The tech world had its most turbulent week in history starting Nov 17. At the start of the gripping soap opera, Sam Altman, the CEO of OpenAI, the pre-eminent Artificial Intelligence start-up, was summarily fired by the board because he was not consistently candid in his communications.
That led to a massive staff rebellion, with 97% of the 770 staffers publicly choosing to back their CEO, forcing the board’s hands and paving the way for Altman’s reinstatement even though just a day earlier, software giant Microsoft had hired him to run its ambitious new AI Lab and bring all his loyal 747 teammates with him.
OpenAI is now back to where it was a week ago. The board has since been reconstituted, Altman is back in the saddle as CEO and his loyal staffers are back to work with him. Microsoft has abandoned its ambitions to create an in-house AI lab. Yet, AI has fundamentally changed over the past week.
The crisis has changed the overall trajectory of the AI boom, with question marks over not just where OpenAI goes but also how the development of this critical technology will be impacted, as well as who wins the AI race from here. Behind the gripping corporate drama was the fight over AI’s future. Irrespective of what you might have read or heard in the media or on the Internet, Microsoft did not come out smelling roses and was not seen as a winner.
Here is how we got here and what is next for AI. OpenAI has long been seen as one of the handful of firms with the keys to the AI kingdom and all the riches that come with it. ChatGPT, a large language model-based AI chatbot developed by OpenAI and launched exactly a year ago, transformed the tech industry when the global economy was slowing, weighed down by higher-for-longer interest rates and dire need of new growth drivers.
In many ways, ChatGPT was the catalyst the world had been waiting for in the aftermath of post-Covid-19 reopening marred by supply chain bottlenecks, labour shortages and soaring prices. As central bankers rushed to raise interest rates at a record-breaking pace to tame runaway inflation, economies worldwide started to slow, and a synchronised global recession seemed imminent. In the aftermath of the pandemic, the world was in dire straits. It needed a saviour.
See also: Without regulator buy-in, scaling AI in financial services will be an uphill battle
Enter AI and Altman, and a bright new spark was lit. No company in history has ever seen the adoption of a product or service or its users grow as fast as OpenAI with ChatGPT.
AI is a game-changing technology like the digital revolution and the industrial revolution before it. AI is already transforming communications and decision-making, finding new ways to fight diseases, discovering new drugs, driving cars, and, indeed, even buying. Not since the heady days of the dotcom boom in the late 1990s or the emergence of smartphones in 2007 have we seen anything as transformative.
The face of the AI boom
That brings me to the OpenAI CEO and why he matters so much. Altman, 38, has been the face of the AI boom since the launch of ChatGPT a year ago. A computer science graduate of Stanford University, the software engineer, start-up founder and venture capitalist is seen as a charismatic visionary who has passionately articulated his vision for AI, launching start-ups in the AI space, tirelessly raising money for those ventures and has successfully made the case the AI can be done safely and OpenAI is one of the entities that can do it safely.
Here’s AI’s back story: In 1950, mathematician and computer scientist Alan Turing, in a seminal paper, Computing Machinery and Intelligence, wondered aloud, “Can Machines Think?” It wasn’t until about 20 years ago when machine learning techniques and deep learning algorithms developed by three Canadian academics — Yoshua Bengio, Yann LeCunn and Geoffrey Hinton, known as the ‘Godfathers of AI’ who jointly won the coveted Turing Award — turned their pioneering work into what we call AI today.
At the forefront of AI was Google’s DeepMind, a sister company of the US search giant that had recruited Hinton and top AI scientists like Mustafa Suleyman. EV pioneer Tesla’s CEO billionaire Elon Musk, who wasn’t comfortable with Google controlling AI, was among the founders of OpenAI.
His idea was to bring top AI scientists to a non-profit organisation funded by billionaires like him and LinkedIn founder Reid Hoffman. The mission was to create safe AGI or Artificial General Intelligence that benefits humanity. Musk wrote a cheque of US$100 million ($133.9 million) and promised another US$900 million more.
Altman, then running the VC firm Y Combinator, was named one of the directors. Three years into the non-profit, Musk told OpenAI executives that it was falling further behind Google and wanted to take over and run the non-profit personally. Tesla, at the time, was struggling to stay afloat. There was resistance to Musk running OpenAI, so he stepped down as director, citing a “conflict of interest”, and Altman stepped in the driver’s seat. Musk poached several of OpenAI’s top talents, including Andrej Karpathy, who now runs Tesla’s self-driving programme. Musk also stopped donating to OpenAI even though he had promised up to US$1 billion.
That left Open AI with no money to fund itself. AI start-ups need a ton of cash to train AI models on supercomputers. To keep OpenAI going, Altman proposed that the non-profit entity should set up a for-profit subsidiary to raise money. Among early investors was Microsoft, whose Azure is a distant second to Amazon’s AWS in cloud computing infrastructure. Microsoft has spent tens of billions on Azure but still doesn’t have even half the revenues of AWS. Microsoft’s deal with OpenAI: We will invest US$1 billion in your company if you use our Azure servers.
After OpenAI unveiled ChatGPT late last year, Microsoft offered an even bigger partnership. It was willing to invest another US$ 12 billion for a 49% stake, but a big chunk of the investment will be barter for OpenAI using Azure servers to train its AI data. Neither Microsoft nor OpenAI has revealed how much of the investment was cash and how much was a voucher for Azure’s computing. Microsoft also agreed to cap its profits from OpenAI at no more than US$100 billion.
However, the dual mandate and complex model of a non-profit entity focused on safety-first AI running a for-profit company came to haunt Altman and OpenAI. It was even more problematic for Microsoft. The software behemoth, which faced an anti-trust suit in 1998 for trying to monopolise the PC market, has been wary of being seen as exercising too much control.
Silicon Valley soap opera
OpenAI’s non-profit model, its 49% stake in the for-profit subsidiary and its vow to cap total profits allowed Microsoft to boast that it did not control the company at the apex of the AI revolution. Yet, in its eagerness to protect itself from embarrassing anti-trust lawyers, Microsoft was left vulnerable when the board abruptly moved against Altman.
Sink your teeth into in-depth insights from our contributors, and dive into financial and economic trends
Altman’s firing left Microsoft investors aghast. How could the world’s second-largest company worth US$2.8 trillion throw US$13 billion on an AI start-up but have no say on what goes on in there and hear about major development just a minute or so before the rest of the world?
Microsoft offered to hire Altman and 747 of 770 staffers ready to flee OpenAI to save face. It turned out that was merely a negotiating tactic with an unfriendly board. In the end, the software firm is better off with Altman running OpenAI rather than working in-house at Microsoft because if he and loyal staffers all came in, there would be nobody left to run OpenAI, which it had poured in US$13 billion, albeit a lot of it as barter for the use of its AI servers.
In the wake of last week’s events, many OpenAI customers openly question how they work with the company. They know that the old OpenAI board is gone and Altman is back as CEO, but the previous weekend shook their world up.
The tumultuous soap opera also forced many AI start-ups to insulate themselves from growing risks and become more model agnostic rather than be heavily reliant on OpenAI’s GPT-4, a large multimodal model that is better at tasks that require more creativity or advanced reasoning, accepts both text and image inputs and churns out human-like text.
Over 18,000 Microsoft customers are building on top of GPT-4 within Microsoft’s Azure. Nobody, not even the start-ups building on the world’s most sought-after technology, wants to be as vulnerable as they were ever again.
The aftermath
Some analysts I have spoken to over the past few days say the weekend’s drama will likely strengthen the hands of those pushing open-source foundation models for AI like Meta Platforms, the parent of Facebook and Instagram. Others say OpenAI’s competitors like Anthropic, Cohere and Google’s Gemini will likely benefit. Users of AI now realise they should not put all their eggs in one basket. That will cut OpenAI’s market share and hurt its main backer, Microsoft.
If there is so much money in AI, why don’t more tech companies flushed with cash build an OpenAI-like platform? Building large language AI models is not only hard but expensive as well. It would take someone three to five years at a minimum and up to US$5 billion to build something like OpenAI today. That is if they can find the talent.
You need 500 software engineers, costing an average of US$200,000 to US$250,000 annually. Add in bonuses, stock options and perks like free food, massages and free laundry that they might demand just for the privilege of working for you. Talent is so scarce that if three or four companies were recruiting more AI people, wages would be 20 to 30% higher. Let’s say you have the money, the talent, and the time to do it all in four or five years. You still might find that incumbents like OpenAI are at least five years ahead by the time you are done building the next AI platform.
For now, the biggest concern is AI safety, which was the core issue that OpenAI’s old board used to oust Altman in the first place. Here is the way I look at it. Though AI has made huge leaps in recent years, we are still many years away from a breakthrough in Artificial General Intelligence, the technology that will allow machines to accomplish most intellectual tasks that humans can perform.
Right now, there is too much doomerism with people screaming that AI development needs to slow down immediately and that we need to pay more attention to safety. The reality is that there is a very low probability of existential risk of AI becoming a potential threat to humanity. Generative AI has helped productivity gains across some sectors. Coming out of the current global slowdown, we will need more productivity gains rather than a broad brush slowing of AI research and development in the name of safety.
Assif Shameen is a technology and business writer based in North America
