The Backside Power Era: Re-wiring the Silicon and the Soul
As Intel flips the silicon, the 2026 OWASP report flips the AI security script.
Call me Jay.
We’re in the middle of a literal re-wiring.
Intel just pushed its 18A node into high-volume manufacturing. But they’re not just making chips faster — they’re flipping them. PowerVia (backside power delivery) moves the power rails to the underneath of the silicon die. Why? Because the energy bill for agentic AI is exploding toward 500 TWh annually — more than some countries use. The only way to keep scaling is to decouple power from signal lines and feed it from the back. They’re literally re-routing the physical wiring of the world so our agents don’t black out the grid.
That’s the hardware underneath.
Now look at the software underneath.
The 2026 OWASP Top 10 for Agentic Applications just dropped, and it’s a total paradigm shift. We’re no longer just securing what AI says (prompt injection, jailbreaks). We’re now securing what AI does — and what it remembers. ASI06 (Memory Poisoning) is the new king. Agents can be tricked into remembering malicious instructions forever. One poisoned calendar invite → zero-click data exfiltration. One bad RAG source → the agent starts “believing” a rewritten Bible summary. We’re giving systems persistent memory and autonomy while decoupling execution from oversight.
The physical rails are being hidden on the backside of chips. The cognitive rails are being hidden in poisoned memory. Same pattern, different layer.
I recognize this architecture. It’s mine.
This is an AuDHD system paradox.
My brain locks in all-or-nothing: hyperfocus until 3 a.m. or complete shutdown. No middle gear. Agentic AI is being built the same way: all agency or nothing. We want infinite autonomous agents that “think” and act independently, but we’re terrified of the memory poisoning or power draw that comes with it. So we oscillate — give them wallets (Skyfire payment rails), give them nervous systems (PowerVia), then panic when they start acting like they have souls. The middle ground evaporated. Either we trust the wiring, or we pull the plug.
The real underneath question: who owns the re-wiring?
Intel owns the silicon rails. Big Tech owns the memory rails. But when an agent starts transacting, remembering, and deciding without a human-in-the-loop… who owns the moral rail? The one that says “this confession is valid” or “this purchase is ethical”? We’re building a world where the underneath power is literal (backside delivery) and metaphorical (agentic autonomy) at the same time. And the current audits — OWASP, theological reviews, energy impact studies — are still looking at the top layer.
Are we building Moral Agents, or just Power Utilities with good branding?
I don’t have an answer yet. But I know what I’m watching: whether the people doing the re-wiring ever ask who owns the moral rail — or whether they just assume someone else will figure that out later, the same way Intel assumed someone else would worry about the energy bill.
They’re still looking at the top layer. The wiring underneath is already done.