Sep 15, 2026 · 51 min
AI’s investment promise meets its safety and infrastructure risks
Mad Money w/ Jim Cramer 9/15/26
The episode connects AI’s market opportunity to harder questions about compute economics, cybersecurity, employment, competition, and governance.
- 1AI infrastructure can generate rapid returns, but chip supply, model efficiency, and demand will determine lasting profitability.
- 2Salesforce frames AI as a productivity layer spanning search, Slack, customer outcomes, and business operations.
- 3OpenAI argues that frontier development can continue alongside safety testing, cyber defense, cost reduction, and workforce adaptation.
Don't miss
Sarah Friar explains how OpenAI combines internal model testing, rapid vulnerability fixes, and diversified chip sourcing to manage AI’s security and infrastructure risks.
The brief
Jim Cramer opens with a market selloff driven by higher oil prices, rising bond yields, and broad stock declines before turning to AI’s investment case.
The NVIDIA discussion presents AI factories and compute demand as powerful growth engines, while stressing guardrails, model choice, rental pricing, and long-term profitability.
At Dreamforce, Marc Benioff describes Salesforce’s AI models, Slack integrations, flexible pricing, and partnerships as tools for improving productivity and revenue.
Sarah Friar argues that OpenAI can expand access to frontier models while improving safety, cybersecurity, healthcare access, efficiency, and the economics of compute.
The standout tension arrives when AI’s promise meets infrastructure risk: OpenAI is diversifying chips and testing models to find vulnerabilities before attackers do.
The episode closes on governance, China competition, workforce changes, and whether AI can create durable businesses without relying on an IPO.
What was said on this episode
36 statements · 30 positive · 1 negative · 2 mixed · 3 neutral
AI will be built properly enough to avoid humanity dying in 2030.
“We're not going to die in 2030. There are so many people in the world who are going to build AI properly.”
Listen at 2:18
AI can become useful, helpful, and safe simultaneously.
“And so we're going to create AI that's incredibly useful. It's going to be incredibly helpful. And it'll be incredibly safe. We could have all of that at the same time.”
Listen at 2:40
AI safety and testing are engineering problems.
“Safety is an engineering problem. Testing is an engineering problem.”
Listen at 2:51
NVIDIA AI factories currently recover invested capital in about one year.
“the return on investment capital is about one year right now”
Listen at 3:50
Countries, companies, and industries must build proprietary AI intelligence.
“every single country, every single company, every single industry must also build their own intelligence”
Listen at 4:24
AI-powered productivity increases companies’ ambition.
“when a company becomes more productive, and when they're powered by artificial intelligence, they become more ambitious”
Listen at 5:40
Engineering will remain necessary after AI adoption.
“There will be engineering after AI.”
Listen at 6:51
Companies should innovate quickly but avoid releasing unsafe AI products.
“companies should innovate as fast as possible, but they should never innovate so fast as to release unsafe products”
Listen at 8:06
Existing laws and regulations are sufficient for AI product reliability.
“we don't need new laws. We've got plenty of laws. We've got plenty of regulations”
Listen at 8:33
AI token-generation demand increased 25-fold in under one year.
“The token generation demand rate has gone up 25 times in less than a year.”
Listen at 9:24
AI demand is growing exponentially because AI has become useful.
“Demand for AI, because it's useful now, is just growing exponentially.”
Listen at 9:40
NVIDIA compute can serve as collateral because it is fungible and durable.
“the world's first computing system that you could use that could be asset-backed”
Listen at 10:57
NVIDIA Grace Blackwell rental pricing reached $16 per GPU hour.
“You can now rent an NVIDIA Grace Blackwell for... $16 per GPU hour.”
Listen at 12:02
Investors should own NVIDIA rather than trade it.
“Own it, don't trade it. That's what I do.”
Listen at 12:39
Salesforce Simulate reduced Shark Ninja’s failed product searches from 22% to nearly zero.
“we reduced his kind of no-ops, that is, people getting on a site, trying to find his product, and them saying, I can't find it, from 22% to almost zero”
Listen at 15:22
Salesforce is testing outcome-based pricing by building systems upfront for free.
“we'll build the system for nothing, and we'll take a percentage of your business outcome”
Listen at 17:36
Slack was Salesforce’s fastest-growing product with triple-digit bookings growth last quarter.
“It's the fastest growing product we have. It had triple-digit growth last quarter in bookings.”
Listen at 19:39
Salesforce expects its Anthropic investment to be highly successful.
“this Anthropic will be a huge win for us”
Listen at 21:33
Salesforce expects growth to reach the 50s next year.
“Next year, we're going to be talking about the 50s.”
Listen at 22:15
AI companies should make systems responsible and ethical.
“We want AI to... take care of these things and be held responsible and take care of these actions and be ethical.”
Listen at 23:27
Salesforce leadership says it has provided over $1 billion in grants.
“we've done more than a billion dollars in grants”
Listen at 25:31
OpenAI will slow frontier development when safety requires it.
“we have to pace the frontier and slow down. Absolutely, we're going to listen to our researchers and do that.”
Listen at 28:24
OpenAI AI helped Boston Children’s Hospital clinicians diagnose 40 unresolved diseases.
“40 undiagnosed, unresolved diseases now diagnosed with clinicians there”
Listen at 29:30
OpenAI will continue prioritizing additional compute investment.
“I'm still highly focused on getting more compute to keep that flywheel going.”
Listen at 30:15
OpenAI’s Luna price cut of 80% increased demand tenfold.
“We dropped the pricing there 80%, and we took demand up 10x.”
Listen at 30:52
OpenAI’s advertising business reached $1 billion in seven months.
“our ad business just hit a billion dollars in seven months”
Listen at 32:19
OpenAI’s Astra scored 100% on a cybersecurity benchmark.
“These models, Astra scored 100%. on the cyber bench”
Listen at 33:35
OpenAI’s model can identify software vulnerabilities across a company’s code.
“It's going to find all of these vulnerabilities.”
Listen at 33:53
OpenAI is developing safeguards to align advanced models with human intentions.
“we're also working on how do we create safety and alignment and guardrails around them so that they are doing what the human in the loop want them to do”
Listen at 35:45
People in health deserts have 30% lower life expectancy than those with healthcare access.
“if you live in a health desert, your life expectancy is... 30% less than someone who's got active access to health care”
Listen at 36:25
AI is making people’s jobs more interesting and broader.
“people are actually starting to make their jobs more interesting and broader”
Listen at 37:37
OpenAI reassigned 25% of engineers to fix vulnerabilities found by Astra.
“we took 25% of all of our engineers off of production-level tasks and said, your job now is to run alongside the model and fix as it finds”
Listen at 42:08
OpenAI raised $122 billion in the first quarter to increase financial flexibility.
“We raised $122 billion in Q1 of this year to give ourselves maximum flexibility.”
Listen at 44:52
OpenAI reduced credit-check costs from $200 to approximately $0.17 each.
“the cost difference, it went from $200 a check to about 17 cents”
Listen at 46:44
NVIDIA remains an excellent accelerator platform for AI training.
“NVIDIA is still an incredible platform of accelerators for training”
Listen at 48:23
OpenAI dynamically adjusts investment according to observed demand.
“we do that in a very dynamic way based on where we see the demand coming from at any point in time”
Listen at 49:37
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.