Oracle reported after the bell last night. Here are the numbers.
Revenue: $19.2 billion, up 21% year over year. Beat. EPS: $2.11 against a $1.97 consensus. Beat. OCI — Oracle’s cloud infrastructure business — grew 93% year over year to $5.8 billion. Beat. RPO, the backlog of contracted but not yet delivered revenue, jumped from $553 billion to $638 billion. That is an $85 billion increase in committed future revenue in a single quarter. Beat — by a lot.
The stock dropped 10%.
Before we talk about why that is completely backwards, let me say something directly.
We Called This. All of It.
Yesterday morning we published a pre-mortem on Oracle’s Q4 earnings. We said beat and raise was the base case. We said the real number to watch was not headline revenue but RPO conversion pace — specifically, whether the $553 billion backlog was translating into delivered megawatts faster than Oracle’s Q3 pace. We said to watch Clay Magouyrk’s specific language on data centre delivery. We said the capital structure was not a crisis, it was a feature. And we said this:
“Beat with a cautious tone. Revenue and EPS above consensus but management vague on the capex timeline and free cash flow conversion. The stock may sell off despite the beat. This happened in December 2025.”
That is exactly what happened.
Oracle beat on every headline metric. Management then guided for FY2027 capex of $70 billion net — with a reported number closer to $90-95 billion once you include $20-25 billion in customer prepayments — plus a plan to raise another $40 billion in debt and equity to fund the buildout. The market read “$95 billion capex” and “$40 billion new debt” and sold. Hard.
We are not going to tell you we feel bad about that. We told you yesterday exactly which scenario to prepare for. We told you the sell-off trigger. We told you what the language on the call would sound like. The pre-mortem is dated and timestamped. You can go read it right now.
What we want to do today is tell you what the market got wrong — and why it matters enormously for what comes next.
The Market Punished Oracle For Having Too Many Customers
Let us start with the RPO number, because it is not getting nearly enough attention.
$638 billion. That is Oracle’s contracted backlog of future cloud revenue as of the end of Q4. It grew by $85 billion in a single quarter. To put that in context: $85 billion in 90 days is more than the annual revenue of Goldman Sachs, Morgan Stanley, and JPMorgan Chase’s investment banking divisions combined. In contractual AI infrastructure commitments. In one quarter.
And it gets more specific. Clay Magouyrk disclosed on the earnings call that Oracle signed $67 billion in new AI infrastructure contracts in Q4 alone. The majority were either bring-your-own-hardware or prepaid — meaning customers are paying Oracle upfront to build data centres around the customers’ own hardware. Oracle gets paid before it spends the capital. The margin on those structures is at or better than standard contracts.
Oracle’s GPU utilisation rate across its entire global fleet: 97.5%. For reference, industry benchmarks for well-run cloud infrastructure hover around 65-70%. Oracle is running nearly every GPU it has. Every single week.
Delivery pace: Oracle delivered more than 1.2 gigawatts of compute capacity to customers in FY2026. Clay said Q1 FY2027 is “approaching 1 gigawatt” — in a single quarter. Nearly the same amount as the prior four quarters combined. The pre-mortem told you to watch this number. It accelerated.
So to recap: more demand than Oracle can build for. Near-perfect GPU utilisation. Delivery pace accelerating. Customers paying upfront. $85 billion in new contractual commitments in 90 days.
And the stock dropped 10%.
The $95 Billion Is Not Oracle’s Problem. It Is Oracle’s Evidence.
Here is the thing about a $95 billion capex number that the sell-side missed entirely last night.
That money does not disappear. Oracle is not setting it on fire. It is spending it. On chips. On memory. On networking. On power. On servers. On cooling. On optical connectivity. Every single dollar of that $95 billion is revenue for someone else.
The question the market should have been asking after Oracle’s call is not “how is Oracle going to fund this.” Oracle has already told you. $75 billion in bring-your-own-hardware and prepaid customer contracts. $40 billion in planned debt and equity issuance. Customer prepayments covering $20-25 billion of the reported capex directly. The funding is accounted for.
The question worth asking — the one that matters for our portfolio specifically — is: who collects Oracle’s $95 billion?
Oracle is not the only hyperscaler with this problem. Amazon, Microsoft, Google, and Meta are collectively spending more than $300 billion on AI infrastructure in 2026. OpenAI just confidentially filed for an IPO. OpenAI is an Oracle customer. SpaceX, pricing its own IPO on Friday, announced it will build AI data centres in Spain. Every major capital allocator in the world is building the same infrastructure Oracle is building.
The AI trade is not a narrative. It is a $638 billion backlog. It is 97.5% GPU utilisation at one company alone. It is $85 billion in new contractual commitments in 90 days. The infrastructure supercycle that we have been writing about since we launched this publication is not slowing. Oracle’s Q4 is the clearest confirmation we have had all year.
The companies that supply that infrastructure — the chips, the memory, the networking, the switching, the optics — are not speculating on demand. They are collecting on contracts. Oracle just told you their customers’ customers are signing 3-year prepaid deals and bringing their own hardware to make sure the supply never runs out.
We have been holding three of the exact companies that sit in Oracle’s supply chain since May 22. One more has been on the Watch List since the day we launched. Oracle’s call last night was the best possible confirmation we could have asked for on every single one of them.
The 3 LENS Positions Oracle Just Confirmed
The Baptista Research LENS Index is OUTPERFORMING THE S&P 500 BY MORE THAN 3% EVEN IN THIS FALLING MARKET. And we have made some VERY SPECIFIC BETS that are helping us achieve this.
Here is our logic. Every dollar of Oracle’s $95 billion capex flows through a specific supply chain. Let us walk through it.
Continue Reading With Our 7-Day Free Trial
GPU chips: Oracle deploys Nvidia GPUs for standard AI workloads and is actively building relationships with custom ASIC providers for hyperscaler-scale deployments. The trend toward custom silicon — building chips specifically designed for inference and training at scale rather than general-purpose GPUs — is Oracle’s largest capex line item and the fastest-growing category in AI infrastructure. One LENS position is directly in this supply chain. It has been in the index since June 5.
Memory: Oracle’s Clay Magouyrk called out component costs specifically on the earnings call. He was talking about HBM — High Bandwidth Memory — which sits directly on the GPU package and is the single most critical bottleneck in AI inference at scale. $95 billion in AI data centre buildout means $95 billion that needs to be populated with HBM. There is exactly one company that matters here in the context of the AI trade, and it has been in LENS since June 5 as well.
Networking: Oracle described on the call how its “Acceleron” network technology delivers the highest performance and lowest cost connectivity in AI data centre design. The switching layer between GPU racks — the infrastructure that decides where every packet goes between thousands of chips operating in parallel — is the LENS Watch List name that has been waiting for its entry trigger since May 22. Oracle’s call just moved the timeline on that trigger.
Here is the current state of all positions:
| # | Ticker | Company | Tier | Status | Alloc % | Entry Date | Entry $ | Current $ | Return % | 12M Target | Hard Stop $ | Thesis Break Condition | Next Catalyst |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | DELL | Dell Technologies | CORE | HOLD | 8% | 22-May-26 | $291.00 | $369.83 | +27.1% | $540.00 | $218 | Q3 AI server rev below $14B OR FY2027 guide cut below $55B | Q2 FY2027 earnings (Aug 2026) |
| 2 | AVGO | Broadcom Inc. | CORE | BUY | 7% | 05-Jun-26 | $385.73 | $373.50 | -3.2% | $575.00 | $289 | Q3 AI bookings below $20B OR hyperscaler reduces XPU commitment | Q3 FY2026 earnings (Sep 2026) |
| 3 | MU | Micron Technology | BASE | BUY | 5% | 05-Jun-26 | $863.57 | $884.00 | +2.4% | $1,250.00 | $605 | HBM pricing collapse OR Samsung closes HBM quality gap | Q4 FY2026 earnings (Jun 24, 2026) |
| 4 | NOW | ServiceNow Inc. | BASE | BUY | 4% | 05-Jun-26 | $113.39 | $107.50 | -5.2% | $170.00 | $79 | Q2 net new ARR fails to accelerate | Q2 2026 earnings (Jul 2026) |
| 5 | GOOGL | Alphabet Inc. | BASE | BUY | 4% | 05-Jun-26 | $368.00 | $363.60 | -1.2% | $470.00 | $258 | Cloud backlog conversion below 40% | Q2 2026 earnings (Jul 2026) |
| 6 | MRVL | Marvell Technology | SPEC | BUY | 3% | 05-Jun-26 | $238.49 | $263.96 | +10.7% | $400.00 | $155 | Scale-up optics below $200M OR FY2028 guide cut | Q1 FY2027 earnings (Jun 18, 2026) |
| — | CASH | Undeployed Capital | — | — | 69% | — | — | — | — | — | — | — | — |
LENS Active Position 1: Broadcom (AVGO) — CORE, 7% The Oracle call confirmed exactly what the AVGO thesis requires: hyperscalers signing multi-year, prepaid AI infrastructure commitments at scale. Oracle’s $75 billion in bring-your-own-hardware and prepaid contracts represents the demand backdrop for Broadcom’s custom AI accelerator business. The XPU roadmap is validated by the same demand signal that drove Oracle’s $638 billion RPO. AVGO is currently -3.2% from our entry following the CPI selloff. The thesis is intact. The Oracle call strengthened it.
LENS Active Position 2: Micron Technology (MU) — BASE, 5% Oracle’s Clay Magouyrk flagged memory prices explicitly on the earnings call. He described a contract structure that protects Oracle from component cost increases — which means those costs are real and elevated. The HBM pricing backdrop that underpins our MU thesis is confirmed by a hyperscaler who just told you he has explicit contractual mechanisms to pass those costs to his own customers. MU is currently +2.4% from entry. Earnings are June 24. The Oracle call is the cleanest possible setup going into that print.
LENS Active Position 3: Marvell Technology (MRVL) — SPEC, 3% Oracle’s network infrastructure is built for AI workloads at a scale that requires purpose-built Ethernet switching and scale-up optics. Marvell’s custom networking silicon and optical connectivity products sit exactly in the supply chain that Oracle is scaling. MRVL is +10.7% from entry. The Oracle confirmation raises our conviction on the thesis without changing the thesis break condition.
LENS Watch List: Arista Networks (ANET) Oracle’s call was the clearest single confirmation yet of the ANET thesis. The company described the complexity of operating “extremely complex clusters that require constant care and feeding, constant maintenance across the network.” That is a description of Arista’s product. Oracle, Amazon, Microsoft, and Google are all building the same infrastructure. The switching layer is irreplaceable. The ANET entry trigger is 10% pullback from current levels. The Oracle call raises our conviction that the pullback, when it comes, is a buy.
The Oracle Verdict for LENS
Oracle is not in the LENS Index. We told you that yesterday. The market has fully priced Oracle’s AI story. There is no narrative gap.
But Oracle just gave every AI infrastructure supplier their most powerful quarterly endorsement in 2026. $95 billion getting spent on the supply chain. $638 billion in contracted demand that has to be fulfilled. 97.5% utilisation. Delivery accelerating.
The companies that collect Oracle’s capex are not speculating. They are invoicing.
Three of them are in LENS. One of them is on the Watch List. None of them dropped 10% last night.
Disclaimer: We do not hold any positions in the above stock(s). Read our full disclaimer here.
