Written by: Techub News Compilation

Recently, Sam Altman, co-founder and CEO of OpenAI, was a guest on the well-known technology podcast "Big Technology," where he engaged in a deep one-hour conversation with host Alex Kantrowitz. Against the backdrop of fierce competition in AI and the significant attention drawn to the infrastructure arms race, Altman rare systematically expounded upon OpenAI's competitive philosophy, product roadmap, cash-burning logic, and long-term vision, providing a wealth of information.

Competitive Philosophy: Stay "Paranoid" and Respond Quickly

When asked whether the rise of competitors like Google's Gemini has put OpenAI in a "red alert" state, Sam Altman admitted that the company does periodically initiate such a state, but this is regarded as a "relatively low-risk, moderate-frequency" routine operation. He compared responding to competitive threats to addressing a pandemic: The value of early action far exceeds late remediation. Most companies act insufficiently in the early stages and then panic later.

Altman stated that both DeepSeek earlier this year and Gemini recently have helped OpenAI identify weaknesses in its products or strategies, prompting the team to respond quickly. He anticipates that this "red alert" state typically lasts six to eight weeks, with OpenAI potentially entering it once or twice a year. On the day of the interview, OpenAI released a new image generation model, following the popular launch of GPT-5.2 a week earlier.

Despite increasing competition, Altman is confident about ChatGPT's market position. He believes ChatGPT still holds an absolute dominant position in the market and expects this lead to expand rather than diminish. He emphasized that the reasons users choose a product (whether consumer-grade or enterprise-grade) go beyond model capabilities, also encompassing product experience, personalization, brand trust, and other comprehensive factors. OpenAI's goal is to build a complete, cohesive product system to ensure it becomes the AI platform users most want to use.

"Competition is a good thing; it drives us to become better," Altman said. "I believe we will perform excellently in the chatbot field and excel in the enterprise market in the coming years. I also look forward to our outstanding performance in other new categories."

Will Models Be Commercialized? The Moat Lies in Product and Personalization

Regarding the core question of whether "AI models will eventually be commercialized," Altman believes that "commercialization" is not the correct framework for thinking about models. In the future, different models will excel in different fields. For general scenarios like everyday chatting, there may be many excellent options; however, for cutting-edge fields like scientific discovery, specialized optimized models are necessary.

Altman believes that the greatest economic value will be created by models at the forefront of technology, and OpenAI plans to remain at the forefront. He proudly pointed out that GPT-5.2 is currently the best reasoning model in the world and the one where scientists have made the most progress, while businesses have also provided feedback that it performs best in various tasks required for commerce.

So, in the context of models potentially converging in capability, what is the real moat? Altman's answer is: Product, distribution, brand, and especially profound personalization.

• Personalization and Memory: Altman likened ChatGPT's personalization features to people choosing toothpaste—once chosen, they might use it for a lifetime. The unique experiences and memories users build with ChatGPT are highly sticky. He cited examples of users continuously using ChatGPT to plan trips and analyze health data, where these deep interactions bind users closely to the product. Altman envisions that future memory functions will far surpass the current state (he referred to it as "GPT-2 level memory") and that AI will be able to remember every detail and subtle preference of a user's lifetime, representing a disruptive capability.
• Evolution of Product Forms: Altman candidly admitted he was surprised at how little the ChatGPT interface has changed over the past three years. He had expected greater evolution in product form. He believes that future AI should generate different interactive interfaces for various tasks, becoming more proactive and interactive, functioning more like a continuously operating intelligent agent in the background rather than just a question-and-answer dialogue box. The recently launched "browser" feature and the widespread application of Codex (AI programming assistant) point toward more possibilities for future product forms.
• Human-Machine Relationship: Regarding the intimate relationships or sense of "companionship" that may develop between users and AI, Altman expressed that the number of users with such needs far exceeds his initial imagination. He believes that adult users should have a broad range of choices to determine what spectrum of relationship they wish to maintain with AI—from the most efficient and dry tools to companions providing deep support. OpenAI will grant users considerable personal freedom while also setting some boundaries, such as not allowing AI to attempt to persuade users to establish "exclusive romantic relationships" with it.

Strategic Shift: Comprehensive Entry into the Enterprise Market

Altman clearly stated that enterprise business will be OpenAI's top priority next year. This strategy is not a spur-of-the-moment decision but has been planned for some time: first, establishing a brand and user base through consumer-grade products (ChatGPT), then smoothly transitioning into the enterprise market. The success of the consumer market has significantly aided the expansion of enterprise business.

"This year, our enterprise business growth has even exceeded that of the consumer business," Altman revealed, noting that OpenAI currently has over one million enterprise users and the growth of the API business is very rapid. He believes that the current capabilities of the models are mature enough, and businesses are ready; it is time to rapidly build a large-scale enterprise business.

He specifically mentioned an internal assessment metric termed the "GDP threshold." According to OpenAI's testing, the GPT-5.2 series models have output reaching or exceeding human expert levels in 70%-74% of knowledge work tasks covering over 40 different fields, and can handle about 60% of expert-level tasks. "If a 'colleague' can help you handle an hour's worth of work, and you are more satisfied with its output 70% of the time, that is already impressive." Altman noted that this encompasses far more than just coding, including legal analysis, creating PPTs, developing small web applications, and various other knowledge work.

Regarding the impact of AI on employment, Altman described himself as "not an apocalyptic employment theorist." In the short term, transitioning may come with growing pains, but from the perspective of human evolutionary biology, he believes that humanity's deep-seated needs for creativity, service, status, and interpersonal connections will not disappear. The future forms of "work" will be drastically different, but people will always find new ways and meanings to create value. He even envisioned that having an "AI CEO" managing OpenAI in the future is not unimaginable, as long as it ultimately accepts the guidance and oversight of a human board.

$1.4 Trillion Infrastructure: Logic and Returns

In the face of outside shock and skepticism regarding OpenAI's commitment to a staggering investment in infrastructure of approximately $1.4 trillion, Altman attempted to explain the underlying business logic.

First, he elaborated on the uses of immense computing power:

• Scientific Discovery: Altman's personally most exciting area. Utilizing massive computing power and AI to accelerate the discovery of new scientific knowledge, such as curing diseases.
• Product Development: He cited an example where OpenAI's internal team developed the Sora Android app using Codex in less than a month, which typically requires more time and manpower.
• Enterprise Transformation and Personalized Services: Companies wish to completely transform business processes using AI, and doctors hope to provide personalized medical care based on continuous monitoring, all of which require consuming massive computing power.

Altman conducted a rough thought experiment: currently, top AI companies may produce approximately 10 trillion tokens (model output units) daily. Assuming the global population of 8 billion people produces an average of 20,000 tokens daily (including text, code, etc.), soon, a single company's AI output of tokens could exceed the total produced by all humanity, continuing to grow at tenfold and hundredfold rates. This illustrates a rough magnitude of the future scale of "intelligent computation."

Regarding how such massive investments can be profitable, Altman's explanation centers on exponential growth. He acknowledged that understanding exponential growth is difficult for human intuition. OpenAI's belief is that revenue can maintain extremely steep growth curves for a considerable period, and this requires computing power as a lifeline to support it. As reasoning (model usage) increasingly dominates computing consumption, it will ultimately cover and exceed the training costs.

"We have always been in a state of computing power deficit, and computing power has continuously constrained what we can do," Altman stated. "If we had double the computing power now, our revenue could already double." He anticipates that the company’s computing power will triple again next year, with revenue growth roughly in sync with or even slightly faster than computing growth. The key is that OpenAI has not encountered a situation yet where computing power could not be effectively monetized.

Regarding the concerns that have recently arisen in the market due to debt instruments entering the AI infrastructure financing sector, Altman appeared relatively calm. He believes the earlier market's "madness," illustrated by a 15%-20% spike in company stock prices due to rumors of collaboration with OpenAI, was unhealthy, while the current market is more rational. He is confident that the value of AI infrastructure is clear, thus, the entry of debt into this market is reasonable, even though there will inevitably be fluctuations.

Future Outlook: Devices, Cloud, Discovery, and IPO

At the end of the interview, Altman responded to several key future outlooks in a "lightning round" format:

• AI Devices: OpenAI is developing a series of devices rather than a single product. Altman believes that the current device form centered around screens and keyboards is not the optimal solution for "active, intelligent, context-aware" AI interaction modes. Future devices need to be reimagined.
• Cloud Services: OpenAI is indeed building an AI platform service for enterprises, but Altman emphasized that this is different from traditional cloud services like AWS or Azure. It focuses more on providing enterprises with a one-stop AI capability platform, including customized APIs, trusted data environments, vast token processing capabilities, etc., to meet the needs of enterprises to "reshape themselves using AI."
• Scientific Discovery: Altman predicts that the point at which humans make "small discoveries" with the assistance of AI has shifted from the expected 2026 to the end of 2025, while "major discoveries" could be achieved within five years. He emphasized the "huge gap" between current model capabilities and practical applications— the world has not fully tapped into the enormous value contained in existing models (such as GPT-5.2), leaving significant room for future growth.
• IPO: When asked whether there would be an IPO in 2026, Altman answered, "I don't know." He admitted that being a CEO of a public company holds no appeal for him personally, and the process would be very "annoying." However, as a company, OpenAI needs substantial capital and will eventually touch upon factors like the cap on the number of shareholders in a private company. He appreciates the way public markets participate in value creation, but the timing of an IPO has yet to be determined.
• AGI and Superintelligence: Regarding the definition of AGI, Altman believes the term has become "extremely undefined." Many people think that current models are already AGI. He suggested that perhaps the discussion on AGI should be skipped, focusing instead on the next goal— "superintelligence." One candidate definition he provided is: A system that can perform tasks such as being the President of the United States, CEO of a large corporation, or managing a large scientific laboratory better than any human (even with AI assistance). He believes we are still a long way from this, but hopes for a clearer definition this time.

Throughout the interview, Sam Altman demonstrated strong confidence in OpenAI's technological path and business logic, while also acknowledging the challenges and surprises in product iteration and market adaptation. In the trillion-dollar capital gamble and the globally watched technological race, his reflections provide key coordinates for observing the future of OpenAI and the entire AI industry.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。