𼽠Metaverse Reality Check
Plus: Salesforce embraces Agents and Marvell bets on Light
Welcome to the Free edition of How They Make Money.
Over 250,000 subscribers turn to us for business and investment insights.
In case you missed it:
âď¸ MongoDB, CrowdStrike, Gitlab
𼽠Metaverse Reality Check
Zuck is finally listening. After burning nearly $70 billion on Reality Labs since 2021, Meta is reportedly preparing to slash its Metaverse budget by up to 30%.
The stock jumped on the news. Wall Street sees this as a signal of discipline and focus. Call it the âYear of Efficiencyâ part deux. Reality Labsâ losses have been an overhang for years, while the core âFamily of Appsâ business has been thriving.
Horizon on the chopping block
The company isnât exiting hardware altogether. Instead, it is a surgical strike on the software that powers the Metaverse vision.
Target: The cuts will reportedly hit Meta Horizon Worlds (the virtual social platform) and the VR unit (Quest) most heavily.
Timeline: Layoffs could happen as early as January as part of the 2026 budget planning cycle.
Meta spent billions on a platform war that no one else showed up to fight. With adoption still stuck in the gaming niche and rivals like Apple and Google pivoting entirely to AI and Spatial Computing, the âexistential threatâ has vanished. Zuck is realizing he is effectively racing against himself, which means he has the luxury to slow down.
To be sure, Zuck still has hardware on his mind. He just poached Alan Dye, Appleâs longtime VP of Human Interface Design, who will lead a new Creative Studio at Reality Labs. Dye led the design for the Apple Watch and Vision Pro interfaces. His move suggests Meta is now serious about building hardware that people actually want to wear.
From âMetaverse Firstâ to âAI Firstâ
Reality Labs has been described as a âleaky bucketâ by analysts. By capping the Metaverse spend, Zuckerberg can reallocate capital to the real war: Generative AI and Smart Glasses (the hardware that actually has traction).
If you recall, Meta and EssilorLuxottica have sold over 2 million pairs of Ray-Ban Meta AI since their launch in October 2023. The company is on track to raise production to 10 million units per year by the end of 2026 to meet demand.
Zuck has already dropped hints that the âFamily of Apps vs. Reality Labsâ reporting structure is becoming obsolete.
âOver time, weâll probably need to find better ways to articulate the value thatâs being generated here across both segments so it doesnât just seem like our hardware costs increase as our glasses ecosystem scales, but all the value flows to a different segment.â
Expect a potential re-segmentation in FY26 that blurs these linesâand conveniently hides the standalone âMetaverseâ losses inside a broader âAI Infrastructureâ cost center.
Takeaway: In late 2021, Facebook changed its name to Meta to signal the future. In 2025, the budget cuts signal the reality. By shrinking the Metaverse ambitions, Zuck is admitting that while the vision remains, the timeline was wrong. Meta is now effectively an AI company that also sells headsets. And investors are thrilled.
đś Marvell & The Nervous System of AI
Marvellâs Q3 FY26 (October quarter) was all about the pivot to AI infrastructure. The company has effectively transformed from a cyclical semiconductor maker into a pure-play AI bet.
You might notice a massive $1.9 billion windfall in âother incomeâ in the top right. This is a one-time gain from the sale of Marvellâs Automotive Ethernet business to Infineon during the quarter. While it boosted the bottom line significantly, it is a non-recurring event, which also explains why the Automotive/Industrial revenue segment dropped sharply year-over-year.
Overall revenue rose 37% Y/Y to $2.1 billion ($10 million beat), and non-GAAP EPS jumped 77% Y/Y to $0.76 ($0.02 beat). Data Center revenue surged 38% to $1.52 billion, accounting for 73% of total revenue. While Enterprise and Carrier rebounded, it was from a very low base.
In just two years, Data Center revenue has nearly tripled from ~$500 million to ~$1.5 billion, effectively offsetting the cyclical decline in the other segments
Margins improved dramatically following a restructuring last year, and operating cash flow improved 9% Y/Y to $582 million. Management leaned into capital returns with $1.3 billion of buybacks plus dividends.
Celestial AI: Buying the Photonic Future
The headline move is the $3.25 billion acquisition of Celestial AI.
The problem: As GPU clusters grow to 100k+ units, copper wires struggle to move data fast enough without overheating the rack.
The solution: Celestialâs âPhotonic Fabricâ uses light to transfer data. Their first chiplet delivers 16 Tb/s of bandwidth, unlocking the next generation of scale-up architecture.
Deal Structure:
Upfront: $1 billion cash + $2.25 billion in stock.
Earnout: Up to $2.25 billion in additional stock if revenue milestones are hit (at least $500 million cumulative by FY29 or more, with full earnout at $2+ billion).
Target: Management expects an annualized run rate of $500 million by Q4 FY28.
Multi-year AI roadmap
The near-term guide was in line, but the multi-year outlook moved the stock.
FY26: Data center revenue expected to grow >25% (excluding Celestial).
FY27: Targeting ~40% growth, driven by custom silicon and the photonic ramp.
Why the confidence? Custom silicon remains the core engine. Marvell expects sales to grow ~20% next year with no âair pockets.â Analysts now see stronger visibility into programs tied to Amazonâs Trainium, Microsoftâs Maia, and a new win with an âemerging hyperscaler.â
Takeaway: Marvell is betting that the bottleneck in AI is not just compute, but connectivity. By acquiring Celestial AI, they are securing the transition from electrical (copper) to optical (light) interconnects. Execution risk is real, and the Celestial payoff is years away. But if they hit their targets, Marvell becomes the indispensable glue holding the AI data center together.
âď¸ Salesforce: The Agentic Awakening
Salesforce posted a solid but unspectacular quarter on the surface, but under the hood, the engine is shifting gears. Revenue grew 9% Y/Y to $10.3 billion (in-line), and non-GAAP EPS landed at $3.25 ($0.39 beat).
The margin expansion story continued, with the operating margin improving to 21% (+1pp Y/Y). Importantly, the business remains a cash machine, with free cash flow growing 22% to $2.2 billion.
Agentforce: Validating the hype
The real story this quarter is the definitive pivot from experimentation to paid production. Marc Benioff called Agentforce and Data 360 the âmomentum drivers,â and the paid deal velocity backs him up.
The combined ARR for these products hit ~$1.4 billion, more than doubling year-over-year (+114%). Agentforce alone has surged to a $540 million run-rate, growing an explosive 330% Y/Y. Of course, thatâs a tiny piece of Salesforceâs ~$41 billion revenue expected in FY26, but the momentum is critical here.
Context: Salesforce revealed a $440 million Agentic AI ARR in Q2 during its Investor Day. The climb to $540 million implies a ~23% sequential growth.
Real signal: The most impressive metric is the deal volume. They closed 9,500+ paid Agentforce deals in Q3, a massive 58% jump from the 6,000 paid deals reported in Q2. And the platform processed a staggering 3.2 trillion tokens.
This is the âAgentic Enterpriseâ in action: moving from sidecar chatbots to autonomous agents embedded directly into sales and service workflows. Crucially, half of these bookings came from existing customers expanding their footprint, a classic signal of platform durability.
Benioff used the call to dismantle the âDIY AIâ narrative, noting that companies trying to build their own agents (like Klarna in late 2024) are hitting a wall on security and governanceâdriving them back to software platforms.
The data foundation
Salesforce is quietly solving the last-mile problem of AI: getting clean, harmonized data into agentic workflows.
The acquisition of Informatica (closed ahead of schedule in November), combined with Data 360, creates a data layer that management expects to become a $10 billion business next year.
The goal is to ingest tens of trillions of records to power low-hallucination agents across verticals like Life Sciences and Public Sector.
The path to re-acceleration
Management nudged the full-year outlook higher, but that was all due to Informatica closing earlier than expected, contributing roughly $330 million.
Q4 revenue: $11.2 billion (+12% on the high-end).
FY26 revenue was raised by $0.3 billion to $41.5 billion, matching the Informatica contribution.
Current RPO is expected to accelerate to ~13% growth in constant currency in Q4 (boosted by Informatica), and remains the best indicator for future revenue growth.
While organic growth is still single digits (~9%), management explicitly signaled a 12â18 month path to re-acceleration as Agentforce scales and the data stack fully integrates.
Takeaway: Salesforce sits in the middle of the AI bifurcation. It is a mature incumbent under valuation pressure (down nearly 30% this year), but the rapid growth of Agentforce offers a path out. By turning into the orchestration layer for agentic workflows, Salesforce is showing the first tangible signs of shifting from âAI victimâ toward âAI enabler.â The clock is ticking, but the pivot is real.
Thatâs it for today.
Happy investing!
Want to sponsor this newsletter? Get in touch here.
Thanks to Fiscal.ai for being our official data partner. Create your own charts and pull key metrics from 50,000+ companies directly on Fiscal.ai. Start an account for free and save 15% on paid plans with this link.
Disclosure: I own CRM and META in App Economy Portfolio. I share my ratings (BUY, SELL, or HOLD) with App Economy Portfolio members.
Marvell made a significant acquisition to address the network wall problem and become the leading option in this category, especially if we look at the best of the network connectivity, but it does have expensive hardware requirements (Disaggregation: Use massive optical bandwidth to separate memory from compute, eliminating the "memory wall"), which is not bad perse but it's harder to scale. My guess once the tech (Photonic Fabric Gen2 - Targeted at 28.8 Tbps (3.6 TB/s) per chiplet) is in production, meaning well tested, it's going to be bought or acquired by any of the hyperscalers, especially Microsoft or Nvidia, as Google and Amazon have gone with slower but more efficient and scalable routes. BTW, great article!
That Agentforce jump from 6k to 9.5k paid deals in one quarter is what matters. When you see 58% sequential growth in actual deployments (not just pilot credits), it means enterprises are moving past testing phase. The Informatica timing also matters because you need clean data pipelines before agents can do anything useful in production.