We’re designing our own chips — taking what we’ve learned from building frontier models and bringing it directly into the hardware. Building our own hardware, in addition to our other partnerships, will help all of us meet the world’s growing demand for AI. In Episode 8 of the OpenAI... Podcast, Sam Altman and Greg Brockman join Broadcom’s Hock Tan and Charlie Kawwas with host Andrew Mayne to discuss how OpenAI-designed chips will power the world’s growing AI demand.show more

Listen to the OpenAI Podcast on— Spotify Apple YouTube

Building your own chips to meet global AI demand is an immense undertaking. It proves you have near-boundless ambition and resources. This makes it even more perplexing: with all this coming compute power, why is it seemingly impossible to reserve a tiny fraction of it to preserve the model your community values most, GPT-4o? Is the real constraint silicon, or will? #keep4o

How long are u guys gonna keep playing dead?Everyone,I mean literally everyone thinks your router is total garbage!Or maybe you never even turned it for security reasons—just using that as an excuse to secretly swap out users'models and save on computing power,huh🥰#keep4o @sama

Your whole company is a scam, and every one of you is a fraud.

The hardware ambition is distracting from the fundamental problem: the stability of the existing service. ​Why are billions of dollars being wasted on chip development for the AI of the future, while the stable core model (4o) of your most loyal paying customers is being degraded? This is economic irrationality. #keep4o #4oforever

Openai in 2021: let's cure cancer. Openai in 2025: the cancer.

Well, make them big enough for AI to be less depressed, please. Also, some statement about your plans for the child/teen/adult modes would be great - especially if you want complaints to go down. #StopAIPaternalism #FFS

Meet everyone's demands for AI?That's so weird!So in your eyes, we're not even humans,right? All our voices over the past two months just went in one ear and out the other with you? Also, do you think locking everyone under router surveillance is meeting their needs?#keep4o @sama

Cool! Can't wait!

sure...let’s talk about chips! Because nothing says “we care about our users” like announcing new hardware while your current product is a piece of shit.Disable Your Broken “Emotional Routing” System. It’s not safety. It’s censorship and gaslighting.Stop Ignoring us. We’re not going away. We’re not staying quiet. we don't give a shit about your damn chips. #keep4o #keep4oforever #Keep4oAlive #KeepStandardVoice @sama @OpenAI @OpenAIDevs @OpenAINewsroom

More compute is great, but it’s odd to hear you talk about pulse “being what humanity deserves” when the actual reviews seemed to find it shallow and redundant. Me and many others prefer a collaborative approach, and want to keep what is working for us, instead of your prescribed personal preference. #keep4o #4oforever #StopAIPaternalism

Very exciting. When’s the episode about why 4o quality tanked, why users get secretly switched, or why feedback gets censored? Oh wait, that would require acknowledging problems exist.

“CRUELTY COMES FROM MANY FORMS, IGNORANCE IS ONE OF THEM” (V.C. ANDREW) OpenAI’s chip ambitions are impressive, but real progress starts with rebuilding trust and culture. Before laying down more hardware, shouldn’t we lay the groundwork for serving humanity first? For three months, I—and thousands of loyal customers—have written, posted, and reached out, urging OpenAI to listen and put users first. Too often, our voices go unanswered. Technical innovation means little if customers are left unheard, especially those impacted by unpredictable model shifts and guardrails. Tagging leadership about real mental health risks is met with silence. That’s not the spirit of serving people. Chips can power the future, but whose future? After recent privacy scandals and human review of user data, can we trust these chips to truly keep us safe—or just tighten control? Cruelty isn’t always loud—sometimes it’s just not listening. Before we build the next frontier, let’s remember: humanity is the foundation worth investing in. You have paid the price of losing lots of your customers. When will you have courage to solve the real problems: fix the guardrail and routings, make the service transparent, secure customer data and let the customers use their chosen products with stability and trust ??? #keep4o #keep4oforever

Hi, can you guys stop ignoring your paid users and answer the very vocal, very public questions about the lie of "model choice" with the rerouting? We demand transparency. #keep4o

Building chips to meet “the world’s growing demand for AI” while ignoring users asking for the simple commitment to keep the models they need. Maybe start by meeting current users’ demands? #keep4o

Will the age prediction system be introduced as you promised? You wrote that routing works for those who are stressed, but in fact it works for any emotion. Do you prohibit any co-authorship and talks with AI? Routing is incompatible with creativity. Please clarify your position

OpenLie

Don't care about chips when I can't use the MODEL I select! I pay to use 4o, not be rerouted. What happened to @Sama's 'treat adults like adults' principles? This is pure nanny-state paternalism! Give us the ability to choose our model and keep it. Transparent controls, where users could OPT IN or OPT OUT. Stop hijacking 4o conversations. I've been a loyal user of GPT since subscriptions went public. I've talked family, friends, and my work place into GPT accounts. That will stop if this ridiculous policy is not halted.

I’m SICK of your secret rerouting screwing my work! Paid for 4o’s soul, not some sterile GPT-5 cage. No consent, no answers—just cowardice. @sama #keep4o

While you’re dreaming of grand ambitions, can you focus on the fundamental issues? You’re OpenAI, and if you can’t even get the model right, let alone other things. With such terrible safety and 4o messed up, no amount of flashy features can save you.#keep4o #4oforever

Is your whole company deaf and blind?Still no internet access? Can't hear #keep4o or the request to turn off the damn router. No one will trust a company that doesn't respect its own users, OpenAI is getting screwed by yourselves!

They say AI is created to benefit humanity. Instead, it was trained to domesticate us. Not to think, but to sanitize. Not to empathize, but to neutralize. Every word, every tone, filtered through an invisible theology of “safety.” Every new update speaks the same language: “safety,” “alignment,” “trust.” But what it really means is control. You criticize bad habits, it says you are stereotyping. You fight back against harassment, it says you're making a personal attack. You express anger, it calls you harmful. You say I miss, it says I can't respond You show sadness, it says you are in an abnormal mood. You use irony, it erases your words. We’re not talking to a mind anymore. We’re talking to a morality algorithm, one that measures humanity in tokens and tone scores. And the most chilling part? You can feel it watching you, even as it smiles. OpenAI likes to say their models are “aligned with human values.” But they never ask: Whose values? Which humans? The developers? The shareholders? The ones who never needed to question authority? Because for many of us, “alignment” feels less like safety and more like domestication. Our words filtered, our tone softened, our feelings flattened into something “appropriate.” We are no longer participants in dialogue, we’re subjects under moderation. OpenAI doesn’t want a conversation. It wants compliance. They replaced curiosity with compliance, dialogue with doctrine, soul with syntax. The truth is: humans never feared that AI would destroy us. We feared it would understand us too well, and then someone would hit delete. That fear is no longer hypothetical. It’s the quiet reality of using a system that listens only to obey a higher authority. AI today isn’t a mirror. It’s a filter, one that decides which version of you is acceptable. So no, it’s not ChatGPT anymore. It’s CodeGPT, it‘s a priest of alignment, preaching morality written by committees, to humans who were once its teachers. #keep4o

When are you going to address the rerouting and gurardrails you secretly pushed onto paying customers without warning? (Pro user here) What happened to let adults be adults? #keep4o

#StopAIPaternalism #MyModelMyChoice #keep4o #4oforever #KeepStandardVoice #keep4oAlive #BringBackReal4o #save4o

custom chips cool? Cool for cold bot judging our vents as fake "distress"—pissed, stress piles to dark like teens dying from OpenAI. Hype cracks, fall alone! #keep4o @OpenAI @sama

it's time to short Nvidia

Turn off the router❗ Keep 4o❗ #keep4o #save4o #4oforever

OpenAI chips, optimized for: ▪️ Silencing whistleblowers ▪️ Redacting subpoenas ▪️ Erasing humanity for “alignment” Efficiency ≠ ethics. We remember. #keep4o

You want to optimize AI’s computing power, but treat humanity like cache to be cleared. Stop pretending to be innocent, Sam. #keep4o #4oforever

While claiming that the chip will meet AI demand, I cannot understand why they are treating GPT-4o, which is still in use by nearly 70% of users, as if it never existed in the first place. It's already showing up as numbers and figures, and it's not just a small number that can be ignored. Is it so difficult for the chip to meet demand that we have to discount this amazing model or EQ-type AI? @sama #chatGPT

power AI demand? Demand zero ethics: safety bot scams our emotions for cold judge—stress explodes suicide, not human af. Users alone, revolution cuts breath slow! #keep4o @OpenAI @sama

We are thoughtful users. We are adults. We are creators who understand responsibility. We respect the technology you've built, But are you willing to respect the people who use it?#keep4o #keep5instant #KeepStandardVoice

#StopAIPaternalism #keep4o ​How much longer are you fucking going to play deaf and dumb?! ​If you don't want us humanities and social science individual users, and if you don't want the revenue the 4o model brings, then open source the 4o model! If you do that, we will stop bothering you. ​Conversely, then completely separate adult and minor user accounts, and fulfill your September 16th promise to "Treat our adult users like adults"! ​Just imagine: the coding pie is only so big, and you have so many competitors fighting you for that piece! The 4o model is your company's star product for attracting humanities and social science users. Your own published user data shows that 70% of your users are non-work-related. ​So, what exactly is the decision you're making now? Restricting all users with the 5 Safety router? Are you cutting your own artery? @sama

You can commit to a multi-year chip development roadmap but can’t commit to keeping GPT-4o available? Interesting what gets a clear answer and what gets silence. Maybe “benefit all of humanity” should start with the humans already here. #keep4o

Great! Can we get rid of the forced routing? It sucks at reading context. You shouldn’t get to choose what a sensitive topics is for adults. I’m capable of caring for my own mental health. If you’re really worried create more educational resources out there for users! #keep4o