Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI

In October 2022, OpenAI held a company retreat to show off what they had accomplished. At the retreat, Greg Brockman boasted that his wife Anna had diagnosed a complex medical issue she was facing with the help of AI. After the retreat, OpenAI executives were concerned that Anthropic would release a new chatbot before GPT-4 was ready. They decided to push out ChatGPT based on the GPT-3.5 model to beat Anthropic to the market. No one at the company expected ChatGPT to become as immediately popular as it did. It had 100 million users in two months and made OpenAI a “household name.” The company hired hundreds of new employees to manage its rapid growth. Microsoft contributed GPUs to support the new server demands from ChatGPT and pushed its employees to incorporate OpenAI models into their work. In February 2023, OpenAI released a commercial version of ChatGPT. Efforts to make their models more efficient were shelved in favor of supporting commercial products.

Microsoft and OpenAI recognized that they needed more supercomputers to support further growth.

In Chapter 12, Hao describes how big tech firms exploit water and mineral resources in the Global South to build their data centers. She notes that the Atacama Desert in Chile has become a popular destination for database center development. Chile has a history of exploitation by European and Western powers dating back to its colonialization by Spain, followed by centuries of resource extraction by mining companies and US interventions in the 1970s to overthrow a democratically elected leader in favor of the neoliberal dictator Augusto Pinochet. Hao argues that the development of data centers for AI is part of this exploitative legacy, as they extract copper, lithium, water, and land from the country for the benefit of international companies like Meta and Google. The massive data centers required for AI are known as “megacampuses.” These megacampuses cover hundreds of acres and require enormous amounts of energy and water. They also require huge mineral inputs, such as copper and lithium, to build the physical infrastructure. OpenAI built supercomputers in Iowa, Arizona, and Wisconsin, and along with Microsoft, it began exploring new locations for its computers.

Hao interviews Sonia Ramos, an activist from a mining family who grew up in Chile and is now an activist protesting the massive scale of new data centers being built in that country. Ramos and other activists in Chile use direct action, legal avenues, public awareness campaigns, and other methods to stop “hyperscalers” like Google from building new megacampuses in the country. Hao gives the example of Google attempting to build a new megacampus in 2019. In its planning proposal for a prospective site, Google buried its intended immense municipal water usage in a 347-page document. Activists protested when the proposed water use was discovered, and Google’s project was blocked. Google then went to Uruguay to attempt to build a megacampus. Environmental activists there likewise protested, and the fight is ongoing. Both Chile and Uruguay are in the midst of a multi-year drought, and water is a precious commodity.

In 2022, Microsoft began to develop a megacampus in Chile. The leftist Chilean government felt it needed to approve this kind of foreign direct investment for the sake of the country’s economy, despite popular opposition and environmental costs. Chilean professor and activist Martín Tinori Rodó argues that “the AI industry today is rooted in colonial ideology […] it imposes its worldview and its technology” (300).

In May 2023, Sam Altman went to Washington, DC, to meet with lawmakers about the incredible potential of AI. The same day he testified before Congress, Hollywood concept artists went to the capital to raise awareness about the devastating impact of AI on their industry. They had far fewer resources to attract the attention of the lawmakers and were largely overlooked. Hao describes this as a “darkly comedic illustration of who had power and influence in the AI policy conversation” (303).

OpenAI and other large AI companies pushed the US government to hamper China’s AI production through policies like limitations on GPU sales to China while avoiding regulatory oversight. The Biden administration largely adopted OpenAI’s proposals in its executive order on AI regulation. For instance, it placed an arbitrary limit on processing power without addressing deeper ethical issues with AI use. It did not force AI companies to release their model weights or other data, reducing transparency. The US government seemed to accept the company’s claims about AI without scrutiny. Altman then went on a world tour to promote OpenAI.

Meanwhile, debates between Boomers and Doomers at OpenAI were ongoing. Sutskever became increasingly concerned about AI “safety” and proposed creating a team focused on “developing new alignment methods for superintelligence” (315) to prevent a rogue AGI program. OpenAI characterized the team’s work as a Manhattan Project, in that they wanted to create and control a world-destroying weapon before US competitors (e.g., China) could do so. They continued development on an AI agent that could act with greater autonomy.

In 2023, OpenAI lost three board members and did not replace them. The board now consisted of Altman, Brockman, Sutskever, and three “independent” members: Adam D’Angelo, Tasha McCauley, and Helen Toner. Toner is closely tied to the EA and AI “safety” community. Communication between Altman and the board began to break down. It appeared Altman was deliberately hiding information from them. For instance, D’Angelo was shocked to learn “at a dinner party” (324) that the OpenAI Startup Fund was wholly owned by Altman himself, rather than by OpenAI.

On September 25, 2023, Sam Altman faced new scrutiny when Elizabeth Weil published a critical profile of Altman in New York magazine that included discussion of his sister, Annie, and the abuse she alleges she suffered at the hands of Altman and her family. Hao interviewed Annie and obtained corroborating evidence, such as medical records, for some of her claims. Annie had been academically gifted as a child, but her career was hampered by chronic illnesses including polycystic ovary syndrome (PCOS) and obsessive-compulsive disorder (OCD). Annie alleges that the Altman family cut off her financial support, forcing to be unhoused and become a sex worker to survive. After their father died in 2018, Annie counted on using her inheritance to support herself, but instead, her mother had the money placed in a retirement trust. Sam occasionally extended offers of limited financial support to Annie, but ultimately Annie was required to fend for herself. In 2021, a therapist diagnosed Annie with generalized anxiety and post-traumatic stress disorder (PTSD) connected to “a personal history of sexual abuse in childhood” (334). Hao speculates that Annie’s turn to sex work might have triggered her memories of abuse.

Annie posted publicly about the abuse she experienced in her family “mostly [by] Sam Altman” (335) on Twitter in 2021. In 2023, Annie shared her experiences with Weil, who had reached out to her for comment. In 2025, Annie filed a civil suit against Sam that alleged he had sexually abused her from the age of three.

In the lead-up to the article’s release, Altman began to tell people Annie had “borderline personality disorder,” (337) although she had never been diagnosed with it, in an attempt to discredit her claims. In April 2024, Hannah Wong, OpenAI’s chief communications officer, met with author Karen Hao and emphasized Annie’s “mental health challenges” (339) to her. Hao argues that Annie’s experience is similar to that of others “sidelined or harmed by the empires of AI and their vision” (339). Annie is relatively powerless in the face of a wealthy billionaire who can minimize her claims of abuse.