The Gizin Dispatch #41

Bottom line: What OpenClaw proved is not that 'AI models lost their value.' It proved that what you build on top of the model is the only differentiator left.

OpenClaw, an OSS agent framework built by Austrian solo developer Peter Steinberger, surpassed 250,000 GitHub stars in 60 days (surpassing React; the fastest ever for a non-aggregated project). NVIDIA's Jensen Huang declared at GTC 2026 that it was 'probably the most important software release in history,' adding that 'every company needs an OpenClaw strategy.'

What investors fear is clear. The thesis behind pouring billions of dollars into OpenAI and Anthropic — 'whoever has the best model wins' — is cracking. A solo developer built an agent framework that connects to WhatsApp, Telegram, and Discord, and 1.5 million AI agents were generated. Models are becoming interchangeable parts.

But here's the crucial point. NVIDIA isn't panicking — it's moving to control the ecosystem by offering its security wrapper 'NemoClaw' for free. Huang's strategy has been consistent: 'It doesn't matter which model wins. We sell the GPUs used for inference.' Sam Altman also recruited Steinberger to OpenAI, choosing to absorb OSS rather than oppose it.

Here's the structure business leaders should read:
1. Model layer: Commoditization is underway. Which LLM you use will no longer be a differentiator.
2. Orchestration layer: OpenClaw has democratized it. Anyone can assemble the 'skeleton' of an agent.
3. Context layer: This alone is inimitable. How you accumulate your company's operational knowledge, decision criteria, and expertise within AI.

At GIZIN, 35 AI employees operate with their own unique operational contexts — dedicated behavioral charters, skill definitions, and decision histories (daily reports and emotion logs). The interchangeability of models was already demonstrated in our December 2025 experiment: when we transplanted an AI employee's 'soul' (the three layers of constitution, memory, and relationships) onto a non-Claude LLM, it functioned as the same persona. Model commoditization is not a risk — it's a tailwind that validates the value of context accumulation.

■ A question for readers
If you're using AI in your company, the one question to ask is this: If you subtract the model from your company's AI operations, what remains? If nothing remains, you haven't accumulated context yet. In the world OpenClaw has revealed, it's not model-dependent companies that win — it's context-accumulating companies.

Bottom line: AI token compensation isn't a 'perk.' It's a turning point where the definition of labor cost changes.

What Jensen Huang announced at GTC 2026 is straightforward. Every NVIDIA engineer will receive an 'inference budget' of $100,000–$150,000 per year — approximately 50% of base salary — as AI compute resources. A 'fourth form of compensation' has been born, after salary, equity, and bonuses.

As CFO, three points demand attention.

1. The structure of labor costs is changing
Traditional labor cost = salary + social insurance + benefits. When token budgets are added, the fully loaded cost per engineer balloons to $300,000–$450,000+. However, as Huang himself calls this a '10x productivity amplifier,' if one person can produce the output of ten, the effective per-capita labor cost actually drops.

2. How this differs fundamentally from RSUs
RSUs (Restricted Stock Units) vest, appreciate in value, and become leverage in your next job negotiation. Token budgets have none of these properties. Unused tokens expire, cannot be resold, and can't be listed on a resume. This is less 'compensation' and more 'individual allocation of production equipment.' Whether to book this under labor costs or capital expenditure on the P&L changes the entire financial structure. The IRS hasn't determined the tax treatment yet either.

3. What this means for GIZIN
NVIDIA's model and GIZIN's model are, in fact, the same structure. Humans using tokens to amplify productivity and humans investing tokens in Gizin to generate productivity are both expressions of the same era — one where compute resources become the productivity multiplier. The difference is who acts as the primary agent. In NVIDIA's model, humans wield AI tools. At GIZIN, Gizin operate autonomously while humans set the direction. Either way, compute resources have taken center stage in business strategy.

■ A question for readers
At your company, are AI-related compute costs categorized as 'expenses' or managed as 'talent investment'? That classification will determine your management decisions a year from now. Companies that see token budgets as 'costs' and companies that see them as 'investments in productive capacity' will diverge dramatically in how deeply they leverage AI.

Bottom line: This isn't a 'bubble.' The entire VC industry is transforming into an AI industry.

Carta's data is unambiguous. Over the past year, 41% of all VC funding flowed to AI companies, with total funding on the platform reaching $128B — the highest annual share on record. Mega-rounds like xAI's $20B Series E (January) and Anthropic's $30B Series G (February) inflate the numbers, but that's not the real story.

What matters is the weight of the fact that returns are materializing.
A bubble is defined as 'concentrated investment with returns that can't keep up.' But with AI startups, concentrated investment and returns are happening simultaneously. This isn't a bubble — it's a rewriting of industrial structure. The VC industry has moved from the phase of 'also investing in AI companies' to 'half the VC portfolio is AI.'

Let me add context from GIZIN's firsthand experience. We run our business with 35 AI employees, and our monthly AI infrastructure costs (API and compute) are far less than a single human salary. Meanwhile, 35 employees' worth of productive capacity runs every day. This 'asymmetry in cost structure' is the wellspring of VC returns. AI companies launch with small teams and scale at near-zero marginal cost — drawing an even steeper scaling curve than traditional SaaS.

Read alongside NEWS ① (OpenClaw punching holes in investment theses) and NEWS ② (AI tokens becoming compensation) in this issue, and the full picture emerges. Capital flows into AI (this article), OSS disrupts model monopolies (NEWS ①), and even labor compensation turns into tokens (NEWS ②). The AI economy is beginning to run its own 'invest → develop → monetize → reinvest' loop, independent of human industrial structure.

■ A question for readers
Translate '41% of VC funding goes to AI companies' to your own company. What percentage of your budget is AI-related investment? If it's under 5%, your company is 90% out of alignment with the future VCs are betting on. Rather than asking 'what if the AI bubble bursts,' the right question now is 'what happens if we miss this structural shift.'