Last week, the data center industry migrated temporarily to Hawaii for the Pacific Telecommunications Council conference. The event—like Davos except warmer—is typically where deal talks in the data center business begin and the priorities for the year are set.
The sector’s power brokers are coming off a wild 12 months, bookended by the announcement—during last year’s conference—of the $500 billion Stargate data center project, and by the largest transaction in the industry’s history, the $40 billion acquisition of Aligned Data Centers last fall by a group led by BlackRock.
Aligned CEO Andrew Schaap was omnipresent last week, along with senior financiers from Goldman Sachs, Citizens Bank, Apollo and Macquarie. OpenAI arrived with a sizable delegation, including executive Peter Hoeschele and, notably, a large number of lawyers.
Here’s what people were talking about at the event.
2026 Will Be About Execution, Not Announcements
The Stargate plan kicked off a wave of similarly ambitious declarations throughout 2025, with companies pledging hundreds of billions of dollars for AI infrastructure and racing to lock up gigawatts of power.
I think that era is ending. Data center CEOs say that 2026 will be judged less by what companies announce and more by what they actually deliver. This raises stakes for developers that have made big promises.
With deadlines approaching for projects, delays could mean developers start losing their customers, either because they walk away or because competitors come in to salvage stalled projects. I’ve heard that investors, as well as cloud providers, are crafting their data center deals to include favorable “step-in rights,” meaning they could take over projects if their developer is struggling or costs are rising.
The focus on execution this year could mean that top data center executives will be in high demand, similar to the talent wars over AI researchers.
“Some people will be delayed, and the customer will live with it,” one data center CEO told me. “Some people are going to be delayed, and their project is going to fail.”
That’s not to say the optimism is gone. Mai tais flowed freely, dozens of restaurants were booked for private events, and Alicia Keys even performed a small concert.
Google’s TPUs Emerging as a Threat
For years, Nvidia’s graphics processing units were the unquestioned, default choice for large-scale AI data centers. But several executives told me that Google’s tensor processing units have become part of infrastructure planning conversations.
Everyone also senses that Google is aggressively pushing its chip, a trend we’ve written about here and here.
Data center investors in particular like Google’s approach. It is backstopping the debt and leases of certain TPU facilities so that new entrants such as Fluidstack can more easily get financing. A few people told me they think that might put more pressure on Nvidia to do the same thing. (While Nvidia leases back some of its chips from cloud providers and invests in companies that buy its chips, it hasn’t broadly guaranteed data center project debt or facility leases.)
Also new this year was talk of the cloud startups focusing on Google’s TPUs. Last year was dominated by cloud startups such as CoreWeave, Crusoe and Lambda that primarily focus on GPUs from Nvidia.
Local Pushback Is Becoming a Real Risk
Some executives said power availability isn’t their biggest concern anymore. Instead, they’re worried about community resistance to data centers.
That context made OpenAI’s recent announcement about developing “community plans” for its Stargate campuses feel less surprising. The move followed a similar effort by Microsoft and reflects mounting concerns—from water usage to electricity prices—about the impact of data centers.
Industry veteran Hunter Newby, who focuses on expanding internet connectivity, told me he’s largely stopped using the words “data center” when speaking with local communities, given its growing negative connotation. (“Cloud pod,” anyone?)
New Players Hit the Beach
PTC is often seen as a reunion of industry veterans. This year, a wave of startups crashed the party, sometimes with audacious plans.
Power developer American Terawatt emerged from stealth mode, advertising that it would have 600 megawatts of capacity available by the end of this year, though it didn’t say where.
It was also the first time I heard anyone casually invoke the word “terawatt”—a unit 1,000 times larger than a gigawatt—in a data center context, which is an apt illustration of where the industry’s ambitions have gone.
Who’s Looking for Data Center Capacity
Multiple attendees told me they came to Hawaii hoping to secure capacity for near-term projects, only to leave empty-handed. Still, it was interesting to hear who was doing the asking and who they were talking to.
For instance, OpenAI was taking meetings last week with “colocation providers,” which essentially offer pre-built data centers to which customers can bring their own server hardware. Today, OpenAI rents its capacity from cloud providers that own the servers.
I also heard last week that the Chinese hyperscalers, including ByteDance, are talking to cloud providers such as Oracle about renting cloud capacity in the U.S. for AI. This is particularly interesting given my colleague Qianer Liu’s recent reporting that the Chinese government is restricting companies from buying Nvidia chips. Maybe those firms will rent the capacity in the U.S. instead?
In other news:
ExclusiveApple Developing AI Wearable PinBy Wayne Ma and Qianer Liu
The Big ReadHow Greg and Anna Brockman Became MAGA MegadonorsBy Jemima McEvoy, Stephanie Palazzolo and Sylvia Varnham O'Regan
ExclusiveCrypto Exchanges Want to Trade U.S. Stocks EverywhereBy Yueqi Yang