CBRS · Forward model · Bear case
The Bear case, 18 quarters out
Each segment is projected from its own operating driver, rolled up into consolidated cash flow, and discounted back to a fair value per share. The assumptions below are editable — change them and every number on this page moves with them.
Built on GAAP. Cerebras reports one segment; the two lines here are the disaggregation of revenue shown on the face of the statements of operations every quarter, and they reconcile to reported consolidated revenue with zero difference in all ten quarters, from $66.6M in 2024 Q1 to $180.1M in 2026 Q2. Only 2026 Q1 is estimated, and it is confirmed twice: six-month less June-quarter arithmetic and the Q1 release's own $110.6M and $82.8M. Nothing else is apportioned. All company guidance is non-GAAP core and core revenue exceeded GAAP revenue by $29.8M in the basis quarter, so every guided figure used here was converted through the disclosed bridge - GAAP equals core plus pass-through less customer-warrant amortisation, which was $44.3M in the June quarter and is expected in a $40-55M quarterly band. What is ours and not the company's: the growth rates and their decay, both margin glides, the capex glide, corporate overhead, the tax rate, the discount rate and the exit multiple. Three choices need stating plainly. First, corporate overhead is a single 20% of revenue for the whole horizon because the engine has no overhead glide; that reproduces roughly today's $118.8M of quarterly core operating expense at 2027 revenue, but it understates the 2026 loss, where actual core opex is 57% of core revenue. Second, valuation uses the 237,564,041 shares outstanding at 2026-08-05 and holds them flat, so the 76.0M of disclosed potential shares - 32% more - are charged nowhere. Third, the September quarter projects high: guidance implies core revenue only 2.4% above June and then a 25% step in December as capacity tranches go live, and a single decaying growth rate cannot be flat and then step, so this model is calibrated to the annual totals instead: the September quarter projects $253M against a GAAP-equivalent guide of about $190M, while FY2026 lands at $970M against a core guide midpoint of $885M. The 2026 Q2 GAAP net loss of $450.5M is dominated by $377.0M of IPO-triggered stock compensation and is not a run rate; adjusted EBITDA was $(53.1)M. Net cash of $7.69bn is cash, restricted cash and investments of $8.61bn less the $918.2M Working Capital Loan; it includes $684.7M of restricted cash and excludes $568.8M of on-balance-sheet operating lease liabilities and about $1.5bn of undiscounted payments on datacentre leases not yet commenced. This model burns about $1.7bn of cumulative free cash flow and spends $9.7bn of capex over the horizon, and charges no interest, no stock compensation and no dilution against any of it.
The build lands and the revenue does not follow on schedule. The June quarter is the existence proof: $416.9M of capex against $180.1M of revenue, a 14.2% GAAP gross margin and cloud cost of revenue up 313% year over year against cloud revenue up 281%. The contractual asymmetry makes a slow quarter worse than a slow quarter. If Cerebras misses a capacity tranche or a service level, OpenAI can terminate part or all of the MRA, and the $25.4bn of remaining performance obligations is that agreement; termination also puts the $1.0bn Working Capital Loan on demand. With 76% of revenue in three customers this is a single-counterparty event, not a diversified miss. Here the 2027 tripling does not happen, the margin glide stalls near today's core level and the multiple compresses to the 3.0x carried by the leased build-out comparable.
Latest: $1.22B (2030Q4E)
| Period | Value |
|---|---|
| 2024Q1 | $67M |
| 2024Q2 | $70M |
| 2024Q3 | $72M |
| 2024Q4 | $82M |
| 2025Q1 | $100M |
| 2025Q2 | $103M |
| 2025Q3 | $136M |
| 2025Q4 | $171M |
| 2026Q1 | $193M |
| 2026Q2 | $180M |
| 2026Q3E | $245M |
| 2026Q4E | $323M |
| 2027Q1E | $412M |
| 2027Q2E | $506M |
| 2027Q3E | $603M |
| 2027Q4E | $696M |
| 2028Q1E | $784M |
| 2028Q2E | $865M |
| 2028Q3E | $936M |
| 2028Q4E | $998M |
| 2029Q1E | $1.05B |
| 2029Q2E | $1.09B |
| 2029Q3E | $1.13B |
| 2029Q4E | $1.16B |
| 2030Q1E | $1.18B |
| 2030Q2E | $1.20B |
| 2030Q3E | $1.21B |
| 2030Q4E | $1.22B |
What drives each segment
Hardware
Growth pathRacked CS-series wafer-scale systems sold to customers who run them on their own premises, recognised at a point in time on delivery. This was the whole company until 2025 and is now the minority line: $54.1M of GAAP revenue in 2026 Q2 against $126.0M of cloud. The CS-4 was announced in August 2026 with first shipments in the September quarter and manufacturing capacity is guided to rise more than 10x in 2026 across Flex, Sanmina and Rocket EMS. No filing discloses system units or an average selling price, so the line cannot be driven by volume and price; it is modelled as a growth line off the reported base.
Latest: $57M (2030Q4E)
| Period | Value |
|---|---|
| 2024Q1 | $49M |
| 2024Q2 | $55M |
| 2024Q3 | $50M |
| 2024Q4 | $57M |
| 2025Q1 | $70M |
| 2025Q2 | $70M |
| 2025Q3 | $97M |
| 2025Q4 | $122M |
| 2026Q1 | $111M |
| 2026Q2 | $54M |
| 2026Q3E | $56M |
| 2026Q4E | $57M |
| 2027Q1E | $58M |
| 2027Q2E | $59M |
| 2027Q3E | $59M |
| 2027Q4E | $60M |
| 2028Q1E | $60M |
| 2028Q2E | $60M |
| 2028Q3E | $60M |
| 2028Q4E | $60M |
| 2029Q1E | $60M |
| 2029Q2E | $59M |
| 2029Q3E | $59M |
| 2029Q4E | $59M |
| 2030Q1E | $58M |
| 2030Q2E | $58M |
| 2030Q3E | $58M |
| 2030Q4E | $57M |
Assumptions & reasoning
- Ten quarters of disclosed revenue and cost of revenue, unusually complete for a company two quarters public, because the IPO prospectus carried an eight-quarter Quarterly Results of Operations table covering 2024 Q1 to 2025 Q4.
- 2026 Q1 is the only estimated point: six-month hardware revenue less the reported June quarter gives $110,593K, and the Q1 2026 release independently states hardware revenue of $110.6 million.
- The 1.8% GAAP gross margin in the basis quarter is not an operating result. $28.0M of customer-warrant amortisation is charged against this line as contra-revenue; core hardware gross margin was 38.8%, which is what the segment margin here is set to.
- Warrant amortisation runs through October 2031 and is recognised in proportion to related revenue, so GAAP hardware revenue understates the commercial line by roughly a third at current scale and the understatement back-loads.
- Point-in-time recognition makes this line genuinely lumpy: sequential moves of -51%, +38% and +26% all appear in the last six quarters, and management said on 2026-08-12 that it manages total core revenue rather than the mix between the two lines.
- No seasonality array is set. Detrended factors from a centred four-quarter moving average put the Q4 factor at 1.101 with a window-to-window spread of 0.287, so the spread beats the signal and the naive and detrended estimators disagree in sign on Q1.
Cloud and other services
Growth pathThe inference service, and now the company. Cerebras contracts multi-year datacentre capacity, leases and fits out the sites, installs its own systems and recognises revenue over time as capacity is delivered and consumed: $125.6M of the quarter's $126.0M was over-time revenue. The line grew 281% year over year on GAAP and is 70% of consolidated revenue. Economically this is a capacity business, but live megawatts, revenue per megawatt and utilisation are all undisclosed - the only capacity figure given is a combined 'more than 600 megawatts of data center capacity that is either live now or will be delivered by the end of 2027' - so it is driven as a growth line anchored to the disclosed $25.4bn RPO recognition schedule rather than on an invented earning base.
Latest: $1.16B (2030Q4E)
| Period | Value |
|---|---|
| 2024Q1 | $17M |
| 2024Q2 | $15M |
| 2024Q3 | $22M |
| 2024Q4 | $24M |
| 2025Q1 | $30M |
| 2025Q2 | $33M |
| 2025Q3 | $39M |
| 2025Q4 | $50M |
| 2026Q1 | $83M |
| 2026Q2 | $126M |
| 2026Q3E | $189M |
| 2026Q4E | $266M |
| 2027Q1E | $354M |
| 2027Q2E | $448M |
| 2027Q3E | $543M |
| 2027Q4E | $637M |
| 2028Q1E | $725M |
| 2028Q2E | $805M |
| 2028Q3E | $876M |
| 2028Q4E | $938M |
| 2029Q1E | $991M |
| 2029Q2E | $1.03B |
| 2029Q3E | $1.07B |
| 2029Q4E | $1.10B |
| 2030Q1E | $1.12B |
| 2030Q2E | $1.14B |
| 2030Q3E | $1.15B |
| 2030Q4E | $1.16B |
Assumptions & reasoning
- Cost of revenue on this line went from $24.6M to $101.4M year over year while revenue went from $33.0M to $126.0M. The cost curve is tracking capacity rather than consumption, which is what a build-ahead looks like.
- $14.5M of the quarter's revenue is zero-margin datacentre pass-through billed on behalf of specific customers. It is excluded from core revenue and it is why the GAAP cloud gross margin reads 19.5% against a core cloud gross margin of 41.8%.
- The segment margin starts at the 41.8% core cloud gross margin rather than the 19.5% GAAP figure, and glides toward 62% - just above the 60%-plus core gross margin target the CFO stated on the call, which is a target for the consolidated company and not guidance.
- Capex intensity starts at 180% of this line's revenue. Trailing-twelve-month capex of $746.5M was 251% of trailing cloud revenue and the June quarter alone spent $416.9M, so 180% is already a step down from the current build rate.
- $1.5bn of undiscounted future minimum lease payments on datacentre leases executed in the June quarter had not commenced at 2026-06-30 and sit off the balance sheet. They will land in cost of revenue as the capacity goes live, which is more cost arriving before its revenue.
- No seasonality array is set. Removing the trend with a centred four-quarter moving average does not shrink the Q4 factor, it inverts it, from a naive 1.273 to a detrended 0.879, and the window-to-window spread of 0.172 exceeds the signal of 0.121.
- The $25.4bn RPO is consolidated, includes variable pass-through consideration and cannot be split between the two lines, so it is used to shape this driver's decay and to anchor the Bull case, never as a cloud-only backlog.
Where each case comes from
Bear case — primary sources
The primary sources this case is built from — filings, calls and posts. Where they are claims by an interested party rather than disclosures, every number in the Bear column is what happens if they are taken at face value.
Q2 2026 Form 10-Q, MRA and Working Capital Loan risk
- Aug 12, 2026 If we fail to deliver such capacity on the stated timelines, or if we experience a certain level of failure with respect to our service levels, OpenAI has the right to terminate a portion or all of the agreement.
- Aug 12, 2026 If the MRA is terminated for any reason other than OpenAI's material uncured breach, or if certain trigger events occur ... we may be required to immediately repay the outstanding principal balance of the Working Capital Loan together with accrued interest.
Q2 2026 release, the cost of the build
Bull case — primary sources
The primary sources this case is built from — filings, calls and posts. Where they are claims by an interested party rather than disclosures, every number in the Bull column is what happens if they are taken at face value.
Q2 2026 Form 10-Q, remaining performance obligations
- Aug 12, 2026 The Company expects to recognize approximately 22% of this revenue over the initial 24 months ending June 30, 2028, 43% between months 25 and 48, and to recognize the remaining balance thereafter.
- Aug 12, 2026 In addition to the Committed Capacity, OpenAI has the option to purchase an additional 1.25GW of AI inference compute capacity (the "Additional Capacity") for deployment in tranches by the end of 2030 for up to a total of 2.0GW.
Feldman case — primary sources
The primary sources this case is built from — filings, calls and posts. Where they are claims by an interested party rather than disclosures, every number in the Feldman column is what happens if they are taken at face value.
Q2 2026 earnings call, growth beyond 2027
- Aug 12, 2026 With the benefit of that progress, we expect to more than triple our core revenues in 2027 and continue to grow at multiples in the years following.
- Aug 12, 2026 In total, over the last seven months, we have secured more than 600 megawatts of data center capacity that is either live now or will be delivered by the end of 2027.
From cash flow to fair value
| Present value of free cash flow, 18 quarters | -$2.40B |
| Terminal-year revenue | $4.81B |
| Terminal-year EBITDA | $1.50B |
| Exit multiple, on revenue | 3.0x |
| Terminal value | $14.44B |
| Discounted at 13.5% a year, terminal value becomes | $8.17B |
| Enterprise value | $5.76B |
| Net cash | $7.69B |
| Equity value | $13.45B |
| Shares | 0.24B |
| Fair value per share | $56.62 |
| Against the current price of $186.67 | -70% |
Both inputs are assumptions and neither is disclosed. 13.5% prices a pre-profit business with 76% of revenue in three customers and a build funded ahead of the revenue that serves it. The exit multiple carries most of the value, because at the 2026-08-24 close of $185.43 the enterprise value of roughly $36.4bn is already 41x the midpoint of the company's own FY2026 core revenue guide and about 14x a 2027 revenue equal to exactly three times that midpoint - so the stock is a bet on the years after the guide. 6.0x the terminal year is defensible against the steady state this model actually reaches: 60% segment margin less 20% overhead less capex that has fallen to 29% of revenue is a high-single-digit free-cash-flow margin in 2030, and the multiple prices the replacement-capex business beyond it rather than the terminal quarter. CoreWeave, the closest listed comparable in this repo, carries 3.0x - but it is a leased, debt-financed fleet with $29.5bn of net debt, where Cerebras is net cash $7.7bn and owns its silicon.
Read the other way round: at $186.67 the market is paying 14.4x terminal-year revenue, holding every other assumption on this page fixed. That is the number to argue about.
The projected path
| Quarter | Hardware | Cloud and other services | Revenue | YoY | EBITDA | Capex | FCF | R40 | PV of FCF |
|---|---|---|---|---|---|---|---|---|---|
| 2026 Q3E | $56M | $189M | $245M | +81% | $38M | $290M | -$252M | -22 | -$244M |
| 2026 Q4E | $57M | $266M | $323M | +89% | $57M | $346M | -$289M | -1 | -$271M |
| 2027 Q1E | $58M | $354M | $412M | +113% | $81M | $393M | -$312M | +37 | -$283M |
| 2027 Q2E | $59M | $448M | $506M | +181% | $108M | $427M | -$319M | +118 | -$281M |
| 2027 Q3E | $59M | $543M | $603M | +146% | $138M | $450M | -$312M | +94 | -$266M |
| 2027 Q4E | $60M | $637M | $696M | +115% | $169M | $461M | -$292M | +73 | -$242M |
| 2028 Q1E | $60M | $725M | $784M | +90% | $199M | $463M | -$264M | +57 | -$211M |
| 2028 Q2E | $60M | $805M | $865M | +71% | $228M | $458M | -$230M | +44 | -$178M |
| 2028 Q3E | $60M | $876M | $936M | +55% | $255M | $448M | -$193M | +35 | -$145M |
| 2028 Q4E | $60M | $938M | $998M | +43% | $280M | $436M | -$156M | +28 | -$114M |
| 2029 Q1E | $60M | $991M | $1.05B | +34% | $302M | $423M | -$121M | +22 | -$85M |
| 2029 Q2E | $59M | $1.03B | $1.09B | +27% | $321M | $409M | -$88M | +19 | -$60M |
| 2029 Q3E | $59M | $1.07B | $1.13B | +21% | $337M | $395M | -$58M | +16 | -$38M |
| 2029 Q4E | $59M | $1.10B | $1.16B | +16% | $351M | $383M | -$31M | +13 | -$20M |
| 2030 Q1E | $58M | $1.12B | $1.18B | +12% | $363M | $371M | -$8M | +12 | -$5M |
| 2030 Q2E | $58M | $1.14B | $1.20B | +10% | $373M | $361M | $10M | +10 | $6M |
| 2030 Q3E | $58M | $1.15B | $1.21B | +7% | $381M | $352M | $25M | +9 | $14M |
| 2030 Q4E | $57M | $1.16B | $1.22B | +5% | $387M | $343M | $37M | +8 | $21M |
Every row is projected. A year-over-year change is shown only where the quarter it compares against exists — an em dash means there is no comparable quarter, not a flat year.
Model revisions
Assumptions are marked to reality as each quarter prints. Every change is appended here, with the fair value the model produced at the time, so the model's own history stays visible.
| Date | Changed | Fair value then | Note |
|---|---|---|---|
| 2026-08-27 | all | $139.08 | First build, on the 2026 Q2 basis. Calibrated to the raised FY2026 core guide and the 2027 tripling converted to GAAP through the disclosed warrant and pass-through bridge, with the decay shaped by the two buckets of the $25.4bn RPO schedule. |