Cost Collapse, Platform Grabs, and a Leak: The Week AI Rewrote Its Own Power Map

Five days. Six stories that matter. Here's what your competitors are already acting on.

1. Google Publishes TurboQuant — And Crashes the Chip Market

March 24

Google Research dropped a paper that sent SK Hynix down 6.2%, Samsung down 4.8%, and Micron down 3.4% within hours. TurboQuant is a software-only compression algorithm that reduces AI inference memory by 6x — with zero quality degradation. Enterprise inference costs, cut 50% or more. No new hardware required.

ZeroForce take: The CFO excuse for not scaling AI just evaporated. The companies in production when this deploys will extend their lead. The companies still in pilot mode will find the gap even harder to close.

2. OpenAI Kills the Sora API — "Unsustainable Economics"

March 24

OpenAI announced it's shutting down the Sora public API with 30 days' notice. The stated reason: video generation at scale isn't economically viable. A proposed Disney deal that never closed was caught in the wreckage.

ZeroForce take: The first major AI product shutdown. Not because of safety concerns, not because of competition — because the unit economics didn't work. A useful reminder that "most-hyped" and "most-viable" are not the same category. Video AI is real; the business model isn't there yet.

3. Anthropic's Most Powerful Model Leaked Before It Shipped

March 26

Fortune broke the story: an unsecured content management system exposed ~3,000 internal Anthropic assets, including draft blog posts describing a model called Claude Mythos (internal name: Capybara). Anthropic confirmed its existence. Their words: "a step change in capabilities" and "the most capable model we've built to date." The catch: the same documents described Mythos as "far ahead of any other AI model in cyber capabilities" — specifically noting it "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace defenders."

Anthropic is releasing Mythos to cyber defense organizations first.

ZeroForce take: The model itself is the news. Anthropic accidentally published a warning about its own creation. Every CISO in a boardroom this week should be reading that leaked draft — not because they can access Mythos, but because it describes the threat landscape they're now operating in.

4. Apple Opens Siri to Rival AI Assistants

March 26

Bloomberg confirmed: iOS 27 will let users route queries from Siri directly to Gemini, Claude, Perplexity, or any App Store-approved AI chatbot. Siri becomes a routing layer. Apple takes App Store margin on every subscription that flows through its devices.

ZeroForce take: Apple can't out-model OpenAI. It doesn't need to. It owns the interface for 2.2 billion devices. The AI labs are becoming utilities. Apple is becoming the grid. Your enterprise AI vendor list just got a new criterion: iOS 27 compatibility.

5. Trump Appoints Zuckerberg, Ellison, and Huang to AI Policy Council

March 26

Per WSJ: the Trump administration is forming a 24-person AI Technology Council co-chaired by David Sacks. Mark Zuckerberg, Larry Ellison, and Jensen Huang are among the appointees. The council's mandate: AI policy.

ZeroForce take: The three people most financially invested in AI infrastructure are now co-writing the rules that govern it. That's not unusual in Washington. But the speed at which this is moving — and the regulatory tailwinds it creates for US AI investment — matters for any organization planning international AI deployments in 2026.

6. SoftBank Secures $40 Billion Loan to Boost OpenAI Investment

March 27

SoftBank closed a $40 billion credit facility specifically to increase its OpenAI position. Both companies are racing toward IPOs later this year.

ZeroForce take: The pre-IPO capital stack for OpenAI is now enormous. This isn't R&D money — it's market-positioning capital ahead of a liquidity event. When OpenAI goes public, it won't be valued on today's revenue. It'll be valued on the infrastructure SoftBank is betting it will own.

The Week in One Sentence

The cost of AI fell, the interface of AI shifted, the most powerful AI leaked, and the people building it are now also the people regulating it.

That's not a technology story. It's a power story.

Sources: Google Research, Fortune, Bloomberg, Reuters, WSJ, TechCrunch, Ars Technica — March 23–28, 2026