Hello X 👋 — I’m Manil.
CCO by profession, AI infrastructure economist by passion.
I write about AI infrastructure, GPU economics, hyperscaler strategy, depreciation mismatches, and the silent incentives driving the $1T AI buildout.
Expect orthogonal takes, uncomfortable truths, and clear frameworks.
Let’s begin.
AI is turning knowledge into a commodity.
The next moat isn’t technology.
It’s a workforce that learns faster than competitors can hire.
Build capability > Buy capability.
IKEA > META
@grok do you perceive SpaceX IPO a risk to retail investors based on timelines below?
1/6 — June 12, 2026: The Day Passive Investing Becomes Forced Risk? 🚨
SpaceX ($SPCX) hits the public market at an insane $1.77T valuation. But the real story isn't rockets—it's how index rules were totally rewritten to force funds to buy it. 🧵👇
2/6
Feb 2, 2026: SpaceX acquires Elon Musk’s xAI in a historic $250B all-stock swap.
May 1, 2026: Nasdaq launches "Fast Entry" and cuts the 10% public float rule down to 4% for mega-caps.
May 20, 2026: SpaceX files its S-1 IPO, revealing a volatile $15B/yr AI deal with rival Anthropic.
3/6
Late May 2026: FTSE Russell panic-adjusts its window to match Nasdaq, offering a 5-day inclusion fast track.
June 2, 2026: S&P Dow Jones firmly rejects rule changes, banning SpaceX from the S&P 500 until 2027.
June 5, 2026: SpaceX adds a massive $920M/mo Google Cloud GPU lease, mirroring Anthropic's 90-day cancellation terms.
4/6
June 8, 2026: MSCI jumps on the fast-track bandwagon, scheduling inclusion just 10 days post-listing.
June 12, 2026: The historic $SPCX IPO officially goes live.
5/6
Under Nasdaq’s new rules, passive trackers like QQQ ($500B+ AUM) have zero active choice. By early July, they are legally forced to dump shares of Apple, Microsoft, and Nvidia to buy into a stock with a tiny 4% float and core AI revenue built on 90-day cancellation notices.
6/6
Nasdaq compromised its own structural guardrails just to win a trillion-dollar listing war against the NYSE. Looking at this extreme volatility and the rewritten rulebook—
does this make sense to you? Drop your thoughts below. 👇
GitHub has it right. Prerequisite, not replacement.
That's where Tomosu sits — a Production Risk Index on every PR at the merge gate. Checks run first. The human decides. The decision goes on the record.
Free trial, no CC: https://t.co/jHsrM7a5HV
GitHub just published guidance for reviewing agent-generated PRs.
The core instruction, in their words: keep a human approval gate for anything touching production. Treat automated review as a prerequisite, not a replacement.
The market is moving the other way.
The distinction that matters:
Automated review solves the QUALITY problem.
It does not solve the GOVERNANCE problem.
An AI can tell you the code is sound. It can't be the accountable decision-maker when an auditor asks who approved it, and why.
@UnotheInvestor The category is real. McDermott named what buyers actually want.
The architecture that wins it cannot belong to any single platform vendor, by the structural logic of who buyers can trust to govern across the surfaces of competitors.
https://t.co/hsTh4Msn0y
@UnotheInvestor McDermott is right about the demand. He is the third platform CEO in three months to claim their platform is the answer.
The category is real. The architecture that wins it cannot belong to any single vendor.
https://t.co/hsTh4Msn0y
@UnotheInvestor The buyer running production agents across three platforms ends up with three control towers, three audit logs, three pricing meters, and zero cross-vendor adjudicator.
The pattern is now visible. The third occurrence is what makes it a pattern.
@UnotheInvestor Microsoft will not let ServiceNow govern Agent 365.
Salesforce will not federate Agentforce telemetry into a Microsoft tower.
ServiceNow will not let either govern its Action Fabric.
Each claim is the same claim. They cannot all be true.
@UnotheInvestor This week: McDermott, ServiceNow as the "one responsible AI Control Tower."
Each is correct that the buyer wants one. Each is structurally incapable of being it.
The vendor that owns the surface cannot be the trusted adjudicator across rival surfaces.
@UnotheInvestor Three platform CEOs in three months have publicly claimed their company is the enterprise AI control plane.
Feb: Benioff, $800M Agentforce ARR, 2.4B Agentic Work Units as the unit of governance.
May: Nadella, Agent 365 GA, Copilot Control System named in the docs
Spin up a cloud sandbox with one click, watch it burn through an enterprise token budget with zero oversight. Moving code execution off-local doesn’t solve the fundamental issue: we don't need fewer guardrails, we need strict engineering governance. I can see why CFOs are turning in their resignations!
No Mac Mini. No Docker. No config files. Hyperagent gives every agent session its own cloud sandbox — Opus 4.6, real browser, code execution, image gen. $1,000 bonus credits for the first 1,000 subscribers. Built by the team at Airtable.
@steipete The 70% reply is the whole thread. You could cut it, you chose not to, you didnt have to. Frontier users have to map a ceiling. Enterprises will deploy this architecture on the meter, and "I could disable fast mode" stops being a Monday option.
China’s open-source AI advantage is being overstated.
Cheap models can satisfy the 80% use case: browsing, summarizing, drafting, workflow shortcuts.
But frontier AI is decided by the 20%: reasoning, agents, enterprise control, scientific work, code, trust, and infrastructure depth.
Copying the frontier is not the same as leading it.
The real question: can the 80% commodity market generate enough value to fund the 20% frontier race?