Arvind Krishna

1. Key Themes

The Quantum Cryptography Clock Is Already Running Because of "Harvest Now, Decrypt Later"

Arvind Krishna (IBM CEO) frames quantum as a staged, observable milestone path, and the real urgency is that data stolen today can be cracked later. DARPA's Quantum Benchmarking Initiative has winnowed from "a couple of hundred" entrants to about 13. Krishna says: "if you pick the middle point, let's call it 2030, well, it will break today's cryptography. That's not a supposition, that's just a reality." 00:00:37 Then the kicker: "if you have an encrypted hard disk today that gets stolen, somebody could keep it for 5 years and break it at that point. The good news is, post-quantum cryptography is understood, it's well known, the code is written, but people have been too lazy to get to it." 00:01:28 Krishna's two-milestone tracking framework: first, error correction on a running machine ("you'll see somebody make claims by next year... one or two will make progress by 2027"), and second, gate-count scaling ("are they adding a zero each year? ... you could be at 10 million by 2029"). 00:03:24

Replace the IT Stack Now: Cryptographic Migration Is a Multi-Year Capex Problem

Nikesh Arora (Palo Alto Networks CEO) argues the date that matters isn't Q-Day, it's the replacement cycle. "It's not like one day all of us are going to go out and say, 'Oh shit, let's go burn $5 trillion of IT infrastructure and just go build it again in one year because now we need quantum safe infrastructure.' So you should be replacing your IT stack today with quantum-safe or quantum-proof infrastructure right now." 00:05:23 He points to the budgeting angle: "everybody in this room has got few hundred million dollars of IT spend that is earmarked for this year, and I'm guessing 50% of your spend is on infrastructure that is not quantum-safe." 00:06:28 Stopgaps exist: "wrapping post-quantum ready sort of keys around traffic." 00:05:23

Model Routing by Task: "You Don't Need a Ferrari for Every Task"

Arora lays out how a $20-per-endpoint-per-year security product dictates model choice: "We inspect all traffic at an end point at the cost of $20 a year... Now if you want to take all that traffic, run that through OpenAI using tokens for every employee in your company... OpenAI will probably generate all their revenue in 1 year from one company." 00:09:36 The playbook: "you take a small language model, you compress it, you make it task-specific, you keep training it until it gets to 99.9% accuracy." 00:09:36 Frontier models are reserved for when value-at-risk justifies it: "let's assume you're undergoing a breach... At that point in time, I can run agent swarms... use the most up-to-date frontier LLM." 00:09:36

75-90% of Token Usage Goes Open-Weight: Economics, IP, and Geopolitics

Krishna makes a quantified prediction on three drivers. Economics: "$20 for a million tokens for probably generation minus one, $50 if you want to go to the most extreme... we could take one of the Chinese open weight models, take a Qwen or Kimi K3, something, 800 billion parameters... a week of engineering work, we could drop the cost to about $1 to $2 for a million tokens." 00:12:46 IP: "you're giving away your IP if that chain of thought can be captured and logged and audited by somebody else... you can't run a frontier model on your own because none of them will give it to you. So then you're back to open weight." 00:12:46 Geopolitics: "The best open weight models today are Chinese... The gap in these things is 6 to 12 months at the most... I don't know whether it'll be 75 to 90%, but that's the range I would put of all token usage will go to open weight models." 00:12:46 Arora adds the auditability angle: open weights with proprietary data "gives us the auditability, the explainability." 00:09:36

Cyber Is Asymmetric Warfare, and AI Finds Bad Code Because It Learned Good Code

Arora explains why attackers have the structural edge: "if you teach a model what good code looks like, it also can figure out what bad code looks like... the reason it's perfect for attackers is they just need one vulnerability. They don't need 10 vulnerabilities. If you're a defender, you need to find every vulnerability and patch it." 00:16:24 Even so, the tools are noisy: "roughly 60% to 80% have been real, and 20% have been hallucinations." 00:16:24 Krishna offers the military playbook: alliances, making defense as cheap as offense ("an attacker is cheaper than the defender, so the defender runs out of money and at some point then concedes"), and technological superiority: "If you get technological superiority on one side, you tend to win." 00:19:06

Agentic Code Is Better, But Not Verifiable

Krishna says IBM uses agents "for probably half the code that is written inside our place, and it is superior to what was being written before" because "the agents can also do evals and basically make sure that everything they can find is actually fixed before it ever goes out. All that said, we're only finding all the flaws that we know how to find." 00:19:06 On full verification: "It's not fully verifiable because that is a task that is going to be almost computationally infeasible right now." 00:20:41

Vertical Integration Returns, and Whoever Owns the Stack Controls Where Profit Sits

Krishna on the IBM mainframe's 65-year run: "In the '60s we went all the way to dirt, meaning we found sand, we made our own wafers, we built our own fabs... What that avoided was any IP leakage, but it allowed us to keep the profit of the complete stack in the company." 00:30:47 Arora identifies Nvidia as the modern analog: "if you control the vertical stack, you can decide which part of the stack to commoditize... if they destroy the profit pool in that stack, you move the profit to the other part of the stack." 00:32:11 Chamath Palihapitiya believes "we're moving back to the world of the conglomerate" as the safe harbor in turbulent times. 00:30:02

Context and Memory Become the Moat; Apps Go Headless

Arora argues the application layer will vertically integrate too: "You can't separate context and memory from the reasoning... Look at what's happening in the personal agent space... If you start training a personal agent with all your context, all your APIs, all your connectivity, now you've built a moat." 00:32:54 He predicts: "in the next 3 months, every frontier lab is going to have a full-on, full-court press" on consumer agents, "Every app that exists in the iPhone has to become headless," and "brute force distribution wins again in the enterprise space." 00:34:19

Capital-Intensive Nations Win; Compute Access, Not Cost, Creates Haves and Have-Nots

Krishna: "This is the classic labor versus capital trade-off... People who spend more on capital than worried about labor tended to win every previous one of the seven or eight cycles like this... Nations that get very hung up on... every existing job has to be protected, are the ones that are going to lose." 00:28:49 Arora's twist: the divide won't be price but supply: "the further away you are from the epicenter, the less likely you will have access to this compute. It's not because of economic reasons, it's purely because of supply constraints." 00:27:36

SaaS Consolidation: 500-5,000 Vendors Collapsing to 5-10 Platforms

Arora: "It's very hard to understand the terminal value of software. It's not clear what software is going to survive the next 5 to 8 years... I know exactly how it should work in an AI-first world. Which means somebody's going to do it." 00:35:18 Krishna quantifies: "The average enterprise uses anywhere from 500 to 5,000 distinct software vendors... could I replace 490 of them with minor AI capabilities, and do I need to go back and leverage 5 to 10 main platforms? That implies massive consolidation." 00:37:49 Boundary condition: interaction-rich software is replaceable, but not where "I need you to be reliable at 10 to the minus 15, which is a database lookup for somebody doing retail banking or closing the books." 00:39:04

2. Contrarian Perspectives

Quantum Can't Be Beaten by Classical AI Shortcuts

Against the view that AI-found math will crack encryption before quantum hardware does, Krishna argues AI can discover quantum algorithms but can't run them efficiently on classical hardware: "you cannot simulate something on a different architecture and expect a more efficient result... otherwise it'll be slower by an order of 1,000 or 10,000." 00:07:20 He concedes a classical number-theory path is possible, but "if it takes 10 years to break one key, most people are going to rest in peace." 00:08:44

Cybersecurity Won't Be "Consumed" by Frontier Models; Small, Cheap Models Win Defense

At the peak of the Mythos panic, the consensus was that AI labs would eat security. Arora counters with unit economics: inspecting traffic via frontier tokens would be economically absurd, so purpose-trained small models at 99.9% accuracy do the work. 00:09:36 Chinese open-weight models are "very little danger if you're putting them in a sandbox and running them by yourself." 00:12:46

Developers Are Lazy; Frontier Token Usage Is Mostly Default Behavior

Krishna, a former developer: "Developers are lazy. They're going to use the most expensive answer unless you put a control on it, because it's easy... doing it for production usage is where I am skeptical that it's of value, except for a tiny number of use cases." 00:14:45 That implies much of today's frontier revenue is behavior, not necessity.

The AI "Dark Side" Narrative Is a Marketing Problem, and Regulation Is Populism

Arora: "We've overrotated on the dark side of the model, where now the entire conversation is about the dark side and the perils of AI and not on the benefits... I think there's a narrative problem." 00:23:00 Krishna on regulation: "The regulation question on AI and data centers is driven, I'm going to be sorry and cynical, by populism, not by any reality." 00:25:54 He thinks the EU path is the warning: "you impose regulations first, and it gets you such a vibrant tech economy there, right?" 00:24:17 On a 50-state data center ban: "there's no chance all 50 will ban. Is it possible 30 will ban? Yes. The remaining 20 will rub their hands in glee." 00:25:54

Protecting Existing Jobs Is the Losing Strategy

Krishna: "Job displacement you should take, because it may result in a better job. I think that's the social philosophy that is going to determine winners and losers here." 00:29:19

3. Companies Identified

IBM

Enterprise technology company led by Arvind Krishna. Mentioned for its 65-year mainframe vertical integration and its quantum program. "The IBM mainframe has lasted 65 years. Why? It was vertically integrated." 00:30:47 It also uses coding agents for "probably half the code that is written inside our place." 00:19:06

Palo Alto Networks

Cybersecurity platform led by Nikesh Arora. Mentioned for running ML inspection at about $20 per endpoint per year and pursuing platformization. "We've tried to find things that are best in their category, stitch them, make them work so our customers don't have to do the stitching." 00:35:18 It was also part of Mythos's early preview circle. 00:15:20

Nvidia

AI chip leader. Arora calls it "the modern-day version" of the vertically integrated stack: "He's trying to fund everybody in the ecosystem to ensure that they don't stray from the vertically integrated stack that he's producing." 00:32:11 Chamath: "Jensen is at the dirt and he's at the application." 00:32:09

Anthropic (Mythos, Claude Code)

Frontier lab whose Mythos model triggered the cyber-disruption narrative. Arora: "the first thing that Mythos did, it figured out what bad code looks like, and then if you let it loose against your code, it'll tell you what is bad." 00:16:24 He also notes widespread Claude Code usage without ZDR. 00:16:24

OpenAI

Frontier lab, referenced for the math release that sparked the cryptography debate, and as a measure of token cost: "OpenAI will probably generate all their revenue in 1 year from one company." 00:09:36

Qwen and Kimi (Chinese Open-Weight Models)

Cited by Krishna as the best open-weight models today, enabling "$1 to $2 for a million tokens" against $20-$50 for frontier. 00:12:46

DARPA Quantum Benchmarking Initiative (QBI)

Government program that narrowed "a couple of hundred" entrants to "maybe 13," described as "what's maybe real." 00:00:37

Salesforce, Microsoft Dynamics, Oracle

Cited by Arora as the endpoint of CRM consolidation, from "23 systems" at Fidelity to off-the-shelf platforms, as the template for cyber consolidation. 00:35:18 Marc Benioff of Salesforce is also credited with aggressive zero-data-retention for customers. 00:15:20

Fidelity Investments

Arora's first employer, used as the CRM-consolidation analogy. 00:35:18

Facebook (Meta)

Arora says Facebook showed how the consumer-agent concept can work: "with Facebook showing us how it can work, I think pretty much everybody has it." 00:34:19

Hugging Face and OpenRouter

Mentioned in discussing open vs. closed stacks. Arora argues OpenRouter-style routing breaks down because "you can't separate context and memory from the reasoning." 00:32:54

4. People Identified

CEO of IBM. Delivers specific quantum milestones, an open-weight token share prediction (75-90%), the economics of $1-$2 per million tokens, and the SaaS consolidation math. "I'll predict you'll see somebody make claims by next year." 00:02:41

Nikesh Arora

CEO of Palo Alto Networks. Lays out the model-routing doctrine, the attacker/defender asymmetry, and the "headless app" prediction. "You don't need a Ferrari for every task." 00:09:36

Jensen Huang

Nvidia CEO, referenced by Arora and Chamath. "If you look at what Jensen is doing, he's a modern-day version of the vertically integrated stack." 00:32:04

Richard Feynman

Physicist whose work on simulation limits underpins Krishna's argument: "this goes back to stuff that people like Richard Feynman did, you cannot simulate something on a different architecture and expect a more efficient result." 00:07:20

Marc Benioff

Salesforce CEO, credited by Chamath as the best example of aggressive zero data retention: "he's been incredibly aggressive about making sure there's zero data retention and doing that on behalf of his customers." 00:15:20

Chamath Palihapitiya

Host and investor. A vertical integration and conglomerate proponent: "I think we're moving back to the world of the conglomerate." 00:30:02 He also flagged that model contracts have synthetic-data "trapdoors." 00:15:20

David Friedberg

Host who raised the AI-versus-quantum cryptography question and relayed a Congress member's prediction of a possible "50-state ban on data centers." 00:24:54

5. Operating Insights

Don't Cost-Optimize During the Experimentation Phase

Arora's rule for rolling out AI: "Don't restrict your teams in terms of capability in the experimentation phase. Don't start trying to cost optimize... 20,000 people have just discovered AI in the last 12 to 18 months... So right now we're saying, 'Go use the best model you can find, the easiest way to use it.'" 00:09:36 Krishna's complement is to move to cheaper models once proven for production. 00:14:45

Always Ask "What Level of Intelligence Does This Task Need, and What Does It Cost?"

Arora: "we're trying to figure out what task requires what level of intelligence, and what is the cost of that intelligence for that particular task." 00:09:36 Pair task-specific small models with the policy of escalating to frontier only when business value at risk is in the billions.

Pool Vulnerabilities Across Companies and Patch Once

Arora describes IBM's initiative: "if we're all finding vulnerabilities, why don't we pool our vulnerabilities and find the defense once, as opposed to every company in the room trying to go test all the open source vulnerabilities... take some trusted parties, give us your vulnerabilities, we'll patch them, we'll distribute the patches." 00:18:20

Triage AI-Found Vulnerabilities Because 20% Are Hallucinations

Budget for validation: "every vulnerability has to then be tested, understood... 60% to 80% have been real, and 20% have been hallucinations." 00:16:24 Even patches from Claude Code should be re-tested before shipping.

Vertically Integrate Only Where It Is Advantage, Not Commodity

Krishna: "where do you want to be, and where, if it's a commodity, then you should not be." 00:30:47 IBM still designs chips, engineering, and most software, but buys metal and outsources fabs for scale.

6. Overlooked Insights

Models Don't Contractually Forbid Synthetic Data Generation

Chamath's aside: "none of the models say that they won't generate synthetic data. It's literally not in there. You guys should look. There's all these outs. There's all these trapdoors." 00:15:20 Even with ZDR, the customer's workflows can feed a model's improvement, which is why Arora's conclusion ("these guys are not going to get into my business") hinges on open-weight, self-managed deployments.

Coding Telemetry Trains Offense: Every Non-ZDR Coder Is Teaching Attackers' Tools

Arora's briefly stated chain: "There's tons of people out there using Claude Code, who have no issue about that data training the models... if you teach a model what good code looks like, it also can figure out what bad code looks like." 00:16:24 Coding usage without data protections is quietly the training set for the vulnerability-hunting capability that favors attackers.

Cost-per-Action Pricing and Stack Friction as the Next Business Model

Arora's passing remark: "it'll be like a CPA you talked about earlier... It's a cost per action. The action will require you to go and traverse the entire logic from prompt till outcome... it'll be too high a friction for you to move your task from one stack to the other." 00:33:29 This implies pricing shifts from tokens to outcomes, and switching costs rebuild around accumulated memory and context, a defensible moat for whoever owns the full stack.