Behind ChatGPT’s tech revolution: OpenAI’s savvy strategy
ChatGPT won the race to mainstream Generative AI; OpenAI’s clever business strategy was instrumental.
Disclaimer: At time of writing, I’m a current LinkedIn and former Microsoft employee. This article is based purely on public information.
Generative AI has been in the headlines since DALL-E went public in July 2022; ChatGPT’s launch months later turbocharged the hype. The news focuses on product evolution, business battles (e.g., Search), and societal impact. But is the hype warranted, and what choices did OpenAI make to win that race?
Generative AI has a killer use case, sufficient quality, and breadth to live up to the hype – and ChatGPT made that vision a reality. OpenAI played its hand well, overcoming near-intractable challenges with an unlikely partnership, clever business model, and curated roll-out.
ChatGPT made Generative AI into the Next Big Thing
The first time I saw a tech paradigm shift was at a dinner with friends. The second we ordered, everyone pulled out their smartphone and was instantly engrossed – I, sans smartphone, puzzled over the breach in dinnertime etiquette. The iPhone did a new job (information anywhere), was capable enough, and had broad utility. Its impact reverberated through tech and society, forcing everyone to adapt.
Generative AI is shifting the landscape again, and ChatGPT is the catalyst. Shital Shah captured the sentiment well: “It’s like you wake up to the news of first nuclear explosion and you don’t know yet what to think about it but you know world will never be the same again.” I reacted similarly to this TikTok. We can look at the hallmarks of a tech shift, and compare generative AI to prior flops:
Solves a need 10x better than alternatives: ChatGPT generates content quickly and flexibly in a way only a person could before. Blockchain failed since its promise (decentralized data storage) copied databases and its risks (new leverage points, scandals) negated the benefits.
Product is “good enough”: It answers complex and esoteric questions well, though it can hallucinate. VR, on the other hand, can’t immerse the user in a virtual world because it can’t deliver the hardware (optics, compute, tracking) in a light, affordable package.
Broad scope: ChatGPT proved versatile enough to write stories, share recipes, draft letters, and answer search queries. This is as opposed to self-driving cars, once envisioned without steering wheels but now only broadly adopted in controlled environments such as warehouses.
An early indicator of this impact is how it’s upended longstanding opinions on Search: One third of Bing Chat users are new to Bing, and Google is panicking. Journalist Kevin Roose highlights the shift best:
I’m going to do something I thought I’d never do: I’m switching my desktop computer’s default search engine to Bing. And Google, my default source of information for my entire adult life, is going to have to fight to get me back.
One could point to LaMDA (an earlier example) or DALL-E (the first mainstream model). But only ChatGPT was both accessible and broadly applicable enough to drive mainstream opinion writers to action and capture the zeitgeist (see Google Trends). That it came from OpenAI was unlikely.
OpenAI executed well in the face of tough challenges
OpenAI’s challenge was unenviable: build a quality product/GTM, with near-limitless resources, without scaring off customers. The product needs extensive usage data and enterprise know-how. The cost ($700K/day by one estimate) is exorbitant and compute needs are vast. A quality brand is critical to win risk-averse enterprises, but the PR risks are myriad (see Tay). Each problem is difficult but solvable: make it free and easy to use to drive adoption; lock funding and sales to stay solvent. Together, the problems are near-impossible and solutions often conflicting – doubly so for a startup.
The OpenAI-Microsoft partnership solved the toughest problems. Altman knew he needed a big tech partner; following the Bing Chat launch he said “there weren’t that many options for us that could have done the scale not only of the capital we need, but compute and hardware and just general muscle as well.” The most obvious benefit was free Azure compute. But Microsoft was also an anchor customer from which OpenAI could learn enterprise needs via co-development (e.g. GitHub Copilot). Microsoft’s brand (internally “Microsoft runs on trust”) and Azure’s GTM opened access to risk-averse enterprises that adopt new tech later. AWS and Hugging Face’s subsequent partnership validates the move.
The product and timing were critical for user growth. ChatGPT was simple, user-friendly, and targeted at the public – OpenAI got hundreds of millions of beta testers trying every query and use case imaginable to improve quality and product-market fit. The timing – four months after DALL-E – let them build on prior hype to grow users (the drumbeat continues with GPT-4 - four months after ChatGPT). OpenAI took calculated risks to be first, mitigating where possible; this is in contrast to Google’s risk-averse approach that left them playing catch-up. Conservative outcry on “woke AI” was the flipside of avoiding controversy, but better than the alternative.
Its freemium business model served their strategy and kickstarted revenue growth. They started with free use in a “public beta” period to get broad user feedback to refine the model (100M ChatGPT monthly active users two months post-launch!), then later offered nontechnical users a SaaS-style capacity-based premium tier (priority access for ChatGPT and PAYG credits for DALL-E). The real money is from their APIs, conveniently available to companies locked in the new AI arms race.
Where to from here?
Generative AI is now OpenAI’s to lose, and competition is heating up. But the space is now in flux, so prognostications feel like guesswork. I’ll focus instead on the questions I’d ask:
Product: Benedict Evans sums this up well: What can and can’t generative AI solve? What is the tech stack and points of leverage? Can it overcome its precision issues? What tech/ecosystem needs to be in place to make it possible?
OpenAI’s role: Should it focus on the platform or services? Where in the tech stack is it, and where can it evolve? Are LLMs winner-take-most or will there be multiple winners? Can competitors catch up? How do its partnerships help or hinder it? How should they adapt, now that they’re the center of attention?
Other questions: I’ll leave this to subject matter experts, but as a sampler: How will work change, and how do we mitigate negative impact like job displacement? How do we manage the content deluge? What about copyright? Ethics and content moderation?