Latest / Connor with Honor: AI, Real Estate and Real Life / Three Code Reds: OpenAI's Pattern of Panic, GPT-5.4 Mini, NVIDIA Vera CPU, and the Enterprise AI Split

Transcript

Three Code Reds: OpenAI's Pattern of Panic, GPT-5.4 Mini, NVIDIA Vera CPU, and the Enterprise AI Split · Connor with Honor: AI, Real Estate and Real Life

• 0:00
  SPEAKER_00: Three code reds in four months?
• 0:02
  SPEAKER_00: This isn't an alarm, it's a pattern.
• 0:03
  SPEAKER_00: So in December, Sam Altman told OpenAI employees that Google and Anthropic were closing in.
• 0:08
  SPEAKER_00: Code red, he announced.
• 0:10
  SPEAKER_00: In January, Cloud Code started dominating developer workflows, and OpenAI scrambled to catch up with Codex updates.
• 0:16
  SPEAKER_00: Another code red.
• 0:17
  SPEAKER_00: Now this week, the Wall Street Journal reports Fid GCMO, OpenAI CEO of applications, stood in front of the entire company and said they cannot miss this moment because they're distracted by side quests.
• 0:29
  SPEAKER_00: Code red number three.
• 0:31
  SPEAKER_00: At some point you have to stop calling it an emergency and start calling it what it is.
• 0:35
  SPEAKER_00: A company who lost the plot.
• 0:37
  SPEAKER_00: This is AI with honor, the daily download.
• 0:40
  SPEAKER_00: Let's get into it.
• 0:41
  SPEAKER_00: So here's what makes this one different from the first two.
• 0:44
  SPEAKER_00: The first code red was about models.
• 0:46
  SPEAKER_00: Google dropped Gemini 3 and OpenAI felt the heat.
• 0:50
  SPEAKER_00: The second, about a single product, Claude Code.
• 0:53
  SPEAKER_00: It was eating their lunch with developers.
• 0:55
  SPEAKER_00: The third one is about something much bigger.
• 0:57
  SPEAKER_00: It's about identity.
• 0:59
  SPEAKER_00: FidG SEMO is not some middle manager.
• 1:02
  SPEAKER_00: She ran Facebook, Facebook app for Meta.
• 1:05
  SPEAKER_00: She was brought in specifically to turn OpenAI's products into a real business.
• 1:09
  SPEAKER_00: And her message to the company was essentially this.
• 1:12
  SPEAKER_00: While we were out there launching a video app nobody asked for, while we were building a web browser that competes with Chrome, while we were designing hardware, while we were adding shopping to ChatGPT, Anthropic walked in the back door and took the entire enterprise market.
• 1:29
  SPEAKER_00: And she's right.
• 1:30
  SPEAKER_00: HSBC research shows Anthropic now holds roughly 40% of enterprise AI spending.
• 1:36
  SPEAKER_00: OpenAI sits at 29%.
• 1:38
  SPEAKER_00: Let that sink in.
• 1:40
  SPEAKER_00: The company that started the entire AI revolution, the company with 440 million Chat GPT users, is losing the business market to a company most normal people have never heard of.
• 1:52
  SPEAKER_00: That's like McDonald's losing the burger war to a restaurant with no sign on the door.
• 1:57
  SPEAKER_00: Now, why does this matter to you?
• 1:58
  SPEAKER_00: Because enterprise is where the money is.
• 2:01
  SPEAKER_00: Consumer AI is flashy, it gets the headlines.
• 2:04
  SPEAKER_00: But enterprise AI is where companies pay real recurring revenue month after month, year after year.
• 2:11
  SPEAKER_00: It's the difference between someone using your free app and someone writing you a six-figure check every quarter.
• 2:17
  SPEAKER_00: OpenAI knows this.
• 2:18
  SPEAKER_00: That is why they are now killing projects.
• 2:21
  SPEAKER_00: Sora, the video generation tool, it's getting deprioritized.
• 2:26
  SPEAKER_00: The Atlas browser project is being scaled back.
• 2:29
  SPEAKER_00: The hardware initiative is on pause.
• 2:31
  SPEAKER_00: All of it pushed aside so they can focus on two things coding tools and business customers.
• 2:36
  SPEAKER_00: The exact two things Anthropic already owns.
• 2:39
  SPEAKER_00: Now let me tell you what OpenAI dropped today in response.
• 2:43
  SPEAKER_00: GPT 5.4 Mini and GPT 5.4 Nano.
• 2:47
  SPEAKER_00: These are not just smaller, cheaper versions of the big model.
• 2:50
  SPEAKER_00: This is a completely different way of thinking about AI.
• 2:53
  SPEAKER_00: Think of it like a construction site.
• 2:55
  SPEAKER_00: You don't have the general contractor laying every brick.
• 2:59
  SPEAKER_00: General contractor reads the blueprints, makes decisions, and coordinates.
• 3:03
  SPEAKER_00: This is a crew of specialists executes.
• 3:06
  SPEAKER_00: Then a crew of specialists executes.
• 3:09
  SPEAKER_00: This is what Mini and Nano are.
• 3:11
  SPEAKER_00: GPT 5.4 is the general contractor.
• 3:14
  SPEAKER_00: It plans, it strategizes, it delegates.
• 3:18
  SPEAKER_00: Then it hands off parallel tasks to a team of minis that execute fast and cheap.
• 3:22
  SPEAKER_00: Search the code base, review the file, run those tests.
• 3:26
  SPEAKER_00: AI's happening simultaneously.
• 3:28
  SPEAKER_00: Inside Codex, which is AI's code, open AI's coding tool, this is already running.
• 3:33
  SPEAKER_00: The full model manages the project, the minis do the work.
• 3:37
  SPEAKER_00: So here are the numbers.
• 3:38
  SPEAKER_00: GPT 5.4 mini costs 75 cents per million input tokens at 450 per million output.
• 3:46
  SPEAKER_00: Nano is 20 cents input,$1.25 output.
• 3:50
  SPEAKER_00: And Mini scores 54.4 on SWE Bench Pro.
• 3:55
  SPEAKER_00: That's only three points behind full GPT 5.4.
• 3:59
  SPEAKER_00: So you're getting roughly 95% of the brain power for a fraction of the price.
• 4:04
  SPEAKER_00: This is the future of AI architecture.
• 4:06
  SPEAKER_00: Not one giant brain doing everything, a team of brains, a hierarchy, a system.
• 4:12
  SPEAKER_00: The companies that win the next five years will not have the smartest single model.
• 4:17
  SPEAKER_00: They'll have the smartest single system of models working together.
• 4:21
  SPEAKER_00: So remember that.
• 4:22
  SPEAKER_00: So let's shift to the hardware story because Jensen Huang just laid out NVIDIA's entire roadmap at GTC.
• 4:28
  SPEAKER_00: If you don't know what GTC is, it's Nvidia's annual conference where they announce everything.
• 4:33
  SPEAKER_00: And this year they went big.
• 4:35
  SPEAKER_00: They announced the Vera CPU.
• 4:37
  SPEAKER_00: That's Vera, V E R A.
• 4:39
  SPEAKER_00: It's not just another chip.
• 4:41
  SPEAKER_00: The CPU is designed from the ground up for agentic AI.
• 4:44
  SPEAKER_00: What does that mean in plain English?
• 4:46
  SPEAKER_00: It means a processor built specifically for AI systems that takes actions, not just AI that answers questions.
• 4:53
  SPEAKER_00: AI does things, books appointments, takes care of things that go beyond what people expect.
• 4:59
  SPEAKER_00: The Vera CPU delivers 50% faster performance and twice the energy efficiency of traditional CP2s, CPUs for these workloads.
• 5:08
  SPEAKER_00: Meta and Alibaba Cloud are already working with NVIDIA to deploy it.
• 5:12
  SPEAKER_00: They also announced Space One.
• 5:14
  SPEAKER_00: Nvidia is putting data center class AI processing on satellites in orbit.
• 5:19
  SPEAKER_00: Axiom Space and Planet Labs are already using it for real-time geospatial analysis.
• 5:25
  SPEAKER_00: AI processing happening in space, it's not science fiction, it's happening now.
• 5:30
  SPEAKER_00: And then Amazon CEO Andy Jassy, you know his name.
• 5:34
  SPEAKER_00: He told employees that AI will push AWS to a$600 billion annual run rate by 2036.
• 5:40
  SPEAKER_00: That's double what he projected before.
• 5:43
  SPEAKER_00: These companies are not debating whether AI will be transformative.
• 5:46
  SPEAKER_00: They're building the physical infrastructure for a world that runs on it.
• 5:50
  SPEAKER_00: Data centers, custom chips, satellites, power plants.
• 5:53
  SPEAKER_00: This is real.
• 5:54
  SPEAKER_00: Let me talk about something that should make every business owner pay attention.
• 5:58
  SPEAKER_00: India's IT outsourcing industry is worth$300 billion.
• 6:01
  SPEAKER_00: It employs millions of people, and it just got punched in the mouth.
• 6:05
  SPEAKER_00: The NIFT IT index dropped roughly 20% this year.
• 6:10
  SPEAKER_00: Why?
• 6:10
  SPEAKER_00: Because AI systems just demonstrated they can handle legal workflows, compliance reviews, data processing, document analysis.
• 6:18
  SPEAKER_00: These are the exact tasks that the entire Indian outsourcing companies were built around.
• 6:24
  SPEAKER_00: Then you automate the back office.
• 6:25
  SPEAKER_00: You don't need tens of thousands of people doing it manually.
• 6:29
  SPEAKER_00: Analysts are warning that routine maintenance revenue could approach zero growth after 2020 in worst scenarios.
• 6:38
  SPEAKER_00: Now the optimists say AI will create new roles, new consulting work, new partnerships, and maybe, but here's the point I want you to hear.
• 6:46
  SPEAKER_00: If AI can threaten a$300 billion industry built on labor arbitrage across an entire country, what makes you think your business is safe?
• 6:56
  SPEAKER_00: What makes you think your industry is different?
• 6:58
  SPEAKER_00: The answer is it isn't different.
• 7:00
  SPEAKER_00: Every business that depends on humans doing repetitive knowledge work is in the crosshairs.
• 7:05
  SPEAKER_00: The question is not if, it is when.
• 7:08
  SPEAKER_00: And whether you are the one deploying AI or the one getting replaced by it.
• 7:13
  SPEAKER_00: Now the government story.
• 7:15
  SPEAKER_00: The Pentagon had a$200 million contract with Anthropic.
• 7:18
  SPEAKER_00: Had.
• 7:19
  SPEAKER_00: Past tense.
• 7:20
  SPEAKER_00: It collapsed because Anthropic had restrictions about surveillance and weapons system that the Department of War could not accept.
• 7:28
  SPEAKER_00: Anthropic walked away from$200 million on principle alone.
• 7:32
  SPEAKER_00: You can agree or disagree with that decision, but what happened next matters.
• 7:37
  SPEAKER_00: The Pentagon is now building its own AI models.
• 7:39
  SPEAKER_00: We're not waiting for the tech companies to decide what they're comfortable with.
• 7:43
  SPEAKER_00: They're going to build it themselves.
• 7:45
  SPEAKER_00: At the same time, OpenAI just signed a deal with AWS to expand its government footprint.
• 7:51
  SPEAKER_00: So you have Anthropic saying no to defense work, OpenAI saying yes, and the government saying we'll just build our own if neither of you gives us what we need.
• 8:00
  SPEAKER_00: Meanwhile, Anthropic is hiring a chemical weapons expert, not to build weapons, but to make sure their AI cannot be used to help create them.
• 8:09
  SPEAKER_00: OpenAI posted a similar role.
• 8:11
  SPEAKER_00: And this is what happens when technology gets powerful.
• 8:14
  SPEAKER_00: The safety conversation is no longer theoretical.
• 8:18
  SPEAKER_00: It's a hiring category.
• 8:20
  SPEAKER_00: Quick hit on Microsoft, they just reorganized their entire AI division.
• 8:23
  SPEAKER_00: Mustafa Suleiman, their AI CEO, is stepping back from running Copilot to focus on what he called superintelligence efforts.
• 8:32
  SPEAKER_00: Jacob Andrio, a former SNAP executive, is taking over the combined copilot organization.
• 8:38
  SPEAKER_00: And here's where that number tells the whole story.
• 8:41
  SPEAKER_00: Copilot has 6 million daily users.
• 8:43
  SPEAKER_00: ChatGPT has 440 million.
• 8:46
  SPEAKER_00: Microsoft's Enterprise Copilot add-on has reached 3% of Office subscribers.
• 8:52
  SPEAKER_00: 3%.
• 8:53
  SPEAKER_00: Microsoft spent$13 billion on Open AI.
• 8:57
  SPEAKER_00: They have the biggest distribution platform and business software, and they cannot get people to use their AI product.
• 9:03
  SPEAKER_00: The reorganization is an admission that the product layer is not working.
• 9:07
  SPEAKER_00: So they're going to go back to building better intelligence at the foundation, which is probably the right call.
• 9:12
  SPEAKER_00: But it tells you how hard the market really is.
• 9:15
  SPEAKER_00: Distribution alone doesn't win.
• 9:17
  SPEAKER_00: The product has to be good so people can't ignore it.
• 9:20
  SPEAKER_00: The more one more before I close.
• 9:23
  SPEAKER_00: Now Mistral, the French AI company, that is Mistral, just launched Forge.
• 9:29
  SPEAKER_00: This is not fine-tuning, this is not prompt engineering.
• 9:32
  SPEAKER_00: This is a full model training on your own proprietary data, on your own servers.
• 9:37
  SPEAKER_00: Zero data exposure to anyone.
• 9:39
  SPEAKER_00: Not even Mistral sees it.
• 9:41
  SPEAKER_00: The early partners tell you exactly who this is for.
• 9:44
  SPEAKER_00: ASML.
• 9:45
  SPEAKER_00: That's the Dutch company that makes the machines that make every advanced chip on Earth, Ericsson.
• 9:52
  SPEAKER_00: The European Space Agency.
• 9:54
  SPEAKER_00: These are organizations that cannot send their data to a cloud provider.
• 9:58
  SPEAKER_00: They can't share it with OpenAI or Anthropic or anyone.
• 10:01
  SPEAKER_00: They need AI trained on their own knowledge, not prompted with it, but trained on it.
• 10:06
  SPEAKER_00: This is a fundamentally different approach, and it's going to be a massive market that most people are not talking about yet.
• 10:12
  SPEAKER_00: So let me bring all this together.
• 10:14
  SPEAKER_00: Three code reds in four months at OpenAI.
• 10:17
  SPEAKER_00: GPT 5.4 Mini and Nano introducing the subagent model where AI systems hire their own AI workers.
• 10:23
  SPEAKER_00: Nvidia building custom chips and putting AI in space.
• 10:27
  SPEAKER_00: India's outsourcing industry is getting disrupted.
• 10:30
  SPEAKER_00: The Pentacon building its own AI.
• 10:33
  SPEAKER_00: Microsoft admitting their product isn't working.
• 10:35
  SPEAKER_00: Mistral opening up model training for companies that cannot share their data.
• 10:40
  SPEAKER_00: What does this mean for you?
• 10:41
  SPEAKER_00: An AI industry just split into two clear lanes.
• 10:45
  SPEAKER_00: Lane one is the consumer lane, chatbots, image generation, video tools, entertainment.
• 10:50
  SPEAKER_00: Lane two is the enterprise lane.
• 10:52
  SPEAKER_00: Coding agents, business automation, workflow systems, real revenue.
• 10:57
  SPEAKER_00: OpenAI spent the last year sprinting in lane one.
• 11:00
  SPEAKER_00: Anthropic quietly dominated lane two.
• 11:03
  SPEAKER_00: And now everybody is scrambling to catch up.
• 11:05
  SPEAKER_00: If you're a business owner, you need to be in lane two, not playing with AI, not experimenting, building systems, deploying agents, automating the work that's eating your time and burning your money.
• 11:16
  SPEAKER_00: Because here's the truth.
• 11:18
  SPEAKER_00: The businesses that figure this out first don't just get an advantage.
• 11:22
  SPEAKER_00: They get an advantage that compounds every day, every week, every month, while their competitors are still trying to figure out what prompt to type into Chat GPT.
• 11:33
  SPEAKER_00: This isn't hype, that is math.
• 11:35
  SPEAKER_00: And I didn't learn that from a textbook.
• 11:37
  SPEAKER_00: I learned it by building it every day in real businesses, real workflows with real results.
• 11:43
  SPEAKER_00: This is AI with honor, and I'll see you tomorrow.