Oracle Cut 30,000. The 17 OCI Engineers Your Title Filter Is Hiding.

On March 31, 2026, Oracle sent 6 a.m. emails to roughly 30,000 employees and cut their badges the same day. Layoffs.fyi has it as the single largest tech RIF of 2026. The press framed it as Cerner cleanup and ERP consolidation, which is half true and half a distraction: OCI Enterprise Engineering, OCI AI, and data-center-ops staff went out the same door as the legacy database people.

Those are the engineers who built the compute layer OpenAI's Stargate runs on. And most recruiters are about to walk right past them, because "Oracle" pattern-matches to legacy database talent in every ATS built in the last fifteen years.

What Oracle actually cut on March 31

Oracle terminated approximately 30,000 employees on March 31, 2026, about 18% of its ~190,000-person global headcount, with the cuts spanning the US, India, Canada, Mexico, and Uruguay. The narrative was Cerner/Oracle Health and ERP consulting. The reality, per WARN filings and firsthand posts, was broader.

Impacted teams, per DataCenterDynamics and state filings:

• OCI Enterprise Engineering division
• Fusion ERP
• Data center operations technicians
• Technical project managers for the AI/ML team
• The broader OCI AI team
• Cerner / Oracle Health (heaviest)
• NetSuite's India Development Centre

The two divisions hit hardest, Revenue and Health Sciences (RHS) and SaaS/Virtual Operations Services (SVOS), each lost about 30% of staff. Seattle's WARN notice covered 161 roles including software engineers, cloud support engineers, and program managers. There was no prior warning. System access was cut immediately. The email was the last working day.

A former Oracle security alert manager posted publicly, per Computer Weekly, that the cuts appeared to follow "an algorithm of high-level individual contributors and mid-level managers, especially those with outstanding stock options." That is a cash-flow-driven cut: higher salaries mean bigger savings. Translation for anyone hiring: the pool skews senior IC, not junior.

Why the OCI pool is mispriced right now

The OCI pool is mispriced because the engineers who built one of the largest AI training clusters on Earth are labeled "Oracle" on LinkedIn, and every keyword filter in the industry is tuned for AWS, GCP, and Azure. The Stargate context makes that mispricing acute.

Here is what OCI has been doing while nobody looked:

• OCI Zettascale10 targets deployments of up to 800,000 NVIDIA GPUs and up to 16 zettaFLOPS of peak performance (Oracle newsroom, October 2025).
• Zettascale10 is the fabric under the flagship supercluster Oracle built with OpenAI in Abilene, Texas, internally called "Project Ludicrous."
• Abilene began receiving NVIDIA GB200 racks in mid-2025 and scales to 1.2 GW across eight buildings (Introl).
• Oracle and OpenAI signed a deal in July 2025 to develop 4.5 GW of additional Stargate capacity in the US, part of a 10 GW, $500 billion program.
• The commercial anchor: a $30 billion annual OpenAI/Oracle deal reported by Fierce Network earlier this year.

800,000

The people who commissioned gigawatt-scale GPU clusters running proprietary RoCE fabric are, by any reasonable count, a group of a few hundred humans on Earth. Meta, xAI, Anthropic, and CoreWeave are the other buyers. Oracle just put a slug of them on the street at Oracle prices.

The numbers in Refolk's index

In Refolk's index of professional profiles, 367 US-based people currently have "Oracle Cloud Infrastructure" or "OCI" in their headline. Filter that same pool by core infrastructure engineering titles and it collapses to 17. That is a 95% miss rate for any recruiter using a naive title filter.

The mechanism behind the miss is boring and important: OCI engineers self-describe by product, not function. Real headlines from the pool read "Master Principal Cloud Architect," "VP of Software Engineering, OCI," "Principal Product Manager, AIML Strategic Customers," and "Lead Software Engineer." None of those hit on a Boolean like ("SRE" OR "infrastructure engineer" OR "distributed systems") AND ("GPU" OR "cluster"). The AWS-shaped filter silently excludes them.

This is the exact gap Refolk closes. You describe the person in plain English ("US-based engineer who worked on OCI's GPU cluster fabric or Zettascale, senior IC, open to work") and get a ranked shortlist that ignores whether their title happens to include the word "infrastructure."

The five keywords that actually surface the pool

The five resume keywords that actually surface OCI infra engineers are Zettascale, Acceleron, RoCE, GB200, and Abilene. If your sourcing Booleans don't include these, you are searching a different pool than the one Oracle just released.

Here is why each one matters:

1. Zettascale / Zettascale10. Oracle's public name for the 800K-GPU cluster architecture. Only engineers who touched it list it. Announced October 2025, so anyone with it on their profile is current.
2. Acceleron RoCE. Oracle's proprietary RDMA-over-Converged-Ethernet fabric for GPU clusters. Non-transferable knowledge, high signal, near-zero false positives.
3. GB200. NVIDIA's Blackwell rack-scale system. The Abilene site is one of the first at-scale GB200 deployments. Engineers who commissioned it are a global handful.
4. Abilene / "Project Ludicrous". The physical site. Anyone who names it worked on Stargate directly. DFW-adjacent, so also a geographic tag.
5. 1.2 GW / gigawatt-scale. Power-density work at this scale is done by maybe a dozen orgs on the planet. Engineers who talk about it on LinkedIn are self-selecting into the pool you want.

The AWS-shaped Boolean is why 350 of these engineers stay invisible for the next four weeks.

Where they live (and where to be)

The pool is geographically narrow: Austin, SF Bay, Denver, Seattle, and the Abilene/DFW corridor. That is where Refolk's index concentrates OCI-linked profiles and where Oracle's WARN filings landed.

Top US metros in the OCI-mentioning pool, per Refolk's index:

• Austin, TX (heaviest concentration)
• San Francisco Bay Area
• Denver, CO
• Seattle, WA (also the confirmed 161-role 2025 WARN metro)
• Add Abilene / Dallas-Fort Worth for the on-site Stargate build engineers

If you're a founder in any of those five metros, you have a physical-proximity advantage over the AI-infra recruiter in NYC or London who will DM these people on LinkedIn next week. Buy coffee. Show up at the March/April meetups. This is a four-week window, not a four-month one.

The 17-to-30 day placement window

Per KORE1's placement data, most displaced senior Oracle engineers are landing in mid-market and PE-backed roles within 17 to 30 days. If you're reading this in April 2026, you have roughly four weeks before half this pool is off-market.

The likely landing pads, all hiring the exact GPU-cluster plus RoCE plus power-density skill set Oracle just released:

• CoreWeave. The obvious buyer. Public, cash-rich, expanding.
• Nscale. UK-based, aggressively hiring cluster engineers in the US.
• Crusoe. Data-center operator with Stargate-adjacent contracts.
• Lambda. GPU cloud, direct competitor to OCI's AI business.
• xAI's Memphis buildout. Same power-density profile as Abilene.
• Anthropic's compute team. Small, senior, and hiring.
• Together AI. Fresh $8.3B round, sourcing hundreds of infra roles.
• PE-backed data-center operators. Quiet, well-funded, and moving fastest.

If you are competing with that list, "we move faster than Anthropic" is the only meaningful pitch. The way to move faster is to source before the LinkedIn Recruiter cohort figures out the keywords above. Refolk is built for exactly that, ask in plain English and pull the shortlist in one query, then reach out the same afternoon.

The senior IC signal, and why it matters for your offer

The Computer Weekly source described the layoff algorithm as targeting "high-level individual contributors and mid-level managers, especially those with outstanding stock options." That means the pool is unusually senior and specifically people with unvested equity they just lost.

Practical implications for your offer package:

• Front-load equity refresh. These candidates just watched a vesting cliff evaporate. A sign-on grant matters more than a sign-on bonus.
• Skip the take-home. Senior ICs from an org that ran 800K-GPU clusters do not need to prove they can implement a linked list. Reference-check and offer.
• Frame the story. "P&L-driven pruning of exactly the people we want" reads better than "Oracle refugee." Say it that way in the first message.
• Assume competing offers by week two. Move on final numbers by day 10, not day 20.

Named entry points into the affected network

The public entry points into the affected OCI network are Mahesh Thiagarajan's org (EVP OCI, the public face of Zettascale10) and the LinkedIn network around Michael Shepherd, an AI operations lead at OCI who publicly reposted colleagues' layoff notices per Computer Weekly.

Two practical moves for anyone sourcing this week:

1. Second-degree from Michael Shepherd's reposts. Every "open to work" comment on those posts is a warm intro. Refolk can pull the commenter set and cross-reference against the OCI Zettascale keyword pool in one query.
2. Abilene contractors and vendors. GB200 commissioning is not just Oracle badges. NVIDIA field engineers, power-density subs, and RoCE network vendors on site in Abilene know exactly who did what. LinkedIn's "worked with" data on the Abilene cohort is unusually clean because the project is small and dated.

The core argument here is short: Oracle just released, into a market that isn't looking, the engineers who built one of the three or four largest AI training clusters on Earth. The window is four weeks. The keywords are five words long. The pool is 367 profiles, of which 350 are invisible to a title-only search. If your Q2 hiring plan includes AI infrastructure, this is the highest-leverage sourcing moment of the year.

FAQ

How many of the 30,000 Oracle layoffs were actually OCI engineers?

Oracle has not published a division-level breakdown, but WARN filings and firsthand accounts confirm OCI Enterprise Engineering, OCI AI, Fusion ERP, and data-center operations were all in scope alongside Cerner and ERP consulting. Refolk's index shows 367 US profiles currently listing OCI or Oracle Cloud Infrastructure in their headline, and the Seattle 2025 WARN precursor alone covered 161 cloud-adjacent roles. The senior IC skew reported by Computer Weekly suggests the OCI infra slice, while smaller than Cerner, is disproportionately valuable.

Why won't LinkedIn Recruiter surface this pool?

Because OCI engineers self-describe by product ("Master Principal Cloud Architect," "Principal PM, AIML Strategic Customers"), not by function ("infrastructure engineer," "SRE"). Standard AWS-shaped Booleans exclude them. The Refolk cut proves it: 367 US profiles mention OCI, but only 17 hit on core-infra title filters, a ~95% miss rate. The real signal lives in project keywords: Zettascale, Acceleron, RoCE, GB200, Abilene.

What is Stargate and why does it matter for hiring?

Stargate is the OpenAI, Oracle, and SoftBank data-center program targeting 10 GW of AI training capacity and $500 billion in buildout. The flagship site is Abilene, Texas ("Project Ludicrous"), running on OCI's Zettascale10 fabric with NVIDIA GB200 racks scaling to 1.2 GW across eight buildings. Engineers who touched Stargate have gigawatt-scale GPU cluster commissioning experience that maybe three or four organizations on Earth can claim, which is why the Oracle layoff temporarily prices them below market.

How long is the sourcing window?

Roughly four weeks from the March 31 termination date, per KORE1 placement data showing most displaced senior Oracle engineers land in 17 to 30 days. CoreWeave, Nscale, Crusoe, Lambda, xAI, Anthropic, and Together AI are all hiring the same skill set. Any article dated after mid-May 2026 is a follow-up piece about who won them, not a sourcing guide.