Oct 21, 2026 · 40 min
Quantum risk reshapes the AI security race
Arvind Krishna & Nikesh Arora: Quantum Threat, Open vs Frontier AI, AI Cyber Warfare & SaaS Collapse
The episode connects quantum preparedness, AI-enabled cyber conflict, and enterprise consolidation into one contest over technological advantage.
- 1Organizations must migrate to quantum-resistant cryptography before practical quantum computers can threaten embedded enterprise keys.
- 2Falling inference costs and open-weight models broaden access while frontier labs compete for control across the technology stack.
- 3AI may accelerate vulnerability discovery and software production, but reliability, liability, and long testing cycles limit immediate disruption.
Don't miss
Arvind Krishna explains why agentically generated code is not automatically error-free or fully verifiable, complicating liability and cyber defense.
The brief
Arvind Krishna and Nikesh Arora shift the AI conversation toward quantum computing, arguing that enterprise systems must prepare for cryptographic threats before the hardware fully arrives.
The panel weighs whether AI or mathematical breakthroughs could undermine encryption first, while distinguishing the discovery of quantum algorithms from running them efficiently.
Open-weight and frontier models are framed as complementary tools: falling inference costs widen access, but production systems still require careful optimization and control.
AI cyber warfare creates an asymmetric contest among states, companies, and alliances, yet agentically generated code remains fallible rather than automatically verifiable.
The discussion ends with a broader market forecast: integrated AI stacks could pressure routers, cybersecurity vendors, and enterprise software companies, though mission-critical systems change slowly.
What was said on this episode
44 statements · 22 positive · 18 negative · 3 mixed · 1 neutral
“Well, it will break today's cryptography.”
Listen at 1:14
Post-quantum cryptography is understood and implemented, but organizations have not widely adopted it.
“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.”
Listen at 1:38
One or two quantum companies will demonstrate error-correction progress by 2027.
“You'll find one or two will make progress by 2027.”
Listen at 3:17
Quantum computers could reach 10 million gate computations by 2029.
“that means you could be at 10 million by 2029.”
Listen at 4:02
Organizations should immediately replace IT infrastructure with quantum-safe systems.
“you should be replacing your IT stack today with quantum-safe or quantum-proof infrastructure right now.”
Listen at 6:02
Organizations beginning quantum migration in 2029 or 2030 will already be late.
“the time— if he says it's 2029, '30, we're already late.”
Listen at 6:23
AI will not break cryptography before quantum computers do.
“It's not going to be possible.”
Listen at 7:26
AI can discover new quantum algorithms.
“Could you have an AI algorithm discover a new quantum algorithm? Absolutely.”
Listen at 7:50
AI cannot run newly discovered quantum algorithms efficiently on current hardware.
“Could it run it effectively and efficiently on today's hardware? Yes, it's not possible.”
Listen at 7:57
Classical hardware may run quantum algorithms 1,000 to 10,000 times slower.
“Otherwise, it'll be slower by an order of 1,000 or 10,000.”
Listen at 8:10
A cryptanalytic method taking ten years per key would not be practically useful.
“if it takes 10 years to break one key, most people are going to rest in peace. That's not a practical answer.”
Listen at 8:51
Organizations should use the best available model during experimentation.
“right now we're saying go use the best model you can find, the easiest way to use it”
Listen at 10:05
Small, task-specific language models can handle many enterprise tasks.
“we're going to use generative AI, small language models, and they're going to do that.”
Listen at 11:22
Most enterprise use cases will rely on trained open-source models.
“for most practical use cases in enterprises, we're going to end up training some version of an open-source model”
Listen at 12:21
Optimized Chinese open-weight models can reduce inference costs to $1–$2 per million tokens.
“we could drop the cost to about $1 to $2 for a million tokens.”
Listen at 13:14
Western open-weight models comparable to Chinese models will emerge within two years.
“over the next year or two, I'm very doubtful that there will not be great models that come up in the West that are also open weight.”
Listen at 14:09
The capability gap between Chinese and Western open-weight models is at most 6–12 months.
“The gap in these things is 6 to 12 months at the most.”
Listen at 14:19
“of all token usage will go to open weight models”
Listen at 14:29
Models trained on good code can identify bad code and vulnerabilities.
“If you teach a model what good code looks like, it also can figure out what bad code looks like.”
Listen at 16:48
AI-identified vulnerabilities are roughly 60%–80% real and 20% hallucinated.
“the number is roughly 60 to 80% have been, have been real and 20% have been hallucinations.”
Listen at 17:37
Cybersecurity providers can distribute defenses against discovered attack techniques to all customers.
“we can take that attack technique, write machine learning code, and ingest that into our product and make sure it gets available to every customer.”
Listen at 18:54
Agentically generated code will outperform human-written code.
“will agentic code be superior to human-written code? Absolutely.”
Listen at 20:08
IBM uses coding agents for about half its internally written code, with better results.
“We use agents and coding agents now for probably half the code that is written inside our place, and it is superior to what was being written before.”
Listen at 20:12
Fully verifying arbitrary agentic code is currently computationally infeasible.
“It's not fully verifiable because that is a task that is going to be almost computationally infeasible right now.”
Listen at 20:41
Governments and companies are not slowing AI development.
“I don't think anybody's slowing down.”
Listen at 23:18
Organizations that release AI systems are accountable for controlling them.
“if you build something and unleash it, you can't unleash and say, oops, I don't know how to control it.”
Listen at 23:33
Digital technology cannot be meaningfully slowed by unilateral regulation.
“a digital technology is impossible to actually slow down.”
Listen at 24:36
AI and data-center regulation is primarily driven by populism rather than technical realities.
“The regulation question on AI and data centers is driven— I'm going to be sorry and cynical— is driven by populism, not by any reality.”
Listen at 26:22
Current AI and data-center policy strikes the correct balance.
“I do think we are on the correct balance right now”
Listen at 26:45
All 50 states will not ban data centers, but 30 states might.
“I think there's no chance all 50 will ban. Is it possible 30 will ban? Yes.”
Listen at 26:54
AI access should become affordable enough for viable enterprises worldwide.
“I don't think that it's going to be so expensive that if you can run an economic enterprise, it should be possible in the envelope to be able to find a way to contract yourself.”
Listen at 27:45
Companies farther from AI compute hubs will have less access to compute.
“the further away you are from the epicenter, the less likely you will have access to this computer.”
Listen at 28:22
Nations prioritizing capital investment over labor protection historically tend to win technological transitions.
“People who spent more on capital than worried about labor tended to win every previous one.”
Listen at 29:04
IBM mainframes have endured for 65 years because of vertical integration.
“The IBM mainframe has lasted 65 years. Why? It was vertically integrated.”
Listen at 30:56
Nvidia is the modern equivalent of a vertically integrated technology company.
“the modern-day version of that is Nvidia.”
Listen at 32:00
Controlling a technology stack lets a company shift profits between layers.
“if you control the vertical stack, you can decide which part of the stack to commoditize.”
Listen at 32:23
Open-router models will become impractical as AI systems integrate memory and reasoning.
“this notion of running something like an open router become Impossible”
Listen at 33:08
AI application providers will build vertically integrated stacks spanning reasoning, memory, and context.
“you will find in the application layer they're going to have a vertical stack as well”
Listen at 34:04
Every frontier AI lab will aggressively pursue consumer personal-agent products within three months.
“In the next 3 months, every frontier lab is going to have a full-on, full-court press on the Muse concept”
Listen at 34:27
AI will cause major disruption in the SaaS market over the next few years.
“I think there's going to be huge disruption in the SaaS space over the next few years.”
Listen at 37:39
AI capabilities will cause massive consolidation among enterprise software vendors.
“That implies massive consolidation because that's where the spend goes.”
Listen at 38:26
Enterprise software will divide into AI-replaceable and reliability-critical categories.
“I do think also software should be distinguished.”
Listen at 38:51
Reliability-critical software will be difficult to replace with AI-generated software.
“I think that's going to be hard to replace that easily, right, with, uh, AI-generated software.”
Listen at 39:16
Long parallel testing eliminates the cost advantage of replacing critical software with AI-generated software.
“And then there's no cost advantage to do that.”
Listen at 39:28
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.