Microsoft wants Windows 11 users to talk to their PCs and let AI agents manipulate files. Cortana failed at exactly this a decade ago, and corporate IT won't enable experimental agents that create new attack surfaces. Third time's the charm?
Oracle booked $65B in AI cloud deals—including Meta—in 30 days, but the stock slipped after-hours despite raised 2030 targets. Investors want proof on 30–40% margins, not bookings. Demand's real. Execution's unproven. The scoreboard moves there.
Anthropic wired Claude into Microsoft 365, chasing institutional memory while Microsoft hedges with model choice inside Copilot. The fight isn't about chat features—it's about who mediates your company's knowledge day after day.
Oracle booked $65B in AI cloud deals—including Meta—in 30 days, but the stock slipped after-hours despite raised 2030 targets. Investors want proof on 30–40% margins, not bookings. Demand's real. Execution's unproven. The scoreboard moves there.
Scarcity is giving way to supply. Now comes the hard part: margins.
Oracle confirmed a multi-year cloud deal with Meta and said it booked $65 billion in new AI-infrastructure commitments over roughly 30 days—seven contracts across four customers, and none tied to OpenAI despite that partnership’s $300 billion headline. Shares rose about 3% Thursday before slipping 2% after-hours as Oracle lifted its fiscal-2030 targets to $225 billion in revenue and $21 in adjusted EPS—above consensus, but not enough to still the profitability debate. Investors heard the story. They want the math.
The Meta contract turns months of chatter into booked business and, more importantly, demonstrates demand that isn’t just one customer. Co-CEO Clay Magouyrk stressed that Oracle now sees “many customers” chasing compute and that supply becomes easier to secure in later years as power, land, and next-gen hardware come online. That’s the pivot. Sell the out-years, deliver in phases.
Oracle also put numbers on its software side of AI, projecting $20 billion in “AI-powered database and platform” revenue by fiscal 2030, up from $2.4 billion in fiscal 2025. It’s an eightfold jump that assumes scarcity fades and deployment scales. Big ifs, but a clear trajectory. The signal is demand. The risk is timing.
Key Takeaways
• Oracle booked $65 billion in AI infrastructure across seven contracts from four customers in 30 days, none from OpenAI
• Meta deal confirmed at roughly $20 billion over multiple years, adding to $66–$72 billion Meta's spending in-house
• Oracle guides 30–40% adjusted margins on AI infrastructure versus mid-teens reported on GPU rentals—lifecycle economics remain unproven
• Stock slipped after-hours despite raised 2030 targets as investors demand proof on margins, not projections on bookings
Meta budgeted $66–$72 billion in 2025 capex and is still turning to outside capacity. That sounds contradictory. It isn’t. External contracts smooth training peaks, hedge schedule risk, and diversify hardware generations while the company builds its own sites. It’s a hedge.
Across the sector, cloud contracts are morphing into options on future compute, not just purchases of today’s instances. OpenAI, xAI, and now Meta are locking in allocations before all the gear is energized. For Oracle, that validates a long, expensive pivot from database licensing into heavy infrastructure competition with Amazon, Google, and Microsoft. For Meta, it pries open a fresh lane when in-house builds and hyperscaler queues collide with power and permitting.
One quiet enabler here is Oracle’s willingness to meet customers where they already run. Its integrations let enterprises use Oracle Cloud Infrastructure alongside native services in the big three clouds. That model eases procurement friction and broadens the addressable base without forcing a one-cloud bet. It also spreads Oracle’s risk. If it delivers.
The challenge is simple to state and brutal to execute: energize multi-gigawatt campuses on schedule, stand up high-bandwidth networks, and keep utilization high from day one. The bookings suggest trust. The grid and supply chain will test it. Execution is everything.
Oracle guided 30–40% adjusted gross margins on AI infrastructure over the life of contracts, after land, data centers, power, and equipment. That’s far better than the mid-teens margins reported on GPU rentals in recent quarters. Both can be true. Early builds carry start-up drag; mature sites throw off better unit economics at scale. Still, “adjusted” does a lot of work here, and early mega-deals often come with pricing concessions to win logos and volume.
This is where the stock’s hesitation lives. Oracle argues lifecycle margins will normalize higher as sites fill and mix shifts toward next-gen hardware. Skeptics want proof in reported gross margin, not pro formas. Show me, don’t tell me. Fair ask.
Magouyrk emphasized that none of the $65 billion in recent commitments came from OpenAI. That line matters. It answers the “one-tenant story” and reduces perceived concentration risk even as the $300 billion OpenAI pact towers over any single contract. In practice, the OpenAI deal likely set a reference point for scale and cadence, giving other buyers confidence that Oracle would underwrite multi-year allocations at speed. That’s how momentum compounding looks in infrastructure.
The flip side: diversification only helps if delivery does. If power slips or hardware shows up late, customers will flex to other providers or squeeze price on delays. Multi-vendor hedging cuts both ways.
Oracle hasn’t disclosed the Meta phasing, hardware mix, or recognition profile. Whether capacity starts on H200 and rolls to GB200—or includes AMD share—will shape realized margins. So will where the compute lands and what the power contracts look like. We also don’t know how much of the $65 billion converts within the next six quarters versus anchoring the outer years. These details decide cash needs, reported margins, and the glide path to that 2030 target.
For now, two truths can sit together. The contracts suggest customers still fear supply. The market reaction suggests investors still fear execution. Numbers are big; execution is bigger.
If Oracle adds “several additional multi-billion-dollar customers,” as hinted, bookings will keep outpacing its historical build tempo. That’s good only if delivery keeps pace and unit economics track the 30–40% guide as cohorts mature. Watch for quarterly disclosures that separate early-stage rental drag from contract-life margins. Watch for campus energization dates and GB200 ramps. And watch for how quickly Meta’s tranche comes online. The scoreboard moves there.
Q: Why did Oracle's stock drop after-hours if it raised guidance above analyst estimates?
A: Oracle guided to $225 billion in 2030 revenue versus the $198.39 billion analysts expected, but investors want proof on margins, not just bookings. The company claims 30–40% adjusted gross margins on AI infrastructure while recent quarters showed mid-teens margins on GPU rentals. That gap—and the word "adjusted"—triggered skepticism about whether lifecycle economics will materialize as promised.
Q: How much is Meta spending on AI infrastructure total if it's also building its own data centers?
A: Meta announced $66–$72 billion in capital expenditures for 2025 alone, primarily for data centers and AI hardware it owns. The roughly $20 billion Oracle deal sits on top of that in-house spending. External capacity hedges schedule risk, smooths training peaks across hardware generations, and provides flexibility while Meta's owned sites come online. It's insurance, not replacement.
Q: What does "adjusted gross margin" mean and why does it matter here?
A: Adjusted gross margin excludes certain costs—in Oracle's case, likely early build-out expenses and one-time items—to show long-term profitability once sites reach steady state. Oracle argues lifecycle economics across multi-year contracts justify the 30–40% guide, but investors remember recent quarters showing 14% gross margin on Nvidia chip rentals. The question's whether "adjusted" hides structural issues or just normalizes for startup drag.
Q: Why would Meta use Oracle instead of AWS, Google Cloud, or Microsoft Azure?
A: Oracle's integration lets customers run its infrastructure alongside services in AWS, Azure, and Google Cloud—a hybrid model that eases procurement without forcing a single-cloud decision. Meta also likely values Oracle's willingness to sign massive forward commitments at scale, while hyperscalers ration capacity more carefully across broader customer bases. Diversification reduces dependence on any single provider's delivery timelines or allocation priorities.
Q: What happens if Oracle can't deliver the infrastructure on schedule?
A: Contracts this large typically include delivery milestones tied to payment schedules. If Oracle misses timelines because of power delays, equipment shortages, or construction slippage, customers can flex to other providers, negotiate price concessions, or push capacity into later quarters. The $65 billion in bookings only matters if Oracle energizes multi-gigawatt campuses, stands up networks, and maintains high utilization—execution nobody's operated at this scale before.
Marcus Schuler
Marcus Schuler
Marcus Schuler
Microsoft wants Windows 11 users to talk to their PCs and let AI agents manipulate files. Cortana failed at exactly this a decade ago, and corporate IT won't enable experimental agents that create new attack surfaces. Third time's the charm?
Anthropic wired Claude into Microsoft 365, chasing institutional memory while Microsoft hedges with model choice inside Copilot. The fight isn't about chat features—it's about who mediates your company's knowledge day after day.
Apple's newly promoted Siri AI chief lasted three weeks before jumping to Meta—the twelfth senior researcher to leave this year. With a March overhaul deadline looming and more exits expected, Apple's talent strategy collides with reality.
Anthropic's Haiku 4.5 delivers May's frontier coding performance at one-third the cost, collapsing the capability-to-commodity timeline to five months. Multi-agent systems just crossed the economic threshold for production deployment.
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
