Confluent Cut 800 Jobs. Databricks Has 840 Open. Get One.
Confluent laid off 800 engineers in March 2026. Databricks is actively hiring from that list. Here's what their interview actually tests and how to land a role.
What this post actually says
IBM closed its $11B Confluent acquisition in March 2026 and cut 800 engineers, 25% of the workforce. The same month, Databricks had 840+ open requisitions. No layoffs, no freeze.
Databricks is pre-IPO at a $170.7B valuation with $5.4B ARR growing 65% YoY, and recruiters are actively sourcing from the Confluent and Snowflake layoff lists.
The Databricks loop does not measure LeetCode count. It tests production code quality, streaming-into-Delta-Lake system design, and operational judgment.
Displaced Confluent hires negotiated $350K to $450K at L4 in the March-May window. Median SWE total comp at Databricks is $504K.
The window is time-bounded: pre-IPO equity narrows as the S-1 approaches, and Confluent severance runs out within months.
800 out, 840 in: the window is open
In March 2026, IBM closed its $11 billion acquisition of Confluent and immediately cut 800 engineers, 25% of the global workforce. Same month, Databricks had 840+ open requisitions posted. No layoffs. No hiring freeze. A $5.4 billion run rate, 65% year-over-year growth, and recruiters actively sourcing from the Confluent and Snowflake layoff lists since February.
Anyone searching for Databricks interview questions right now is already thinking about this correctly. The window is open. This post covers exactly what is on the other side of it: what the loop tests, what the offers look like, and how Confluent experience translates.
What the 800 Confluent engineers carry into the market
The Confluent layoffs were not a slow bleed. IBM closed the acquisition on March 17, announced the cuts on March 18, and gave 4 months of severance with offboarding through the end of April. That is 800 mid-to-senior platform engineers and Kafka specialists entering the candidate pool inside a single quarter.
These are not generalists. Confluent employed the people who built and operated Kafka at the vendor level: broker replication internals, the metadata layer, Schema Registry, connector frameworks, exactly-once semantics at scale. The operator pool for senior Kafka cluster work was already thinning as managed offerings absorbed routine operations. Now 800 of those deep specialists are looking at once.
Kafka appears in 24% of data engineering job postings, and demand has not dropped. But candidate density surged. A Kafka specialist competing for the same roles as 800 ex-Confluent engineers who literally built the thing needs positioning more specific than “I ran Kafka.”
Know the patterns before the interviewer asks them.
“The differentiator is no longer “I know Kafka.” It is “I know what breaks at scale and how to prevent it.””
Streaming pay compression: the 2026 numbers
Commodity-tier DE comp is compressing while specialized infrastructure defies it. The market is not shrinking; it is splitting.
| Segment | 2026 comp | Direction |
|---|---|---|
| Senior Kafka engineer (production experience) | $202K median base ($100K-$306K range) | Premium eroding as 800 specialists flood in |
| Commodity-tier DE (batch ETL, A-to-B moves) | $133K median, down from $153K in early 2025 | Compressing |
| Specialized infra (Databricks, Stripe tier) | $180K-$240K+ base | Holding firm |
| Platform engineer, senior+ | $128K-$205K base, $260K-$385K total | Demand outstrips supply |
| Entry-level DE | 2.3% of total postings | Functionally gone |
Why Databricks is the destination company right now
Databricks is pre-IPO at a $170.7 billion valuation (Forge price, June 2026), up from $134 billion in December 2025, with an S-1 expected H2 2026 or early 2027. It is the only profitable AI/data company in the IPO pipeline: $5.4B ARR, 65% YoY growth, 140%+ net revenue retention, positive free cash flow. 840+ open roles. Zero mass layoffs.
The hiring mix skews toward Solution Architects, Field Engineers, platform engineering, and ML/applied AI. This is expansion, not backfill. Recruiters have a script for Confluent-to-Databricks moves because the buyer conversation is similar, the technical depth is real, and the migration story Databricks sells against Snowflake plays better when the person delivering it used to be on the other side.
Most displaced senior Confluent profiles are landing at Databricks with competing offers closing in under 30 days. Loops that typically run 6 to 8 weeks are compressing to 3 to 4. That timeline favors candidates who can demonstrate pattern recognition from prior incumbency over fresh LeetCode prep.
What Databricks actually interviews for
Databricks explicitly does not measure interview readiness by LeetCode problem count. The bar is whether a candidate can write correct, maintainable code, reason about concurrency, and handle real-world engineering with good judgment. The loop runs 5 to 6 stages over 4 to 7 weeks: recruiter screen, 1 to 2 technical screens, then a virtual onsite with 3 to 5 rounds (2+ coding, a system design round, and a behavioral/hiring-manager round).
The system design round is 45 to 60 minutes, open-ended, conducted in a Google Doc rather than on a whiteboard. The flagship problem domain is real-time fraud detection: Spark Structured Streaming plus Kafka ingestion plus MLflow inference plus Delta Lake ACID guarantees. For a Confluent engineer, that problem is 60% familiar and 40% new vocabulary for the same concepts.
This is what a medallion-architecture answer looks like when reasoning about streaming into Delta Lake. Bronze is raw schema-enforced ingestion from Kafka; silver is deduplicated, quality-checked, and enriched:
CREATE OR REFRESH STREAMING TABLE bronze_transactions
COMMENT 'Raw fraud detection events from Kafka'
AS SELECT
current_timestamp() AS ingested_at,
key AS transaction_id,
value:user_id::STRING AS user_id,
value:amount::DECIMAL(12,2) AS amount,
value:merchant_id::STRING AS merchant_id,
value:event_time::TIMESTAMP AS event_time
FROM STREAM(read_kafka(
bootstrapServers => 'broker:9092',
subscribe => 'transactions'
));
CREATE OR REFRESH STREAMING TABLE silver_transactions (
CONSTRAINT valid_amount EXPECT (amount > 0) ON VIOLATION DROP ROW,
CONSTRAINT valid_user EXPECT (user_id IS NOT NULL) ON VIOLATION DROP ROW
)
AS SELECT
t.*,
m.risk_category,
m.avg_transaction_amount AS merchant_avg
FROM STREAM(LIVE.bronze_transactions) t
LEFT JOIN LIVE.dim_merchants m
ON t.merchant_id = m.merchant_id;Delta Live Tables, bronze and silver layers. The intermediate layers are queryable Delta tables, not ephemeral Kafka state. That is the mental-model shift: Confluent engineers default to topics and external stores for state; Databricks expects Delta Lake as the authoritative sink.
The streaming knowledge translates directly. Schema evolution: Schema Registry becomes Delta Lake schema enforcement. Exactly-once semantics: same concept, different guarantee mechanism. ACLs and governance: Unity Catalog is Schema Registry for everything. What does not translate is the frame. Confluent engineers think in unbounded streams and event brokers. Databricks interviewers want real-time pipelines that land in governed, queryable, ACID-compliant data assets. Same destination, different frame.
The bridge pattern the system design round probes, in Python:
from pyspark.sql import functions as F
stream_df = (
spark.readStream
.format("kafka")
.option("kafka.bootstrap.servers", "broker:9092")
.option("subscribe", "transactions")
.option("startingOffsets", "latest")
.option("kafka.isolation.level", "read_committed")
.load()
)
parsed_df = (
stream_df
.select(
F.col("key").cast("string").alias("transaction_id"),
F.from_json(
F.col("value").cast("string"),
"user_id STRING, amount DECIMAL(12,2), event_time TIMESTAMP",
).alias("data"),
F.col("timestamp").alias("kafka_timestamp"),
)
.select("transaction_id", "data.*", "kafka_timestamp")
)
(
parsed_df.writeStream
.format("delta")
.outputMode("append")
.option("checkpointLocation", "/checkpoints/bronze_transactions")
.trigger(processingTime="30 seconds")
.toTable("bronze.transactions")
)Structured Streaming, Kafka source to Delta Lake sink. A candidate who can walk through this code, explain the checkpoint mechanism behind exactly-once, and articulate how read_committed isolation interacts with the Delta transaction log is speaking the loop's language. That is production fluency, not LeetCode.
Pre-IPO Databricks comp by level
A concrete March 2026 L4 offer: $190K base + $600K RSU grant (4-year vest, 1-year cliff) + $30K target bonus, roughly $430K year-one total. Median SWE total comp across Databricks: $504K.
| Level | Total comp range | Typical profile |
|---|---|---|
| L3 (entry) | $253K - $380K | New grad / 0-2 YOE |
| L4 (mid) | $380K - $550K | 3-5 YOE, production systems |
| L5 (senior) | $550K - $800K | 6-10 YOE, tech lead |
| L6 (staff) | $800K - $1.2M | 10+ YOE, org-level impact |
| L7 (principal) | $1.2M - $1.65M | Company-level scope |
Reading the equity: real money, with a haircut
Confluent’s median total comp was $261K. Displaced Confluent hires negotiated $350K to $450K at L4 in the March through May window. A meaningful uplift, but below the $500K+ Databricks’ internal bands suggest: pay compression on incoming senior hires is real when 800 candidates hit the market simultaneously.
One thing that changed in 2025: Databricks removed the “second trigger” on RSUs. Vested units now settle into actual shares before IPO, which means tender-offer optionality mid-year. That is material. Pre-IPO equity is real money if the IPO lands on timeline and lockup expires without dilution. Databricks’ 140% NRR and profitability tilt the odds, but model a 25% to 30% haircut as the base case.
Also: no 401(k) match, at $500K+ median comp. Price that into the comparison.
Snowflake's Cortex move: what it signals
Snowflake cut roughly 700 positions, including its entire technical writing team of about 70 people, and replaced them with Project SnowWork: Cortex AI models generating documentation. This is not financial distress. Snowflake reported 30% product revenue growth and 9,100+ customer accounts. It is strategic: free budget for AI talent by eliminating roles AI can approximate.
The catch: Cortex Analyst accuracy is 85% to 90% on well-defined semantic views but drops to 47% without inference context. They replaced humans with a system that is wrong half the time on ambiguous inputs. For interview prep, that gap is the opportunity. A candidate who can articulate why 47% baseline accuracy, stateless conversation handling, and 100M-row ceilings break real customer workflows signals they understand the distance between marketing narrative and production reality. That distance is where hiring happens.
The bifurcation is the story
Databricks hiring at scale while Confluent sheds 800 is not one company winning and another losing. The market is splitting into 2 tracks.
Track one: commodity pipeline work
ETL, basic orchestration, moving data from A to B. Being absorbed by platforms like Fivetran and Airbyte. Entry-level roles down to 2.3% of postings. Median salary compressing toward $133K.
Track two: infrastructure architecture
Real-time feature stores, LLM output governance, medallion/DLT-native systems, platform engineering with self-service discovery. 80% of large orgs have dedicated platform teams by end of 2026; real-time workloads are 60% of new pipelines. Demand outstrips supply.
The clock is running
Databricks’ 840-role sprint is time-bounded. The pre-IPO equity window narrows as the S-1 approaches: engineers interviewing this quarter get better equity valuations than those hired after a public offering. Confluent’s severance runs out soon. And the compressed 3-to-4-week loop favors engineers who prep for what Databricks actually tests instead of defaulting to generic LeetCode grinding.
A displaced Confluent engineer with 5+ years operating Confluent Cloud sits squarely on the winning side of the market split. Operator skills (broker tuning, partition rebalancing, disaster recovery) are scarcer than baseline Kafka knowledge despite the supply surge. 800 out, 840 in. The math is simple. The prep is specific.
The prep sequence that matches the actual loop
- 01
Coding rounds: production code, not tricks
Structured, maintainable solutions with edge cases and testing strategy discussed out loud. Practice writing code you would defend in review, not code that merely passes.
- ▸Talk through failure modes as you write
- ▸Concurrency reasoning comes up; refresh it
- 02
System design: streaming into Delta Lake
The flagship domain is real-time fraud detection: Kafka ingestion, Structured Streaming, MLflow inference, Delta Lake ACID guarantees. Practice decomposing into bronze/silver/gold and defending each boundary.
- ▸Know Delta transaction log semantics cold
- ▸Unity Catalog RBAC is the governance answer
- 03
Translate the streaming background
Build the narrative that maps Schema Registry to schema enforcement, exactly-once to checkpointing, ACLs to Unity Catalog. Interviewers reward the lakehouse frame delivered by someone with broker-level depth.
- ▸Lead stories with production incidents you owned
Common misconceptions vs hiring-manager reality
Databricks hiring in 2026: direct answers
Does Databricks have a hiring freeze in 2026?+
Did Databricks do layoffs in 2026?+
Is Databricks hiring data engineers right now?+
What does Databricks pay in 2026?+
How hard is the Databricks interview?+
Prep for the loop Databricks actually runs
- 01
Reading a solution is not the same as writing one
Every engineer who has frozen on a query they had read a dozen times knows the gap. The only preparation that closes it is producing the answer yourself, under time, before the interview does it for you
- 02
76% of hiring managers reject on the coding task, not the resume
From HackerRank's 2024 Developer Skills Report. Candidates who look strong on paper still fail the live screen if they haven't done timed, executable practice
- 03
System design comes down to the calls you defend out loud
Ingestion, batch vs streaming, the bronze/silver/gold layers, idempotency, backfill and replay. Sketching the pipeline and naming the failure modes is the signal, not the boxes
Related interview prep
Databricks Data Engineer process, Spark internals, lakehouse architecture, Delta Lake questions.
Snowflake Data Engineer process, micro-partitions, query optimization, warehouse architecture.
Streaming Data Engineer interview, Kafka, Flink, exactly-once, event-time vs processing-time.