Jul 9, 2026

Welcome to The Hidden Layer. I’m Ian Krietzberg.

Two weeks ago, OpenAI announced that the release of its latest model family would be limited and staggered at the request of the White House. Today, GPT-5.6 made its public debut, cementing worries about what frontier model releases will look like amid the administration’s new quasi-licensing scheme.

Meanwhile, Mark Zuckerberg is reportedly pondering re-reinventing Meta as a cloud company—an unsurprising move that nevertheless is loaded with implications for the robustness of the A.I. business. We’re getting into all of it today, plus news and notes on a few different aspects of the A.I. jobs story, and Microsoft’s $2.5 billion bet.

If this email was forwarded to you, consider subscribing here for access—not just to The Hidden Layer, but to the industry-leading reporting from all my incredible colleagues across Puck. Your support makes our work possible, and we appreciate it.

Also mentioned in this issue: Sam Altman, Youssef Squali, Susan Li, Rishi Bommasani, Doug Anmuth, Gil Luria, Sundar Pichai, Joshua Achiam, Teresa Carlson, and more.

Let’s get into it…

Three Things You Should Know…

Another side of the jobs story, part one: Plenty of ink has been spilled on A.I.’s impact on the labor market, but less on the technology’s impact on hiring practices. Researchers at Stanford recently found, unsurprisingly, that around 90 percent of employers use A.I. “to some extent” while hiring, but that screening algorithms often made racially biased recommendations. In an analysis of more than 4 million applications, researchers found that the algorithms for some jobs recommended Asian and Black candidates significantly less. “Some companies think that A.I. will help them be more fair in their decision-making,” said lead author Rishi Bommasani, a researcher at Stanford’s Institute for Human-Centered Artificial Intelligence. “That’s not necessarily what our results suggest.”
Another side of the jobs story, part two: Ramp, the fintech firm, recently published an analysis suggesting that companies spending around $33 per employee per month on A.I.—their definition of high-intensity adopters—grew their overall head count by around 10 percent in the two years following adoption. Entry-level head count for high-intensity adopters climbed 12 percent over that same period. Low-intensity adopters, which spend less than $3 per employee per month on A.I., saw no statistically significant head count changes, according to Ramp. “The results counter predictions that A.I. adoption will lead to broad job loss,” the company contended.
The rotating door: Teresa Carlson, who founded and led the General Catalyst Institute, departed the organization earlier this week for Anthropic, where she will be its global head of public sector. … And, on Tuesday morning, Joshua Achiam—a nine-year OpenAI veteran who was most recently serving as its chief futurist—announced his resignation. “There’s not a specific reason for me leaving, or a specific reason for why now,” he wrote in a post. “But it’s something I have been thinking of for a while and it feels right. The world is in on the secret now and it feels possible to work on the mission from outside the walls of a frontier lab.”

Capital Intelligence

Microsoft kicked off July by pouring $2.5 billion into the Microsoft Frontier Company, which will hire 6,000 experts—essentially forward-deployed engineers—to help commercial customers utilize A.I. This comes just a week or so after Amazon committed $1 billion to forward-deployed engineers of its own. (This has been IBM’s whole business model for forever, by the way.)
Amazon is also planning to raise at least $25 billion in an eight-part bond sale, according to CNBC. The company raised $54 billion in bonds this year, all as its free cashflow has dwindled to almost nothing. Building all those data centers isn’t cheap, is it, Andy?
Together A.I., an inference platform, secured an $800 million Series C round at an $8.3 billion valuation.
Resolution, a nonprofit A.I. alignment org, was given a $160 million grant from Coefficient Giving (formerly known as Open Philanthropy).
8090, an A.I. coding startup building “software factories,” raised a $135 million Series A round, led by Salesforce.

And now for the main event…

Zuck’s A.I. Landlord Era

Meta’s new plan to start selling compute raises a bunch of uncomfortable questions, and not just for the neoclouds it might put out of business: After years of Meta needing more compute than it could produce, what does its sudden surplus say about demand for A.I. overall?

Ian Krietzberg

Meta, as you may have heard, is developing a plan to get into the cloud infrastructure business. This latest pivot, which was reported by Bloomberg earlier this month, has been slightly misunderstood as an echo of the company’s previous, quixotic attempt to redefine itself around the nebulous “metaverse.” Instead, this latest reorientation is a little more complex and could take shape in a few different ways.

On the one hand, Facebook and Instagram’s parentco could end up selling access to different A.I. models that it self-hosts, similar to some of the hyperscale cloud services, like AWS Bedrock. Indeed, Meta kicked off an early stage of what this might look like today, allowing developers to pay to access its latest coding model through a new A.P.I. Alternatively—or, perhaps, additionally—Meta could sell access to raw compute, akin to the fundamental sales pitch of the debt-riddled neoclouds such as CoreWeave and Nebius. (Meta did not return a request for comment on the report, but C.E.O. Mark Zuckerberg confirmed the thought process to Bloomberg today, saying: “The offers that you get for using the compute are so high that it may make sense, in some cases, to rent out or consider those kind of deals instead of your own internal uses.”)

Either way, the plan has gone over quite well with investors—in the week since the Bloomberg story, shares of Meta are up about 11 percent—and it’s not hard to figure out why. Like its hyperscaler contemporaries, Meta has spent the past few years pouring tens of billions into compute infrastructure. By year-end of 2025, the company was telling investors to brace for between $115 billion and $135 billion in capital expenditure in 2026, to “support our A.I. efforts and core business.” By April, Meta had hiked the top end of this year’s capex guidance to $145 billion.

But unlike Google or Amazon, Meta doesn’t operate a cloud service, so there was never a clear path to recoup all those dollars outside of the core business of selling advertising across its social platforms. Nevertheless, Zuckerberg and C.F.O. Susan Li had often argued for the opportunistic need to relentlessly build compute anyway: “Being able to make a significantly larger investment here is very likely to be a profitable thing over some period,” Zuck said during a call with investors late last year. Anyway, now Meta has a couple of clear paths to monetization outside of its core business. “Meta is under pressure to monetize A.I. beyond advertising, and providing access to its infrastructure would generate incremental revenue and meaningful returns,” J.P. Morgan analyst Doug Anmuth wrote in a note.

And yet there’s a riddle inherent in this news. Meta had justified its capex spend largely on account of its own growing business needs; now, as Zuck recently suggested, the company may have inventory to rent. So is Meta tapping out on A.I. needs and looking to monetize its capex in alternative ways? And, simply put, can a company as large as Meta achieve an actual A.I. glut in this environment? What about the company’s plans to stand up a $10 billion, 1-gigawatt, A.I.-optimized data center in Canada—is this infrastructure going to service the company’s needs, or merely serve its landlord strategy? And does the delta between current demand, current supply, and future supply matter at all, at least right now?

The New A.I. Landlords

Zuckerberg is not the first mogul to find himself in a position to arbitrage this investment cycle, merging compute necessity with some extra feudal recurring revenue to boot. CoreWeave, the most prominent example of an A.I.-focused neocloud, started in 2017 as a crypto mining company, then pivoted hard to A.I. in 2023 to put its on-hand G.P.U.s to use. Crusoe, another former Bitcoin miner, made a similar transition in 2025. But the closest analogue to Meta lies with SpaceX, which started racking up compute deals in the lead-up to its June I.P.O.—including its billionish-per-month deals with Google and Anthropic, and a more recent deal for $150 million-per-month to Reflection A.I.

Even Sam Altman, who seems to be in a perpetual compute crunch, floated a comparable idea in November, when he wrote: “We are also looking at ways to more directly sell compute capacity to other companies (and people); we are pretty sure the world is going to need a lot of ‘A.I. cloud,’ and we are excited to offer this.” (Anthropic’s Dario Amodei has yet to suggest any similar concepts.)

So much is still unknown, but Meta’s plan “doesn’t bode very well” for the neoclouds, Truist analyst Youssef Squali told me. The CoreWeaves of the world now look like subscale competitors in the A.I. leasing marketplace next to SpaceX and now Meta, and whichever hyperscaler follows. If all the A.I. companies start morphing into partial cloud companies as well, he posited, “we’re going to get to a point where there may be oversupply.” This impression was echoed by D.A. Davidson analyst Gil Luria, who said in a note that more than half of Nebius’s backlog, and a third of CoreWeave’s, was accounted for by… Meta. “If Meta is selling compute, it probably doesn't need much external capacity,” Luria said, which opens the neoclouds to “significant disruption.”

The Glut Factor

Analysts are torn about how to decipher Meta’s move: Does the company have capacity, or is it strategically arbitraging its neocloud hombres? As Squali acknowledged in a note, “This news is interpreted by some as an indication that Meta already has excess capacity, a prospect that would be negative for the A.I. ecosystem as a whole” given that the investment thesis in the sector posits that demand currently, and radically, outpaces supply. “From every conversation we’ve been having with industry participants, there remains an acute demand for compute,” Squali told me.

Adding yet another layer to the fuzziness around the supply question are recent potential hits to A.I. demand in general. Squali mentioned a more recent trend of enterprises establishing stricter limits on token spend, for instance—something that, widespread, could compress demand. Then there’s competition from cheaper Chinese models, which European companies especially are leaning on, according to Squali. “If we compress the demand side, then the overbuild will be here before we know it,” he said.

Meta’s cloudification, Squali continued, has added a lot more weight to the company’s 2027 capex numbers, which will ultimately settle the question of whether the company is actually experiencing a supply glut. And that will certainly have ramifications across the A.I. pipeline. “If Meta comes out and says, Well, we’ve had enough with the capex spend, it’ll mean that they’re also topping on top-line growth,” Squali said.

Then again, earnings are coming up in just a few weeks, and Squali thinks that Meta, like Google, will soon need to raise another equity offering. This new cloud potential, like the company’s new subscription offerings and model releases, might just entice investors before such an equity offering goes live. (As always, this is not investment advice.)

That’s all for today. I’ll see you next week.

Ian

Puck sports correspondent John Ourand and a rotating cast of industry insiders take you inside the executive suites and owners boxes where the decisions that shape the entire sports business are made. You’ll hear interviews with players, network execs, and everyone in between. The Varsity is an extension of John’s private email for Puck by the same name. New episodes publish every Wednesday and Sunday.

Puck’s daily political newsletter from Washington on what’s really happening in this town, from the White House to the Pentagon to Capitol Hill, K Street, the campaign trail, and beyond.

Inside Platner’s Campaign Collapse

PETER HAMBY

A $26B I.P.O. Mystery

WILLIAM D. COHAN

Sun Valley Intel

DYLAN BYERS

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St., New York, NY 10006