Investing/Nov 13 edition/Reporting & analysis
StocksMarkets

StocksEquities, earnings & valuations

Cerebras shares jump about 17% after Q3 revenue beat as CEO names Microsoft as new cloud customer

Cerebras reported third-quarter core revenue of $230.9 million, up 70% and above estimates, guided fourth-quarter sales well above consensus, and said Microsoft signed a cloud deal in October, though growth slowed from Q2 and GAAP losses remain large.

From the CNBC · Earnings article of November 13, 2026 and cited sources: Q3 core revenue of $230.9M beat the roughly $214.7M expected, and Q4 core revenue guidance of $310–330M topped the $271M consensus. [8] [1]

SOURCES[8] Cerebras Q3 2026 results, SEC 8-K Ex. 99.1 sec.gov[1] Cerebras surges 17% after beating on quarterly revenue and… cnbc.com

AI-assisted, not human-edited. Methodology Corrections Who owns this site

KEY FACTS
What happened
Q3 core revenue of $230.9M beat the roughly $214.7M expected, and Q4 core revenue guidance of $310–330M topped the $271M consensus. [8] [1]
Original source
CNBC · Earnings article, independent perspective
Brief by
AT Pulse research desk AI-assisted; publisher and contact on the desk page
When
November 13, 2026 article published
Evidence
9 sources across 8 sites

Reach: not measured yet

Illustration from CNBC: Cerebras shares jump about 17% after Q3 revenue beat as CEO names Microsoft as new cloud customer
Image: CNBC — Original article ↗
THE CORE IDEAS4 TAKEAWAYS
01

Q3 core revenue of $230.9M beat the roughly $214.7M expected, and Q4 core revenue guidance of $310–330M topped the $271M consensus. The 'core' figures exclude stock compensation, customer-warrant amortization and data-center pass-through items, which turns a $161.2M GAAP net loss into $29.8M of core net income. [8] [1]

02

CEO Andrew Feldman named Microsoft as the new cloud customer in a CNBC interview and on the earnings call. The earnings release refers only to a top-three hyperscaler. He said Microsoft signed in October and that revenue should begin in Q1 2027. Size and terms were not disclosed. [1] [9]

03

Momentum is cooling. Core revenue growth fell from 103% in Q2 to 70% in Q3, and the implied 2027 outlook of about $2.85B sits slightly below the roughly $2.92B analysts expected. OpenAI still accounts for most of the $26.1B backlog. [1] [9] [5]

04

Management pitches its systems for latency-sensitive inference. It says cost per token beats small-batch GPUs above roughly 250–300 tokens per second, with CS-4 now in full production and CS-5 targeted for summer 2027. These are company figures, and the reviewed sources describe no independent benchmarks. [9] [8]

WHY IT MATTERS

a $26.1B contracted backlog and about 700 MW of capacity due by end-2027. Implication: Cerebras inference could reach buyers through more major clouds, but delivery depends on data-center buildout and on its heavy reliance on OpenAI.

Executive brief

Microsoft signed a cloud deal with Cerebras in October 2026, but Cerebras' own earnings release doesn't name Microsoft. The press release refers only to "a top three hyperscaler." Microsoft was named by CEO Andrew Feldman, in a CNBC interview and on the earnings call. No Microsoft statement appears in the reviewed sources. Feldman expects revenue from the deal to start in Q1 2027. On Nov. 12, Cerebras reported Q3 core revenue of $230.9M, up 70%, against an expected $214.7M. Its shares rose about 17% after hours, though they are still down about 47% since the May IPO.

What changed and event timeline

  1. Nasdaq debut

    Cerebras priced 30M shares at $185, above its raised range, and raised $5.55B, per the. Reports said the stock opened at $350 ().

  2. Q2 results

    Cerebras filed its second-quarter results (). CNBC reports Q2 core revenue growth of 103%, which sets up the slowdown to 70% in Q3 ().

  3. Microsoft signs

    Feldman told CNBC that Microsoft signed in October and that revenue should start showing up in Q1 2027 ().

  4. Q3 beat and stock jump

    Adjusted EPS was $0.12, versus an expected loss of $0.14. Shares rose about 17% in extended trading; reported 18% ().

  5. Nvidia CEO exit, same day

    Jensen Huang resigned as Nvidia CEO and left its board after reports of extramarital affairs. Colette Kress became interim CEO (;).

Capabilities and access

All product details below come from the company (Q3 release; call transcript):

Read the full section

All product details below come from the company (Q3 release; call transcript):

  • CS-3: currently deployed. Customers are expected to switch to CS-4 in Q1 2027.
  • CS-4: in full production. Management says it doubles CS-3 performance.
  • CS-5 / WSE-5: the fifth-generation wafer-scale chip is back from the fab. Production is targeted for early summer 2027, aiming at up to 10,000 tokens per second per user.
  • Access channels: Cerebras' own cloud, AWS, Oracle (launch targeted for summer 2027), Microsoft, and smaller cloud providers such as Fireworks.

Technical analysis for researchers and developers

  • Where it fits: management says its cost per token beats GPUs running small batches once speeds pass about 250–300 tokens/s, and is "vastly more economical" at 500–800 tokens/s (transcript).
  • Disaggregated inference: the company says this setup raises token capacity by up to 5x.
  • Reproducibility: none of the reviewed sources describes benchmark methods or independent evaluations of these speed figures.
Read the full section
  • Where it fits: management says its cost per token beats GPUs running small batches once speeds pass about 250–300 tokens/s, and is "vastly more economical" at 500–800 tokens/s (transcript). That points to latency-sensitive inference rather than high-throughput batch work.
  • Disaggregated inference: the company says this setup raises token capacity by up to 5x. It hosts Kimi K3.1, which it describes as a 2.8-trillion-parameter model (release).
  • Reproducibility: none of the reviewed sources describes benchmark methods or independent evaluations of these speed figures.

Claims and evidence

  • Q3 results: GAAP revenue was $210.8M (up 55%) and core revenue $230.9M (up 70%). GAAP gross margin was 20% and core gross margin 41%.
  • Backlog: remaining performance obligations (contracted revenue not yet recognized) were $26.1B.
  • Guidance: Q4 core revenue of $310–330M versus the $271M consensus. Full-year 2026 core revenue of $940–960M (release).
Read the full section
  • Q3 results: GAAP revenue was $210.8M (up 55%) and core revenue $230.9M (up 70%). GAAP gross margin was 20% and core gross margin 41%. GAAP net loss was $161.2M (8-K release).
  • Backlog: remaining performance obligations (contracted revenue not yet recognized) were $26.1B. Most of it is the OpenAI contract, which is reported net of estimated credits for delays (transcript).
  • Guidance: Q4 core revenue of $310–330M versus the $271M consensus. Full-year 2026 core revenue of $940–960M (release).
  • 2027 outlook: tripling revenue implies about $2.85B, against the $2.92B analysts expected (CNBC).
  • Microsoft deal: this rests on management statements only. Its size and terms were not disclosed.
  • Performance claims: the "30x faster inference" figure for CS-4 is the company's own claim.

Context and prior work

  • Cerebras builds wafer-scale processors and is trying to compete with Nvidia GPUs.
  • Before Microsoft, Cerebras already had deals with Amazon and Oracle.
  • OpenAI is its largest customer. Its only debt is an OpenAI working-capital loan from January 2026 (transcript).
Read the full section
  • Cerebras builds wafer-scale processors and is trying to compete with Nvidia GPUs. Feldman says long GPU lead times and high GPU costs create openings for its hardware sales (CNBC).
  • Before Microsoft, Cerebras already had deals with Amazon and Oracle.
  • OpenAI is its largest customer. Its only debt is an OpenAI working-capital loan from January 2026 (transcript).

Limitations, safety and contested findings

  • Slowing growth: core revenue growth fell from 103% in Q2 to 70% in Q3. ShiftDelete puts the GAAP operating margin at about -106% (ShiftDelete).
  • Adjusted vs. GAAP: "core" figures exclude stock-based compensation, customer-warrant amortization and data-center pass-through items.
  • Delivery risk: management says delays in land, power and building shells are the only constraint.
Read the full section
  • Slowing growth: core revenue growth fell from 103% in Q2 to 70% in Q3. ShiftDelete puts the GAAP operating margin at about -106% (ShiftDelete).
  • Adjusted vs. GAAP: "core" figures exclude stock-based compensation, customer-warrant amortization and data-center pass-through items. That turns a $161.2M GAAP loss into $29.8M core net income (release).
  • Delivery risk: management says delays in land, power and building shells are the only constraint. It puts cancellation risk on delayed OpenAI capacity at under 5% (transcript).
  • 2027 guide: the implied ~$2.85B is slightly below consensus.

Business and practitioner implications

  • More ways to buy: with AWS, Oracle and now Microsoft, Cerebras inference may become available through mainstream clouds.
  • Capacity: about 700 MW of data-center capacity is contracted for delivery by end-2027 (release). Cloud growth is limited by that capacity until then.
  • Funding: Cerebras holds $7.8B in cash and investments plus an undrawn $850M credit line.
Read the full section
  • More ways to buy: with AWS, Oracle and now Microsoft, Cerebras inference may become available through mainstream clouds. Microsoft's revenue is expected only from Q1 2027, and Oracle's launch in summer 2027.
  • Capacity: about 700 MW of data-center capacity is contracted for delivery by end-2027 (release). Cloud growth is limited by that capacity until then.
  • Funding: Cerebras holds $7.8B in cash and investments plus an undrawn $850M credit line. Feldman says the company is still evaluating financing options to meet demand (CNBC).
  • Customer concentration: heavy reliance on OpenAI remains, even with more hyperscaler customers.
FOLLOW THE EVIDENCE

The source trail.

Sources (9)
A LITTLE LESS NOISE. A LOT MORE CONTEXT.

Stay curious.
Follow the evidence.

Independent perspectives, the original sources, and room for the questions that don't have easy answers.

How we build the brief
Connect with us

Find us where you already read.