Why one of the world's most controversial investors thinks open source AI is the best thing that could happen to frontier labs
Who is Cathie Wood? She's a 70-year-old American investor and founder of ARK Invest — an investment firm focused exclusively on disruptive innovation. Her flagship fund, the ARK Innovation ETF (ARKK), was the world's best-performing global equity fund in 2020, returning over 170%, and it topped the charts again in 2023 and 2025. But her career has also had dramatic downturns: in 2023, Morningstar named the ARK funds the third-worst "wealth destroyers" of the preceding decade.
Wood is known for going against the grain. She was early on Tesla, Bitcoin and CRISPR — often when no one else believed in them. Her mentor was economist Arthur Laffer (of Reagan-era Laffer Curve fame), and she holds an almost religious conviction that technological disruption compounds value exponentially over time.
What did she say?
On August 9, 2026, Wood laid out a thesis that turns the conventional narrative on its head:
"Ironically, contrary to the narrative, open-weight models are becoming a major reason that OpenAI, Anthropic, and in our view SpaceXAI, are going to capture the majority of the revenue from AI-powered models."
She specifically named OpenAI, Anthropic (which has just filed an S-1 for an IPO valuing it at nearly $1 trillion) and SpaceXAI (Elon Musk's xAI) as the three big winners.
The argument: the security paradox
Wood's logic rests on a paradox:
Open-weight models keep getting better. Meta, Mistral and DeepSeek are releasing models that trail frontier labs' latest by only ~3 months — at roughly 87% lower token prices.
But the better the open models get, the more dangerous they become in the wrong hands. The UK's AI Security Institute found that open-weight models now match frontier cyber capabilities from just four to seven months ago.
That forces companies to buy premium security. When anyone can download a powerful AI system and run it without restrictions, the attack surface grows dramatically. Companies respond by investing more in managed security infrastructure — exactly what frontier labs sell.
Wood identifies four reasons the money is flowing to frontier labs rather than open weights:
| Factor | What frontier labs deliver |
|---|---|
| Security | Hardened, controlled environments with guardrails |
| Reliability | Managed uptime, SLAs and support |
| Updates | Continuous frontier upgrades |
| Compliance | Documentation, audit trails and accountability |
"The cheaper AI becomes for attackers, the more companies need the best defense."
Nvidia's $500B bombshell: infrastructure becomes the biggest bottleneck
On the same day as Wood's statement — August 10, 2026 — Nvidia announced a deal that cements her point in an entirely different way.
Nvidia has struck agreements with six of the world's largest asset managers — Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — to mobilize $500 billion in third-party capital for AI infrastructure. Jensen Huang called it "the first time technology chips have become an investable asset class," comparing GPUs to commercial real estate and highways.
"The computer is now part of the infrastructure, just like electricity, just like the internet." — Jensen Huang, CNBC
Larry Fink (BlackRock) called it "the next frontier of financial engineering" — a nod to the creation of mortgage-backed securities in the 1970s.
What does this mean in practice?
Nvidia's $500B package isn't just a big number. It's a signal that infrastructure has become the biggest bottleneck in the AI economy:
| Who | What they get |
|---|---|
| Hyperscalers | Financing for data centers without straining their own balance sheets |
| Frontier labs | Secure access to Nvidia hardware at unprecedented scale |
| Everyone else | A wall of capital requirements growing exponentially |
The four largest US tech companies are expected to spend $725 billion on data centers and AI equipment in 2026 alone. Moody's warns that these investments are squeezing free cash flow and pushing the tech giants into heavier debt.
And here's the point: not everyone can, or should, own the hardware themselves.
When infrastructure becomes a Wall Street asset class, access to compute becomes a matter of capital — not just technology. That creates a divide between those who can afford to buy Nvidia clusters outright, and everyone else.
Who is Cathie Wood, really?
To understand why her statement is turning heads, it helps to know her background:
- Graduated summa cum laude in finance and economics from the University of Southern California
- Worked at Capital Group (1977), Jennison Associates (18 years) and AllianceBernstein (CIO for global thematic strategies)
- Founded ARK Invest in 2014 after AllianceBernstein rejected her idea for actively managed ETFs based on disruptive innovation as "too risky"
- Her first four ETFs were seeded with capital from Bill Hwang (later convicted of fraud)
- Has consistently bet on Tesla, Bitcoin (25% of her net worth in BTC), gene editing and AI
- Lives in St. Petersburg, Florida, is a practicing Christian, and has donated funds to her old high school for an innovation institute for girls
Wood isn't impartial — ARK Invest holds positions in OpenAI, Anthropic and SpaceXAI alike. But her point about the security paradox is backed by independent data: the UK AISI's measurements show an accelerating curve in which open-weight models' offensive capabilities lag frontier ones by an ever-shrinking margin.
What does this mean for the open-weight movement?
Wood's thesis and Nvidia's $500B package point to the same reality:
The more powerful open models become without control mechanisms, the greater the demand grows for closed systems that can govern them.
And the more expensive infrastructure becomes, the fewer companies can host it themselves.
That doesn't mean open weight is bad — quite the opposite: MiniMax M3 (#1 on BenchLM), Qwen3.8-27B (14 GB VRAM at INT4) and Kimi K3 (2.8T parameters) prove that open models compete directly with closed frontiers on quality.
But it does mean that the infrastructure around the model is becoming at least as important as the weights themselves.
The Liviate perspective: neutral hosting as the bridge
Cathie Wood is right about the diagnosis: open-weight models are creating a security gap that needs to be filled.
Nvidia is right about the mechanics: infrastructure has become a capital asset class, and access to compute keeps getting more expensive.
But the prescription doesn't have to be yet another proprietary API from a US frontier lab. And it doesn't have to be a $500B Wall Street package either.
For European companies, there's a third option:
When you use a US frontier lab as your security provider:
- Your data passes through American servers
- You're locked into proprietary APIs with vendor lock-in
- You pay premium prices for something open weight can already deliver technically
- EU AI Act compliance sits on your desk alone
When you buy your own hardware:
- Multi-million-dollar investments in GPU clusters
- Ongoing upgrades every 18–24 months
- Operations, cooling, power and maintenance
- Still the full weight of the compliance burden
At Liviate, we combine access to the best open-weight models with the infrastructure that makes them accountable — without you having to own the hardware or send your data to the US:
✅ Pre-configured deployment of MiniMax M3, the Qwen3.8 family and more ✅ GPU pooling so you avoid multi-million-dollar hardware investments ✅ Data isolation + audit logs + GDPR compliance ✅ Guardrails that close the security gap without sending your data to the US ✅ Fixed monthly pricing, no surprises
Freedom requires control. Control requires infrastructure. Infrastructure doesn't require you to own it — only to be able to trust it.