MATTERS: This deal crystallizes a shifting AI power dynamic: Musk's ability to offer massive compute resources creates new leverage with labs like Anthropic that might otherwise cut ties. OpenAI's financial upside in Cursor also buys goodwill. Neither lab wants to cede distribution; expect Cursor to stay multi-model, but as a less neutral platform—one where SpaceX's own models gradually gain shelf space and preferential treatment as Cursor's compute advantage grows.
Anthropic in talks with Samsung on a custom AI chip
Anthropic is in talks with Samsung to build a custom AI chip, joining OpenAI, Google, and Amazon in the race to reduce reliance on Nvidia. The move addresses both cost and supply constraints—custom silicon lets labs optimize for their own architectures while negotiating leverage against a single dominant supplier.
THE DETAILS:
OpenAI unveiled its own custom inference chip called Jalapeño last week, built with Broadcom and claimed to offer better performance-per-watt than competitors.
Samsung already manufactures Nvidia chips for AI training and is building a joint AI chip factory in South Korea with Nvidia, positioning it at the center of the supply chain.
Anthropic's project remains early-stage; the chip's purpose, server role, and performance targets haven't been finalized, suggesting a multi-year development timeline ahead.
WHY IT MATTERS: Every frontier lab is now building or planning custom silicon—this is no longer a Google/Amazon edge case, it's table stakes. Samsung's dual role as Nvidia partner and potential Anthropic vendor is shrewd positioning; it can serve both sides of the supply chain without direct conflict. Expect the next two years to see multiple custom chips ship, finally breaking Nvidia's monopoly grip on accelerator margins.
Meta puts smart glasses features behind a Meta One Premium subscription
Meta is putting advanced smart glasses features behind a paid subscription, starting with Conversation Focus—which enhances voice clarity in noisy settings—capped at 3 hours per month free, 15 hours for $10/month Meta One Premium subscribers. This is a deliberate monetization play, not a cost recovery one: the feature runs entirely on-device with no server cost to Meta, signaling how AI hardware makers will extract value once installed bases scale.
THE DETAILS:
Meta's new $299 Meta-branded glasses are sold at break-even to grow adoption, following the classic razor-and-blades model where recurring software subscriptions drive long-term revenue.
Google launches competing smart glasses this year with Samsung, Warby Parker, and Gentle Monster; Google's own inference efficiency gains may let it bundle more features than Meta without aggressive tiering.
Apple and Google already gate advanced AI features behind subscriptions: Pixel Video Boost requires a specific Google One tier, Gemini Live sits behind Gemini Premium, and iOS 27 photo-editing locks into higher iCloud+ plans.
WHY IT MATTERS: Meta is making the clearest bet yet that on-device AI hardware is a platform play, not a one-time sale. By gatekeeping features that cost them nothing to serve, they're signaling that the real money lives in building habit and then layering subscriptions once users can't imagine life without the glasses. Google's incoming lineup could flip this if their efficiency advantage lets them undercut Meta's tiering—but either way, smart glasses just became a subscription business.
Microsoft commits $2.5B to Frontier, a new AI deployment company
Microsoft is launching Frontier, a $2.5 billion dedicated unit that will embed 6,000 engineers directly into enterprise customers to deploy AI systems end-to-end. This escalates a deployment arms race with AWS, which announced a rival $1 billion venture just two days earlier—both hyperscalers betting that enterprises need hands-on implementation help, not just software licenses.
THE DETAILS:
At 6,000 engineers, Frontier is roughly six times larger than Palantir's entire FDE organization at IPO and double AWS's announced commitment.
Early partners include London Stock Exchange Group, Unilever, Land O'Lakes, and Accenture—a mix of data-heavy, supply-chain, and systems-integrator customers.
Microsoft's existing Azure and Microsoft 365 relationships give it embedded engineers already inside the Fortune 500, a structural advantage AWS will have to build from scratch.
WHY IT MATTERS: Microsoft just signaled that AI deployment—not model capability—is the real bottleneck. By doubling AWS's bet and leveraging an existing installed base, Microsoft is converting its enterprise stickiness into a moat around the last mile. AWS will have to outspend or outmaneuver; both OpenAI and Anthropic are watching to see if the hyperscalers squeeze out pure-play AI labs from the implementation layer.
Everything else in AI today
Google's electricity use jumped 37% in 2025 as AI buildout accelerates — Data centers consumed 42 million MWh, more than New Zealand or Denmark; operational emissions still fell 2%.
Nvidia opens AI-factory model with Sharon AI and Firmus taking 210,000 GPUs — Nvidia is fronting infrastructure to AI clouds in exchange for revenue sharing, with Sharon AI booking 40,000 GB300s and Firmus scaling to 360 MW.
Bhavin Turakhia stakes $30M of his own money on Neo, an AI-native Office rival — The Directi and Zeta founder is bootstrapping a work platform he says can't be built by retrofitting chatbots onto pre-AI software.
AI backlash hardens: 71% oppose local data centers, $130B in projects blocked — Public opposition, junior-worker displacement, and memory-chip inflation are converging into an economic and political headwind the industry hasn't priced in.
Woodside Energy pushes agentic AI into LNG plant operations — The Australian energy producer is layering AI agents over a decade of operational data, starting with an LNG plant Startup Advisor copilot.
