Artificial intelligence has thoroughly permeated business and politics, but it is also beginning to create divisions between companies and investors. That was the core message delivered by executives attending this week's WTF Summit hosted by The Information, which brought together senior women from the technology and finance industries.
A global co-head of investment banking said investors are becoming increasingly proactive in sorting out the winners and losers within the AI space, favoring companies that can independently train models and control the underlying chips and data centers. "This is not a shift in market sentiment toward AI," she said. "What is happening is a repricing and growing divergence within the AI trade itself."
At the same time, major technology giants have begun jockeying around the key policies that will determine the direction of this technology, with chip export controls being one example. Google's global head of data center energy, Amanda Peterson Corio, warned that fast-growing emerging cloud providers are forcing the industry to introduce new standards and regulations governing data center resource usage across the board. Speakers at the summit came from the above companies, the NYSE Group, Goldman Sachs, as well as a senior figure from CBS News and the co-founder of Bumble.
Anthropic policy chief: hoping a compromise on chip export controls can be reached
Anthropic's head of public affairs, Sarah Heck, said the AI lab has been lobbying the U.S. government to include chip export control provisions in the annual National Defense Authorization Act, which represents the best window for passing such rules before the end of the year; but she also stressed that whether the bill ultimately passes is "not guaranteed." The developer of Claude has been calling for tighter chip export controls and warned that without such controls, other governments could use American technology to develop AI. Heck said Anthropic has been communicating with Republicans, Democrats, and the White House to secure text acceptable to all parties in the NDAA. Congress is still deliberating whether to include three bills restricting the sale of chips and chipmaking equipment to China. These bills have also created divisions within the tech industry. In September, Anthropic withdrew from a tech industry organization whose members include Nvidia and OpenAI, after that group sent a letter to Congress opposing the inclusion of chip export control provisions in the NDAA. The industry group argued that some controls would weigh on U.S. AI companies and push some customers toward China, where regulation is relatively looser. Heck added about the bill: "Some room for compromise has already emerged, but if a final version passes, the only vehicle will be the NDAA." Heck disagreed with the claim that China could catch up to or even surpass U.S. AI leadership without using American chips. "If you accept that China is six to 12 months behind us, then their pace of chip self-development is not enough to close that gap." Heck acknowledged that in the future Anthropic and other U.S. AI companies may need to cooperate with China on AI safety, but that cooperation remains feasible even with chip export restrictions. "The two things can proceed in parallel; we can absolutely do both."
NYSE president: AI is accelerating changes in trading models
Lynn Martin said AI is yet another technology fundamentally changing the shape of trading. She recalled that when she first entered the industry, a trader's physical attributes could directly affect efficiency on the trading floor, a completely different era from today. "A man who was six-foot-eight could shout orders far louder than a woman who was five-foot-four, and at the time that really mattered." After the rise of electronic trading, that advantage disappeared, and market liquidity and efficiency improved as a result. AI is now accelerating this process further. Martin noted that the NYSE processes 1.2 trillion to 1.3 trillion order messages every day. She predicted that AI will cause the number of messages exchanges need to handle to surge dramatically. Although technology can handle greater volume, she believes the surge in messages will actually make human intervention more important. "If models are trading with models, that is not an ideal scenario; or if there is a flaw in the model code and the model itself lacks judgment and the ability to adapt autonomously, that is harmful to the market," Martin said. "In my view, AI combined with human intervention is the optimal market structure." Martin also mentioned that Anthropic's IPO has been hanging over the entire IPO market, weighing on other companies preparing to list.
Google data center energy chief: "My timeline is 2031"
Google's global head of data center energy, Amanda Peterson Corio, said the production iteration cycle for AI server chips is far shorter than the construction cycle for the supporting power infrastructure. This means she now has to plan for data centers that will not be built and come online until 2031 or even later. Grid-connected power construction takes five to 10 years, but Google's tensor processing units, its self-developed AI chips, are updated roughly once a year. "Power construction takes a seven-year cycle, while chips iterate annually, so the two are inherently mismatched." To address this mismatch and avoid the risk of chip oversupply, Corio predicted that power solutions such as gas generators and on-site clean energy for data centers will increase, while the market will step up upfront capital investment to build new grids and transmission facilities. She had just returned from a visit to India to study pumped-storage hydropower options. Local community demands are also a key variable determining whether data centers can be built and powered on time. In response to widespread external concerns, she pushed back on claims that data centers consume too much water, create too few jobs, and pay too little tax. She also noted that many newcomers are rushing data center projects into development without Google's mature experience, making industry-wide regulation critical. "We strongly support unified industry regulation."
Goldman Sachs co-head of investment banking: a "repricing divergence" is taking place within the AI boom
Goldman Sachs global co-head of investment banking Kim Posnett said that as the cost of capital continues to rise, weaker segments are already emerging within the broad AI market. She is very bullish on AI overall, but investor screening standards are becoming stricter. Posnett pointed out that if interest rates remain high for an extended period, many AI projects will fail to meet expected returns. "AI itself is not a bubble, but localized bubbles can breed within AI." She added that investors are beginning to favor vertically integrated companies: those that develop frontier models in-house, deploy industry applications, and also control AI chips. This week, Goldman Sachs was deeply involved in underwriting major deals for several leading AI companies, including financing and IPO-related matters for Anthropic and Meta. Asked about circular financing issues in the AI ecosystem, Posnett said she focuses more on the demand side, but acknowledged that the model in which tech giants like Nvidia invest in upstream and downstream suppliers and customers carries inherent risks. "They want to build closer partnerships, which is logical; but it also amplifies risk. Both things are true."
Flock co-founder: whether police hand footage to ICE is not up to us
Paige Todd, co-founder and head of people at license plate recognition startup Flock Safety, said the hardest decision in running the company was determining early on not to restrict law enforcement agencies from using its technology within legal bounds. "I have personal views on many laws, just like anyone else. But we decided it should not be up to us to decide how laws are carried out with our technology. If cooperating with U.S. Immigration and Customs Enforcement is legal in a given jurisdiction, that is their choice." The nine-year-old, venture-backed company focuses on license plate recognition technology used by local law enforcement to investigate traffic violations and track suspects. Against the backdrop of the Trump administration dispatching more federal agents nationwide to crack down on illegal immigration, local residents worry these cameras could assist in forced deportations. Todd said the company has not signed a contract with ICE, but "local law enforcement agencies can choose on their own to share footage with immigration authorities." "That is not up to us. At the end of the day, ownership of the footage is not in our hands." She also said that despite protests targeting Flock cameras, the atmosphere inside the company has remained relatively stable. In September, the company offered voluntary severance packages to U.S. employees. The reason: "When employees are having dinner with family or out with friends on weekends, they get asked, 'You actually work at that company?' Over time, no matter how much employees believe in what the company does, that pressure causes harm."
Arm and Nscale executives: there is no chip glut
A growing number of data center developers, analysts, and investors worry that delays in power delivery, political resistance, and technical problems will leave large quantities of specialized AI server chips idle in the future, with no usable data centers to pair with them. Arm Chief Marketing Officer Ami Badani said she has not heard of any Arm-based chips sitting idle in warehouses due to a lack of power. Nidhi Chappell, AI infrastructure president at data center and cloud provider Nscale, said Nscale is also continuously monitoring the supply chain to avoid mismatches between power supply and chip supply and demand. In recent months, the market has worried that a shortage of data centers with power could mean delivered chips cannot be put into use, prompting Nvidia to address data center power bottlenecks. Badani also said customer demand for custom application-specific chips is now at a record high. Arm provides chip design support for Nvidia, Apple, and major cloud providers. Custom chips require higher upfront investment, but companies gain greater control. Chappell and Badani noted that changes in consumer AI usage habits are also reshaping chip demand. One trend is deep integration of various hardware components inside data centers, with different chips handling different tasks to optimize resource utilization; chips themselves are also becoming more narrowly specialized, for example dedicated to model training or dedicated to inference.
23andMe founder Wojcicki: AI can uncover new medical discoveries
Anne Wojcicki, head of 23andMe Research Institute, said that leveraging the nonprofit organization's vast DNA database, AI can uncover entirely new medical clues and offer practical health improvement recommendations for ordinary people. "This is the power of scale." Wojcicki reviewed 23andMe's entrepreneurial journey: the company was once a pioneer in direct-to-consumer genetic testing, but it never found a sustainable business model and ultimately filed for bankruptcy in 2025. Last year, she spent more than $300 million of her own money to buy back the company and transform it into a nonprofit research institution. At the summit, she cited a recent scientific collaboration that demonstrated the database's research potential: it contains DNA samples from more than 3.3 million participants who volunteered for research. Researchers at Boston's Dana-Farber Cancer Institute and the 23andMe Research Institute jointly discovered a rare genetic mutation that puts carriers at extremely high risk of lung cancer. People who carry the variant but do not smoke have 60 times the lung cancer risk of the general population. Wojcicki's sister, former YouTube CEO Susan, died of lung cancer in 2024. "I pay very close attention to this area," she said. She met the Dana-Farber researchers at an academic conference, where they were studying the mutation but had only 20 samples and found it difficult to recruit more participants. Wojcicki told them: "I have 500 participants carrying this genetic variant." Now 23andMe can contact these people and recommend screening to detect cancer early. "I may be able to save these lives." The 23andMe Research Institute is developing a feature called "Health Summary" that uses AI to uncover connections between users' DNA and other health data and generate health recommendations. "The more data we have, the more medical insights we can provide."
Comments