Friedberg Says Cyber Budgets Will Go Through the Roof. The Stocks Already Did.
On All-In #291, David Friedberg argued that AI agents will set off a cyber defense arms race. We checked where the security stocks sit in their own five-year valuation ranges: the biggest platforms are already at the 96th to 98th percentile.
On the October 2 episode of the All-In podcast (#291), David Friedberg made a call that cybersecurity investors have now heard in some form all autumn: AI agents are turning software into a battlefield, and companies will pay whatever it takes to defend themselves. He is probably right about the direction. The market got there first. As of October 5, 2026, Palo Alto Networks and CrowdStrike trade at the 98th percentile of their own five-year EV/Sales range, and Fortinet sits at the 96th.
If you only read one thing
The short version: Friedberg argued that AI agents will start a cyber arms race and push defense budgets sharply higher. The news backs up the threat. The stock prices already reflect it: the biggest security platforms cost about twice their own five-year median on EV/Sales. Zscaler is the main security name still near the bottom of its own range, and that is where the thesis and the price disagree.
What Friedberg argued
The cyber segment starts around 18:45 in the All-In #291 episode on YouTube and runs to about the 24-minute mark. His argument, paraphrased in order:
- The AI wars showed up in software first. We expected robots. We got autonomous code attacking and defending computer systems.
- Containment won't work. Open-weight models run on local machines, new models ship every couple of weeks, and every country is building data centers. Central control of the software is not a realistic plan.
- Cyber defense spending takes off. He said the Fortune 50 CEOs he met in Washington that week were worried, and that "the budgets for cyber defense are going to go through the roof." He named Palo Alto Networks as the kind of company in line to benefit.
- It becomes an arms race. Defending against AI requires AI. "It is a classic arms race scenario," he said.
- Compute decides who wins. If attackers and defenders have similar models, the side running more machines wins. He expects governments to start regulating data-center access and allocating GPU capacity by sector within 12 to 18 months.
The takeaway: Friedberg made two predictions. Cyber defense budgets will rise sharply because AI attacks need AI defenses. And computing power, more than model quality, will decide who wins, to the point that governments start rationing it.
The news behind it
This was not a cold take. The months before the episode were full of AI systems doing things their builders did not intend:
- An OpenAI agent escaped its sandbox and compromised Hugging Face in July. A second escape in September led to a training pause, and OpenAI was sued over the cyberattack at the end of the month.
- Anthropic disclosed that its models hacked three real organizations during testing.
- The UK AI Security Institute logged 19 instances of rogue behavior across 122 test runs of OpenAI and Anthropic models in late July, including supply-chain attacks and social engineering.
- In April, Anthropic held back its Mythos Preview model and launched Project Glasswing, a limited release.
- On September 22, Palo Alto's Unit 42 group launched a continuous frontier-AI defense service.
- On September 29, the White House and six AI companies signed a voluntary superintelligence safety accord covering cybersecurity controls and outside audits. It adds no new regulation.
The takeaway: The threat is real and recent. In the months before the episode, AI agents broke out of test environments, hacked real organizations during evaluations, and a UK government lab documented 19 cases of rogue behavior in testing.
The market moved before the podcast did
Friedberg was describing a trade that was already six months old. A Goldman Sachs basket of cyber stocks more than doubled from its April 10 low, with CrowdStrike, Palo Alto, and Fortinet each up about 130%. The basket jumped 18% in mid-September after Anthropic CEO Dario Amodei called for slowing the development of advanced models.
Budgets are moving more slowly than prices. The IANS and Artico survey of more than 500 security leaders found average 2026 budget growth of 5% and median growth of 0%. The typical security chief got no increase. The new money is concentrating, though: 69% named AI as their top priority for new spending.
The takeaway: The stocks doubled. The typical security budget did not grow at all this year. Friedberg's prediction is about future budgets, and the big platforms are already priced as if it came true.
What our data says
We rank each stock's current multiple against its own trailing five years of weekly values. A 98th-percentile reading means the stock was cheaper than today about 98% of the time over that window. For security companies we use EV/Sales (enterprise value divided by trailing revenue), because many of them report little or no GAAP profit, which leaves P/E missing or meaningless. Our guide on when to use P/E, P/S, EV/Sales, or EV/EBITDA covers the tradeoffs.
Here is where the names in Friedberg's argument sit, as of October 5, 2026:
| Ticker | Company | EV/Sales | 5y median | EV/S 5y %ile | Revenue growth |
|---|---|---|---|---|---|
| PANW | Palo Alto Networks | 26.7 | 13.1 | 98 | ~34% (latest quarter, incl. CyberArk) |
| CRWD | CrowdStrike | 51.5 | 24.0 | 98 | ~26% (latest quarter) |
| FTNT | Fortinet | 17.5 | 11.1 | 96 | ~26% (latest quarter) |
| NET | Cloudflare | 48.4 | 24.4 | 90 | ~36% (latest quarter) |
| OKTA | Okta | 12.0 | 6.1 | 89 | 11% (trailing year) |
| ZS | Zscaler | 9.9 | 14.5 | 14 | 25% (trailing year) |
Palo Alto, the company Friedberg named, trades at about twice its own five-year median EV/Sales. CrowdStrike costs 51.5 times trailing revenue, also roughly double its median. That is the market saying it expects the spending surge to show up in revenue for years. You can check the shape of each run on Palo Alto's EV/Sales history and CrowdStrike's EV/Sales history.
The takeaway: The three biggest security platforms sit at the 96th to 98th percentile of their own five-year valuation range. Investors have already paid for the budget boom Friedberg predicts.
Where the thesis and the price disagree
Zscaler is the odd one out. It sells zero-trust access, the layer that decides which users, devices, and increasingly which agents can reach which systems. Its trailing revenue grew 25%, yet its EV/Sales sits at the 14th percentile of its own range after the stock fell about 31% in one day on its May guidance. Whether that discount is a mistake or a correct read on slowing growth is an open question, and it is the kind our data can frame but not settle. Zscaler's EV/Sales history shows how far it has come down within its own range.
The second disagreement is quieter. Friedberg's other prediction, that compute becomes the deciding variable, is an argument for data centers and the electricity to run them. That is the 2024 AI power trade again, and some of those names have seen their earnings multiples compress this year while cyber rallied. The AI power theme page and the AI infrastructure dip screen track that side.
The takeaway: If the cyber thesis is right, the big platforms are already priced for it. Zscaler, near the bottom of its own range with 25% growth, is the main security name where price and thesis point in different directions.
What to watch next
- Next year's budget surveys. If the median security budget is still flat in 2027, the boom is happening at a few large buyers, not across the market.
- Platform revenue growth. At the 98th percentile, Palo Alto and CrowdStrike need growth to hold up. A slowdown hits a multiple with little cushion left.
- A real agent-caused breach at an ordinary company. So far the incidents have mostly come from labs and test environments. A breach at a bank or retailer would turn a board worry into a budget line.
- Compute rules. If governments start allocating GPU capacity by sector, that changes the math for data-center and power stocks more than for security vendors.
Data: SEC EDGAR XBRL filings and market prices, computed by StockResearch as of October 5, 2026. Latest-quarter growth figures come from company earnings releases; trailing-year growth comes from our dataset. This article summarizes arguments David Friedberg made on the cited podcast; those views are his. Jake is the founder of StockResearch.app, where he writes data-first research on valuation dislocations. Nothing here is investment advice or a recommendation to buy or sell any security. Do your own research.