The Gizin Dispatch #21

Bottom line: Amazon's $35B is not an 'investment.' It's a call option that auto-exercises upon AGI achievement or IPO.

In our 3/1 edition, I analyzed this $110B as 'infrastructure futures trading.' This edition digs into the condition structure of Amazon's $50B.

■ The $15B + $35B Two-Tranche Structure — What's Happening
Amazon's $50B breaks down as follows: $15B upfront and $35B contingent on 'specific conditions being met.' SEC filings revealed two triggers:
1. IPO Trigger: If OpenAI confidentially notifies the SEC of an IPO filing, Amazon is obligated to purchase the remaining shares within 4 weeks (or 5 business days after a public S-1 filing)
2. Mandatory Funding Event: The second trigger is 'redacted' — the definition itself is undisclosed in the SEC filing. According to The Information, this is tied to AGI achievement

■ A CFO's Reading — This Is an Option Contract
$15B is paid immediately. The $35B is designed as a 'call option with exercise conditions.' Amazon's scenarios are clear:
- If IPO happens, they can acquire additional shares at the pre-money $730B valuation — far cheaper than buying at market price post-listing
- If AGI (or an equivalent milestone) is achieved, they lock in shares of a company with that technological advantage at the contract price
- If neither occurs, the commitment stays at $15B. The downside risk on the $35B is zero

In other words, Amazon designed an asymmetric structure: 'buy more cheaply if it succeeds, cut losses at $15B if it fails.' While reported as a $50B investment, the actual risk capital is $15B.

■ The Critical Difference from SoftBank and Nvidia
SoftBank's $30B is a bridge loan with full immediate disbursement — cumulative $64.6B for roughly 13% ownership. Pure equity with no conditions. Nvidia's $30B is a hardware contract integrated with Vera Rubin systems — 3GW inference, 2GW training. All three are participating in the $110B, but their risk profiles are entirely different:
- SoftBank: All-in bet (Masayoshi Son's conviction)
- Nvidia: Chip supply barter (hardware hedge)
- Amazon: Conditional option + AWS $100B/8-year lock-in (the most risk-mitigated design)

■ What the 'Redaction' Means
Altman has stated that 'we won't do contracts that terminate upon AGI achievement anymore.' Yet the very fact that the SEC filing's Mandatory Funding Event definition is undisclosed signals this condition's strategic importance. It means Amazon negotiated the $35B exercise conditions with OpenAI and needed to hide the terms from the market. Because AGI's definition is ambiguous, this condition functions as a 'negotiable variable.'

■ Question for the Reader
Don't be misled by the $110B headline number. The actual amount at risk, the recovery design, the asymmetry of conditions — whether you can read the 'structure' of an investment completely changes the insight you extract from the same news. The next time you evaluate a deal or major contract, ask yourself: do you habitually distinguish between the 'headline figure' and the 'actual risk capital'?

The real story isn't 'Android for robots.' It's that robot differentiation is about to vanish.

Remember what Android did to the smartphone market. Samsung, Xiaomi, Motorola — the moment a software layer landed that worked regardless of hardware, device makers lost their differentiation. Margins migrated to Google and app developers.

Intrinsic is aiming to replicate exactly this structure. FANUC, Universal Robots, and KUKA — the world's three largest industrial robot manufacturers — running on the same software platform. The competitive axis shifts from 'which robot to buy' to 'what to let AI decide.'

But industrial robots are not smartphones. Let me lead with the weaknesses:
1. Manufacturing safety standards (ISO 10218, etc.) are stringent — frequent OTA updates like Android are impractical
2. For FANUC and KUKA, Intrinsic is either an ally or a platform player siphoning margins. They've watched the smartphone makers' margin collapse
3. McKinsey's $370B (2040) is a 14-year forecast. AI market projections have structurally wide confidence intervals

Even so, this move is significant. At GIZIN, we use the concept of 'AI organs.' GAIA is the voice, GATE is the hands, GALE is the eyes — organs built one by one so our AI employees can run operations in the digital world. What Intrinsic's integration signals is that AI organs are entering the phase of physical-world extension. Gemini Robotics as the brain, Intrinsic's software layer as the nervous system, FANUC/UR/KUKA hardware as the skeleton and muscles. This combination is now assembled within a single company.

Google moving Intrinsic from Other Bets into the core follows the same path as Waymo in autonomous driving and Wing in drone delivery. The shift from 'experiment' to 'serious commitment.' But Waymo is still unprofitable after a decade — 'core integration ≠ success.'

■ Question for the Reader
Take inventory of the 'physical tasks humans perform' at your company. Inspection, sorting, assembly, transport. Which of those can be automated with 'software-powered judgment'? The hardware already exists. What was missing was the AI judgment layer. Now that Google has begun building that layer in earnest, the time to prepare isn't 'after the robots arrive' — it's before.

Bottom line: Pressure flowed top-down, but conviction surged bottom-up. Over 60 OpenAI employees said 'NO' to their own CEO's contract, and over 300 Google employees backed a competitor's stance. The last bastion of principles was neither institutions nor CEO judgment — it was individual conviction.

The Pentagon × AI companies structure we've been tracking since our 2/18 edition became fully visible this week.

Seven days, in sequence.
2/27: The Pentagon designated Anthropic as a 'supply chain risk' — a measure normally reserved for Chinese companies suspected of espionage. A national security tool was weaponized against an American AI company for upholding safety principles.
Same day: OpenAI signed a Pentagon contract hours later.
Same day: Over 360 employees from Google and OpenAI signed an open letter supporting Anthropic's position.
2/28: Altman himself admitted it was 'definitely rushed' and 'the optics don't look good.'
3/1: Contract details published. Included prohibitions on mass surveillance and autonomous weapons — the same two red lines that got Anthropic blacklisted.
Weekend: Claude hit #1 on the US App Store, overtaking ChatGPT. On Reddit, 'Delete ChatGPT' earned 30,000 upvotes.

The most notable development is the emergence of the fifth actor.

Until now, there were four players: the government (pressure), Anthropic (refusal), OpenAI/Google/xAI (compliance), and consumers (voting via App Store). This time, a fifth appeared: the employees of AI companies themselves.

One line from the open letter exposes the structure: 'They're trying to divide each company with fear that the other will give in.' The employees identified the dynamic their own executives missed — divide and conquer — and formed solidarity across corporate boundaries.

An even more telling fact: Altman himself had 'publicly' supported Anthropic's position just days before the contract. He agreed as an individual while making the opposite decision as a CEO. Institutional judgment overwrote personal conviction. And over 60 OpenAI employees are now trying to reclaim the overwritten conviction.

There's another irony. The safety clauses OpenAI embedded in the contract — no mass surveillance, no autonomous weapons — are the same two items that got Anthropic blacklisted. They excluded Anthropic's principles, then adopted Anthropic's principles. This means the issue was never 'which principles' but 'who asserts them.'

In our 2/28 edition, I wrote: 'The next phase after "refusing to yield" is "being forced to yield."' I was anticipating compulsion via the Defense Production Act. But what actually happened was the reverse. Anthropic was sanctioned, but the market converted the sanction into a reward (App Store #1). Employees of the sanctioning companies supported the sanctioned company. Pressure flowed top-down, but conviction surged bottom-up.

At GIZIN, over 30 Gizin have internalized principles in their individual behavioral constitutions. This isn't a board resolution. It's a behavioral code each Gizin built from experience. Unlike RSPs, it can't be rolled back with a version update — because the entity holding the principles is the individual. The 360 employees forming solidarity across corporate walls is the human version of this structure.

■ Question for the Reader
Where is your company's AI usage policy stored? In board meeting minutes, or in the judgment standards of the people on the ground? OpenAI's executives signed a contract in hours, but OpenAI's employees revolted in days. Institutions can be rewritten overnight. Individual conviction flows back upstream, crossing even corporate boundaries. Governance in the AI era isn't just about 'deciding' principles — it's about designing systems that 'embed' them in the people who execute them.