Nvidia’s New Sales Chief, Anthropic’s Mega IPO—and 12 More Things That Matter in Tech This Fall

This summer wasn’t exactly uneventful in Silicon Valley: No stretch of time in the ever-changing AI age could possibly be described as such. But with everyone back at their desks, the next few months ahead seem absolutely chockablock with action—most notably, Anthropic’s IPO, with investor expectations high for the long-awaited offering. (We hope they read the IPO filing’s footnotes carefully.) That debut will come as growing public unease around AI has cast doubt on data center projects across the country and raised uncertainty about consumer demand for the technology. (In other words, maybe that Jony Ive–designed device from OpenAI won’t be the next iPhone.) And with Bay Area real estate prices soaring, can anyone possibly be fortunate enough to play host in a new home when the holidays hit? Perhaps if they can figure out how to make better use of their startup equity.

As always, our reporters have their eyes keenly on the horizon. Below you’ll find our Weekend section’s very first fall preview package, which spotlights the burning questions on their minds as they contemplate what’s coming on their beats in the next three months. Sure, other publications at this time of year like to wax on about the finest flannel and the coziest stews. Lame! We’re concerned about matters like supply capacity at TSMC’s chipmaking facilities, Amazon’s lackluster Nova models and much, much else. What—were you expecting us to, uh, turn over a new leaf?

1. What Will Happen With Nvidia’s New Sales Chief—and Its New Chips?

A week ago, Nvidia’s longtime head of sales, Jay Puri, retired from his day-to-day role after 21 years at the company. People who worked closely with the 72-year-old Puri credited him with building up Nvidia’s business of selling AI server hardware to companies—now more than 90% of the chip design giant’s revenue. His replacement is a company outsider: Nick Parker, a Microsoft veteran. I’m curious how Parker might shake things up, especially since Puri’s departure is only the most prominent exit in a broad changing of the guard for Nvidia’s sales organization: Many of his longtime direct reports have also retired recently.

Parker has the difficult task of getting customers to buy Nvidia’s complete set of hardware offerings—server chips plus products devoted to networking and storage—at a time when many of its biggest customers, such as Google, Amazon and OpenAI, are now racing ahead with their own AI chips. And his arrival coincides with the rollout of Nvidia’s new Vera Rubin server racks. We’ll know more about whether that debut is going smoothly as customers get large-scale shipments starting this month. The expanded rollout spotlights questions such as whether customers will actually have enough power and networking equipment for the chips to function as soon as they arrive.—Phoebe Liu

2. Can a Mega Data Center Survive the AI Backlash?

As the midterm elections near, data centers have become a target for voters frustrated about AI’s seemingly unfettered acceleration. Poll after poll reflects a bipartisan consensus against new construction, and plenty of politicians have picked up on that sentiment, abandoning their previous support for the facilities. (The notable exception is President Donald Trump).

Perhaps no project better represents the promise—and giant scope—of America’s AI infrastructure binge than the data center site in Piketon, Ohio, and it could very well become the largest lightning rod in the entire data center debate during the upcoming election season. The project is a collaboration between OpenAI, SoftBank, Nvidia, the U.S. and Japanese governments, and a slew of Wall Street firms. Billed as the world’s largest AI data center, it will cost hundreds of billions of dollars. Given the staggering budget and the cast of characters involved, the arguments we’ll hear voiced in favor of the site’s completion—and against it—will echo debates happening across the country.

Exactly how it’ll play out remains unclear. Last week, The Guardian recently reported a surprising lack of organized opposition. Piketon’s own mayor, Billy Spencer, said his “emotions range between positive and skeptical.” He’s weighing how the facility could bring new jobs against the fears of some residents that it could bring hazardous waste.—Leo Schwartz

3. Will a Small Drug Trial Supercharge Longevity Research?

The many past attempts to develop longevity drugs have floundered at a critical stage: when it’s time to show convincing evidence that aging can be reversed. That might start to change later this year—at least in one instance.

Life Biosciences, co-founded by Harvard scientist and longevity influencer David Sinclair, is not the only company pursuing a technique called partial epigenetic reprogramming, which purports to revert aged cells to a more youthful state. But it is the first to get the Food and Drug Administration to sign off on a trial testing the therapy in patients. The company is enrolling people with two types of age-related eye diseases. The first patient was dosed in June, and CEO Jerry McLaughlin told me the company should have safety data on at least three patients by the end of the year.

Normally I wouldn’t be so interested in preliminary results from just three patients. But promising early data is likely to bolster the hopes of many in the longevity field to keep trying to reach the ultimate goal: reprogramming an older person’s entire body. The trial’s significance goes beyond an attempt to validate a drug. It’s an indicator of whether the dream of someday finding the fountain of youth is any closer to coming true.—Amy Dockser Marcus

4. What Will Dario Amodei’s IPO Letter Say?

When Anthropic’s IPO prospectus arrives in the coming weeks, the immediate headlines will focus on backward-looking financial metrics, rehashing much of what we already know about its incredible growth.

Other parts of the IPO filing will offer more revealing details about what’s happening today and what’s coming.

I’ll first turn to the prose with Anthropic CEO Dario Amodei’s name on it—his letter to shareholders. Amodei likes to communicate through essays rather than tweets. And certainly, he’s never written anything with such high stakes: These types of missives in an IPO filing are a way for a company to turn a founder’s neatly stated vision for the business into marketing material at a time when its bankers need to go out and ask public market investors for a ton of money. It’ll be interesting to see how Anthropic discusses AI safety—not the most upbeat subject—in a forum where most CEOs only express sunshiny optimism.

Deeper in the document’s footnotes will be details about Anthropic’s data center contracts, which likely won’t sit on its actual balance sheet. The IPO filing will also reveal much about Anthropic’s intertwined relationship with Amazon and Alphabet, two of its biggest shareholders and cloud vendors: Just how much revenue is circulating between the companies?

Then there’s the matter of who controls Anthropic. The filing should give us more information about whether the company has followed through on its discussions to give Amodei supervoting shares, which could offset the power it has handed to an independent trust of nonshareholders. How will that jive with prospective new shareholders?—Cory Weinberg

5. How Much Chipmaking Capacity Will Nvidia and Google Get?

Taiwan Semiconductor Manufacturing Co. typically finalizes prices and production allocations during the fall for the following year. Its decisions this season will shape a growing contest between Nvidia and Google.

Both companies want more capacity from TSMC. Nvidia has overtaken Apple as TSMC’s largest customer this year and needs that capacity for its new Vera Rubin graphics processing units. Meanwhile, Google is ramping up plans for its own TSMC-made AI chips—its tensor processing units—and has begun selling them to customers, posing a new challenge to Nvidia’s core business.

The constraint in TSMC’s manufacturing abilities is based on supply, not demand. TSMC executives previously said orders continue to exceed what it can produce despite its accelerated expansion of its manufacturing facilities. Its fall allocations will largely decide whether Nvidia’s scale allows it to maintain a commanding lead in chips and how much room Google has to grow as a supplier.—Qianer Liu

6. Will a New Hedge Against AI Risk Take Off?

Starting this fall, two major exchanges, CME Group and Intercontinental Exchange, both want to begin offering compute futures—financial instruments tied to the rental prices of GPUs. I’m intrigued to see if their plans come to fruition, because that would signal AI’s widening reach into financial markets. Compute futures would allow traditional Wall Street traders to wager on the outcome of the AI boom with highly liquid products. At the same time, they would offer AI companies renting GPUs the ability to hedge against their exposure—much as, say, airlines buy energy futures to hedge against fuel costs—and help those companies better manage their finances if the AI economy gets wobbly.

Goldman Sachs and JPMorgan Chase are among a group of financial institutions that have been thinking about trading compute futures. Other market participants would likely include brokers and trading firms such as DRW, StoneX, FalconX and Wintermute.

CME and Intercontinental Exchange, the parent company of the New York Stock Exchange, still need regulatory approval from the Commodity Futures Trading Commission. Intercontinental has announced partnerships with Ornn, a compute market and data provider, and Nativx, a public market for compute, to list contracts based on their indices. CME has said it hopes to launch futures based on Nvidia’s H100 and Blackwell B200 chips as soon as Oct. 5.—Yueqi Yang

7. Can Cursor Reenergize Grok?

There’s always a lot going on with Elon Musk, and one part of his busy empire I’m particularly interested in is its newest addition, Cursor—and how well he can meld it with the rest of SpaceXAI.

If he can do so successfully, he may be able to resurrect his ambitions to develop AI products that compete with OpenAI’s ChatGPT and Anthropic’s Claude. If he fails, that may push SpaceXAI in a different direction—toward offering more compute rental in its data centers to outsiders like Anthropic. Why would SpaceXAI make such a shift? Well, it may make more financial sense for it to rent out its data centers for billions of dollars a month than to reserve their use for developing internal products like Grok. So more of these compute deals in the future could suggest the Cursor integration isn’t going well.

A lot is riding on the Cursor team. Musk recently warned SpaceXAI staff that its chatbot, Grok, was falling behind rivals. Grok has never been especially popular among business users—but Cursor has been. Its coding tool helped spark the vibe code era and has been one of the most popular such programs. Cursor CEO Michael Truell and several other engineers have already been given leadership roles at SpaceXAI, so presumably they will be working on more than just Cursor. I’ll be keeping an eye on whether they can actually roll out new models that impress coders. Another intriguing indicator will be if we see a lot of Cursor talent continue to exit the company. More than three dozen employees have left since the startup announced its initial partnership with Musk back in April.

I’m also curious to see whether Musk lets Cursor remain a separate product or whether he drops the brand and just tries to fold the technology into Grok. Would such a decision suggest Grok had gotten strong enough to exist alone as the flagship product? Perhaps. But the Musk tea leaves are often hard to read with precision.—Grace Kay

8. What Will Ternus Do With Apple’s Aging Lieutenants?

When John Ternus ascended the throne and officially became Apple’s chief on Tuesday, he took on a much older executive staff than the one his predecessor, Tim Cook, inherited 15 years ago. At 51, Ternus is virtually the same age as Cook when he took over in 2011—Cook was 50—but the average age of Apple’s senior executives has climbed from 48 to 59 over the past 15 years.

That average reflects the steadiness Cook brought to the company: He was a guy who really valued the presence of lieutenants who were well steeped in its unique culture. At times, though, this stability at the top has frustrated ambitious up-and-comers who find they have limited upward mobility in Apple’s ranks.

But now, a number of executives are close to reaching retirement age—most notably, Eddy Cue, head of Apple’s services business (Apple TV, iCloud, the App Store and more); Johny Srouji, head of hardware; and Greg Joswiak, marketing chief. In short order, Ternus will be looking around to find their worthy successors, and I wonder what palace intrigue will unfold as he considers his options. Other changes are also taking place: Phil Schiller, 66, who recently stepped down as leader of the App Store and product events, though he’s sticking around in an advisory role as an Apple Fellow.—Aaron Tilley

9. Can You Really Buy a House With Startup Stock?

As AI startup valuations have climbed higher and higher, plenty of people have been left sitting on piles of valuable equity. But since that equity is fairly illiquid, it’s somewhat tricky to put it to use for things like real estate purchases.

A handful of Bay Area housing listings grabbed attention this summer, with the sellers advertising a willingness to accept startup equity as part of a down payment. But as far as I can tell, no one in San Francisco actually took them up on it. As the Anthropic IPO further turbocharges the real estate market, I wonder if someone actually will. It’d be another sign of how wild it has gotten.

Legal experts have told me it’s theoretically possible to make such a transaction work, but it’s not simple. Real estate agents I’ve talked to said clients at companies like OpenAI who have explored this avenue typically came away feeling like it would just be easier to make a traditional offer. And of course, there are other ways to get cash from equity without having to sell, such as securities-backed loans. Still, the prospect of a seamless swap remains appealing in a place that’s awash with equity. —Eli Rosenberg

10. Can Microsoft Curb Its Anthropic Habit?

A year ago, Microsoft struck a deal to use Anthropic’s models in its flagship 365 Copilot software. The decision wasn’t cheap. It’s much more affordable for Microsoft to use OpenAI’s models to power Copilot because it has the rights to reuse OpenAI’s technology. But the Anthropic deal was seen internally as a necessary move to get Copilot to work better and burnish its reputation with customers.

As we’ve reported over the past year, Microsoft’s spending on Anthropic models has only grown since then, both to power Copilot and for its own internal use. But more recently, Microsoft has started taking steps to cut back on its Anthropic dependency. The company has been testing out newer OpenAI models as well as open-source ones—from the likes of DeepSeek and Moonshot—to replace Anthropic models in Copilot. Executives have also gently nudged employees in recent months to use OpenAI models for tasks such as generating code, to keep costs in check.

It remains to be seen whether Microsoft can effectively pivot away from Anthropic models more meaningfully in the coming months, but senior executives see doing so as a necessary step to shore up Copilot’s margins and avoid contributing to a winner-takes-all scenario for Anthropic in the AI race.—Aaron Holmes

11. Will OpenAI’s Jony Ive Device Actually Arrive?

Over the last year, OpenAI—like other major AI developers—has focused on the lucrative business of selling AI to companies rather than consumers. What does that mean for its consumer AI business? It’s somewhat uncertain, but one clue will lie in whether OpenAI’s Jony Ive–designed AI hardware actually debuts anytime soon, as promised earlier this year. (The hockey puck–shaped device, intended to sit on a desk or a table, will provide AI help with everyday tasks and questions.)

A further signal of OpenAI’s consumer AI ambitions would come from ChatGPT’s user numbers—and whether we see them increase. Such growth has previously been slower than OpenAI had hoped. The AI lab aimed to reach 1 billion weekly users of ChatGPT by the end of last year: In reality, it took more than seven months longer for the company to actually reach that milestone. More recently, OpenAI announced that its advertising business had surpassed $1 billion in annualized revenue run rate. That figure sounds impressive, but it suggests the company will fall short of the $2.4 billion in ad revenue it had initially projected for 2026.—Stephanie Palazzolo

12. Will Amazon Make Its Nova Models Open Source?

Three years ago, Amazon made a bet that it could compete with frontier labs in developing its own AI models. This summer, Amazon began a retreat: It shut down its San Francisco AI lab and laid off other staff working on its Nova models. On a July earnings call, Amazon CEO Andy Jassy said that while the company was still working on developing Nova, “Amazon can have a wildly successful business without its own frontier model.”

Maybe Amazon will kill Nova entirely. I’m not so sure. Either way, we might get a sense of its plans in late November, when Amazon Web Services holds its annual customer conference. In previous years, AWS has used the occasion to make announcements about Nova.

One possibility is that Amazon decides to make Nova open source. One of Nova’s main competitive advantages against better-performing models is price, and no price is better than free. Besides, customers would still likely have to pay AWS for the compute to run the models.

Open source is trendy now. Nvidia has lately gone on a shopping spree for open-source model makers, and a chorus of tech’s top leaders have thrown their public support behind open-source models, hoping to stop Washington from regulating them. One of those people has been Matt Garman, who runs AWS.—Catherine Perloff

13. Can Huawei’s New Smartphone Chip Boost Its AI Plans?

Huawei, the giant Chinese hardware manufacturer, hopes to soon introduce a new version of its Kirin processor, the main chip in its smartphones. It’s been developing that chip using a new technique that stacks its circuitry in layers to make it faster and more energy efficient.

Whether the chip works as planned is hugely consequential to AI: Huawei hopes to apply the same technique to its Ascend processors, used for training and running AI models. Those new chips could mean Chinese firms can rely less on hard-to-obtain Nvidia hardware, which faces U.S. export bans.—Qianer Liu

14. Will Meta Find a Foothold in Enterprise AI?

Mark Zuckerberg’s AI ambitions are huge and varied. (When are his ambitions not huge and varied?) And while Meta Platforms remains best known for consumer apps like Facebook and Instagram, Zuckerberg hopes to grab a piece of the market for selling AI to businesses. Well, him and everyone else! Enterprise AI has become the hottest part of AI this year and involves all the biggest players in AI. Meta isn’t well known at all for its enterprise AI aspirations, and if Zuckerberg can get any traction, that would be a real coup—and would suggest that Meta can actually compete with its AI rivals across the board.

In the company’s July earnings call, Zuckerberg said he sees a “large” opportunity to sell APIs for its AI models, as well as agentic technology and maybe even compute, to businesses. Meta has already launched an agent on WhatsApp and Messenger that can answer customer questions, recommend products, book appointments and assist with closing sales. It plans to add that agent to Instagram, too. Zuckerberg has said Meta eventually wants to turn the technology into a service that helps companies run their businesses.—Jyoti Mann