2 weeks ago
If it wasn’t already clear, Elon Musk and Sam Altman dislike each other.
While nan 2 men were erstwhile cofounders of OpenAI, they’re now locked successful a vicious feud, playing retired successful each its theatrics successful beforehand of a judge and assemblage successful a California courtroom. Musk is suing, alleging that Altman and OpenAI president Greg Brockman tricked him into forming and backing nan statement arsenic a non-profit earlier they subsequently restructured it to person a for-profit entity. OpenAI says Musk was good alert of those plans and frames nan suit arsenic an effort to derail a competitor.
I cognize this communicative each excessively well. I’ve been reporting connected OpenAI since 2019, embedding wrong its agency for 3 days soon aft Musk stepped distant and Altman formally took up nan CEO position. If there’s thing I’ve learned from my years of pursuing this institution and nan AI industry, it’s that this world breeds bitter rivalries.
It’s not a coincidence that astir each of OpenAI’s original founders near nan institution nether acrimonious conditions, nor that each tech billionaire has a mostly identical AI company. The frenetic AI title is inseparable from nan petty, clashing egos of nan unfathomably rich, hellbent connected dominating 1 another.
Indeed, if Musk were to triumph his bid, that could beryllium devastating for OpenAI, particularly arsenic it prepares this twelvemonth for a imaginable first nationalist offering. Musk seeks $150bn successful damages from nan institution and 1 of its apical investors, Microsoft. He besides seeks to return OpenAI to a non-profit, to region Altman and Brockman arsenic leaders of nan for-profit, and to footwear Altman disconnected nan non-profit board.
Yet, to presume that nan early of AI improvement will beryllium wished by a characteristic title misses nan point. Yes, Brockman’s diary entries are revealing, arsenic was erstwhile OpenAI main exertion serviceman Mira Murati’s grounds astir Altman pitting executives against each other, confirming my erstwhile reporting.
But fixating connected questions of whether Altman is untrustworthy, aliases whether Musk is moreover little truthful distracts from a acold deeper problem. If OpenAI mislaid its footing arsenic nan AI manufacture frontrunner, different hardly distinguishable competitor – Musk’s xAI aliases different – would simply switch it. That includes companies for illustration Anthropic, who bask a amended estimation yet prosecute successful galore akin behaviors compromising, for illustration careful decision-making for speed, disregarding intelligence property, aggressively scaling their computing infrastructure to nan detriment of communities.
Nothing astir this proceedings aliases OpenAI’s financial building will alteration nan imperial thrust of these companies to consolidate ever-more information and capital, terraform nan earth, exhaust and displace labor, and embed themselves heavy wrong nan authorities to summation leverage complete its apparatuses of violence. We would still beryllium successful a world successful which a mini fewer person nan profound powerfulness to formed it successful their image and dictate really billions of group live.
As overmuch arsenic Silicon Valley would wish you to judge it, AI does not necessitate imperial conquest, nor could broad-based use from nan exertion ever look from specified a foundation. Before nan manufacture made a difficult pivot into processing extraordinarily resource-intensive AI models, a afloat breadth of different types of AI flourished: small, specialized systems for detecting cancer, for reviving disappearing languages, for forecasting utmost upwind events, for accelerating supplier discovery. So, too, did ideas to create caller AI technologies, including those that didn’t request overmuch information astatine all, and those that required only mobile devices, not immense supercomputers, to train.
Even now pinch ample connection models, an abundance of research and examples specified arsenic DeepSeek already show that different techniques tin nutrient nan aforesaid capabilities pinch a mini fraction of nan standard that AI companies usage to warrant their planet-consuming ambitions.
“Scaling is simply a inexpensive look for getting much performance, but it’s besides a highly imprecise formula.” Sara Hooker, nan erstwhile vice-president of investigation astatine Canadian AI institution Cohere, erstwhile told me. “We emotion it truthful overmuch because it benignant of fits predictable readying cycles. It’s easier to opportunity ‘throw much compute astatine nan problem’ than to creation a caller method.”
But these myriad paths wither successful nan empires’ shadow. In nan first 4th of past year, nearly half of each task money went to conscionable 2 companies: OpenAI and Anthropic. That’s nan extremity of nan iceberg to a yearslong superior consolidation that has hollowed retired academia and starved investigation antagonistic to, aliases simply retired of measurement with, nan firm agenda. From 2004 to 2020, nan percent of AI PhD graduates who chose to subordinate manufacture jumped from 21 to 70%, according to a study by MIT researchers successful Science. And it’s not conscionable nan diverseness successful AI improvement that’s suffering. In 2024, backing for ambiance tech plunged 40% arsenic investors redirected their dollars successful portion to nan brute-force scaling of nan AI empires.
It doesn’t person to beryllium that way. And complete nan past year, arsenic I’ve traveled to dozens of cities astir nan US and globally, I’ve seen this realization dawning. People everyplace are picking up nan mantle of corporate resistance. Most visible and vibrant person been nan information halfway protests popping up successful communities crossed geographies and governmental divides. In New Mexico, I met pinch residents eager to amended themselves astir nan AI manufacture complete potluck, to request transparency and accountability for section projects, specified arsenic a monolithic multi-billion dollar OpenAI supercomputing field being projected successful nan authorities arsenic portion of nan company’s $500bn Stargate computing infrastructure buildout.
At a gathering successful New York, I listened arsenic KeShaun Pearson, a leader successful nan conflict successful Memphis, Tennessee, against Musk’s Colossus supercomputers, gave a heartfelt reminder of nan toll that nan facility’s dozens of methane state turbines were having connected his community. “Take 2 heavy breaths,” he said to nan audience. “That’s a quality right” that was being taken from them. As of this month, Anthropic is using Colossus.
At nan aforesaid event, Kitana Ananda, different organization leader from Tucson, Arizona, mobilizing against Project Blue, an Amazon hyperscale AI facility, described nan deep-seated emotion that she and her chap residents shared: that they fought not conscionable for their ain organization but for each organization being steamrolled by nan AI industry. And connected a 114F day, arsenic they packed into metropolis hallway successful a show of unit and watched nan assembly ballot 7-0 to region nan task successful its existing form, they whooped and cried pinch nan elation that their triumph was each community’s victory.
Workers are besides striking crossed sectors and countries: successful bluish California, much than 2,000 healthcare professionals astatine Kaiser Permanente walked out complete nan threat of AI being utilized to automate their activity aliases degrade diligent outcomes. In Kenya, information workers and contented moderators contracted by AI companies to train and cleanable up their models are organizing to bring world attraction to their exploitation and request amended moving conditions.
In much than 30 countries, taste workers from sound actors to screenwriters to manga illustrators are mobilizing to denounce issues ranging from nan training connected their activity to nan usage of AI systems to rip their likeness aliases switch them, according to the Worker Mobilizations astir AI database, a investigation effort led by nan Creative Labour & Critical Futures group astatine nan University of Toronto.
Educators and students are pressuring their institutions. Victims and their families are suing. Tech employees themselves are campaigning. Group chats for much organizing abound. People are marching.
The upwelling of corporate pushback seems to beryllium forcing nan AI manufacture to downsize its ambitions. Already, much than $150bn worthy of infrastructure projects were blocked aliases stalled successful 2025, according to Data Center Watch, an effort search nan guidance by AI investigation patient 10a Labs. Investors are taking note and opening to discount their projections of really overmuch AI companies tin present connected their promises.
OpenAI shuttered its video-generation app Sora, erstwhile lauded by institution executives arsenic 1 of its astir important products and a caller frontier successful AI development. As the Wall Street Journal reported, Sora’s demise yet stemmed from respective intersecting considerations shaped by grassroots action: flatlining usage, rocky nationalist perception, tightening financials, and dense constraints connected computational resources.
Here’s nan point astir empires. They don’t conscionable activity to devour everything – they dangle connected it for their survival. In different words, nan very point that appears to springiness them paramount spot is their top vulnerability. When moreover a fraction of nan resources they request are withheld, nan giants statesman to stumble. So if you’re wondering what will present existent accountability to nan AI manufacture and a different imagination of nan technology’s development, look beyond nan billionaire mudfight. The existent activity is happening everyplace else.
-
Karen Hao is nan writer of Empire of AI: Dreams and Nightmares successful Sam Altman’s OpenAI
