Toronto · Independent Shariah screening · Updated daily
Market news · Researched from public market reporting · Content-only (no affiliate programs)

Cerebras Surged 15% on a Microsoft Deal and a Raised Forecast — but Its Halal Screen Hinges on the Cash Pile (2026)

Friday's biggest large-cap winner was not a household name. Cerebras Systems, the wafer-scale artificial intelligence chipmaker that listed on Nasdaq in May at 185 dollars a share and had fallen 47 percent from that price by Thursday's close, reported third-quarter results after the close on Thursday, beat estimates, raised its full-year forecast, and announced it had signed Microsoft as a cloud customer. The shares rose about 17 percent after hours, added about 15 percent in early trading on Friday, then eased to about 104 dollars by late morning, still up about 7 percent on the day. This article lays out the numbers, the Microsoft deal, the OpenAI concentration behind the backlog, the IPO journey that brought the stock to this point, and the halal read, which is unusual: Cerebras has never been screened by this site, and the raw numbers from its own filings say the debt gate passes, the cash gate is the question, and the income gate cannot be verified yet, while the two rating services split, with Zoya calling it compliant and Musaffa calling it not halal.

Updated November 2026 · Reviewed by the Canadian Halal Investor team

What happened: the beat, the raise, and the Microsoft deal

Cerebras reported its third-quarter 2026 results after the market closed on Thursday 12 November, for the quarter ended 30 September, and the release carried three pieces of news at once. Core revenue, the company's non-GAAP measure, was 230.9 million dollars, up 70 percent from a year earlier and above the 214.7 million dollar consensus compiled by LSEG, and above the company's own August guidance of 214 to 216 million dollars. Adjusted earnings were 12 cents a share, against an expected loss of 14 cents. And the company raised its full-year 2026 core revenue forecast to 940 to 960 million dollars, from 880 to 890 million dollars, the second raise of the year after the June range of 855 to 865 million dollars was lifted in August. The release also announced an important cloud deal with a top-three hyperscaler, without naming the customer, and chief executive Andrew Feldman said in an interview that the customer is Microsoft, that the deal was signed in October, and that revenue should begin to show up in the first quarter of 2027, with more details to come next year.

The market's answer was immediate and large. The stock rose about 17 percent in extended trading on Thursday to about 114 dollars, after closing at 97.75 dollars, and added about 15 percent in early trading on Friday to about 113 dollars, before easing to about 104 dollars by late morning, still up about 7 percent on the day. The move is the biggest confirmed earnings reaction of the week, and it arrived into a market that has spent the week selling the artificial intelligence trade: Nvidia reversed lower on Friday after its founder's resignation, AMD fell 4.7 percent, and the Nasdaq was down at mid-morning. Cerebras was the exception, and the exception needs explaining, because a 15 percent rise in a stock that has fallen 47 percent from its IPO price is a different event from a 15 percent rise in a stock near its highs. The first is a relief rally in a beaten-down name. The second is a breakout. This one is the first, and the halal read has to start from that.

There is also a timing point worth stating plainly. The main IPO lock-up expired on Monday 9 November, three days before the results, releasing the remaining restricted shares into a stock already down about half from its offering price. The results landed on the first trading day after the lock-up expiry window closed, and the beat, the raise and the Microsoft name arrived together into a shareholder base that had just absorbed the largest supply event of the company's short public life. That does not make the numbers better or worse. It makes the reaction harder to read, because a stock that has just absorbed a lock-up expiry and then jumps 15 percent on earnings is telling two stories at once, and the second one, the supply story, is finished, while the first one, the demand story, is the one the next quarter has to confirm.

The fourth-quarter guidance is the steepest step in the story, and it is the number the market is really buying. Cerebras guided fourth-quarter core revenue to 310 to 330 million dollars, against about 271 million dollars expected, with core gross margin of 47 to 49 percent and core operating margin of about breakeven, between a 1 percent loss and a 1 percent profit. The implied fourth quarter is up 34 to 43 percent sequentially from the third quarter's 230.9 million dollars, and the full-year range of 940 to 960 million dollars implies a year that ends roughly three times the size of the one that started it, with the chief executive telling investors he expects to triple revenue again in 2027. Those are the numbers behind the 15 percent. They are also the numbers that have to be delivered from data-centre capacity that does not fully exist yet, against a backlog that belongs mostly to one customer, and from a cash pile that earns interest while the company burns it. Each of those threads gets its own section below, because each of them is where the story can break.

The numbers: a cloud business now, with a profit that came from the balance sheet

The headline numbers are strong, and the composition underneath them is the story. GAAP revenue was 210.8 million dollars, up 55 percent from 135.7 million dollars a year earlier, a record for the company. Core revenue, which strips out data-centre pass-through revenue and adds back the amortisation of customer warrants, was 230.9 million dollars, up 70 percent. The split inside the quarter is the important part: hardware revenue fell to 68.9 million dollars from 96.8 million dollars a year earlier, while cloud and other services revenue rose 265 percent to 141.9 million dollars on a GAAP basis and 272 percent to 144.9 million dollars on a core basis. Cloud is now 63 percent of core revenue, up from 29 percent a year earlier. Cerebras is no longer mainly a company that sells chips. It is mainly a company that rents out computing power, and the chips are the means to that end. That changes the business, the margins, and the screen, because a cloud business carries a balance sheet a chip seller does not.

The profit line needs the same treatment, because the headline and the operations point in opposite directions. The company reported a GAAP net loss of 161.2 million dollars, or 67 cents a share, against a 22.2 million dollar loss a year earlier, and the loss includes about 155.1 million dollars of stock-based compensation in the quarter. On the company's core measure it reported net income of 29.8 million dollars, against a core loss of 10.9 million dollars a year earlier, its first core profit. But core operations lost 33.1 million dollars in the quarter, against a 15.1 million dollar core operating loss a year earlier. The difference between a core operating loss of 33.1 million dollars and core net income of 29.8 million dollars is other income, net of 63.9 million dollars in the quarter, against 5.8 million dollars a year earlier. The release does not split that line out, and the split matters, because a company with 7.8 billion dollars of cash, restricted cash and short-term investments earns yield on that pile, and yield is interest income. Strip out the other income and the core business lost money in the quarter. That does not make the beat false. It makes it a balance-sheet profit, not an operating one, and the distinction is the one the halal screen cares about most, because the income gate measures exactly that yield.

The nine-month picture is rougher, and it is worth stating because it is the context the quarterly beat sits inside. Revenue for the nine months to 30 September was 584.3 million dollars, up 72 percent from 338.5 million dollars a year earlier, while the net loss was 625.8 million dollars, against net income of 263.4 million dollars a year earlier. The year-earlier profit is not comparable, because it included a 363.3 million dollar one-off gain from extinguishing a forward contract liability, and the 2026 loss includes 541.7 million dollars of stock-based compensation for the nine months. Research and development spending tripled to 573.5 million dollars for the nine months, from 170.8 million dollars a year earlier. Operating cash flow was negative 388.9 million dollars for the nine months, and purchases of property and equipment reached 1.05 billion dollars, the cost of building the data-centre capacity the backlog requires. The company is growing fast, spending faster, and paying for the gap with the 6.23 billion dollars of net proceeds from its May IPO and 1.01 billion dollars from a Series H preferred share sale. That is the funding model in one sentence: public-market cash, spent on capacity, while the operating business is still loss-making on a GAAP basis.

Two more numbers belong in this section because they frame the fourth quarter. Core gross margin was 41.0 percent in the quarter, up from 40.6 percent in the second quarter, and the company guided it to 47 to 49 percent for the fourth quarter and 42 to 44 percent for the full year. The margin is improving as cloud revenue, which carries higher margins than hardware, becomes the larger share of the mix. And the balance sheet at 30 September shows total assets of 11.94 billion dollars, up from 2.33 billion dollars at 31 December 2025, almost entirely because of the IPO: cash and cash equivalents of 3.07 billion dollars, restricted cash of 444.0 million dollars, and investments of 4.29 billion dollars, a combined liquidity pile of 7.8 billion dollars, against total liabilities of 2.78 billion dollars and equity of 9.16 billion dollars. The pile is the company's runway, and it is also, as the screen section below shows, the company's screening problem.

The Microsoft deal: a hyperscaler, signed in October, paying from 2027

The Microsoft deal is the most consequential line in the release, and it is also the least detailed, which is why it needs careful handling. The release announced an important cloud deal with a top-three hyperscaler, without naming the customer and without giving a value, a duration, or a capacity figure, and said more will be shared next year. Feldman said in an interview that the customer is Microsoft, that the deal was signed in October 2026, and that he expects revenue to begin showing up in the first quarter of 2027. That is the whole of what is confirmed: a customer, a signing month, a revenue start, and a promise of details next year. There is no contract value, no megawatt figure, and no disclosure of whether the deal is a capacity reservation, a minimum commitment, or a pay-as-you-go arrangement. The market priced it as a validation anyway, and the validation is real even without the numbers, because Microsoft is the largest buyer of artificial intelligence computing in the world and it does not sign cloud deals with chip startups lightly. But the size of the validation is unknown, and a halal investor should not treat an unnamed-value deal as a known quantity.

The Microsoft deal is the second hyperscaler-scale commercial win of the year, and the pattern matters. In June, Cerebras signed a commercial agreement with Amazon Web Services and gave an Amazon affiliate a warrant to buy up to 2,696,678 shares at 100 dollars each, according to its quarterly report. Oracle is integrating Cerebras' fast inference into Oracle Cloud Infrastructure, with availability in 2027. The release also said the company won commercial deals with two new hyperscalers in the quarter, and that its pipeline for high-speed artificial intelligence inference increased by more than ten times in the quarter. On the enterprise side, Cerebras' high-speed inference will power the new Bilt Neighborhood OS platform and the Bilt Concierge artificial intelligence assistant, serving eight million consumers and 50,000 businesses. The customer base is broadening, and that broadening is the answer to the concentration question in the next section. But broadening is a direction, not a destination: the new names are real, and they are small next to the one customer that still dominates the backlog.

The technology behind the deals is worth one paragraph, because it is what the customers are actually buying. Cerebras builds the Wafer Scale Engine, the world's largest computer chip, and the CS systems built around it, and it sells access to that hardware as a cloud service for artificial intelligence inference, the work of running trained models rather than training them. In August the company unveiled its next-generation CS-4 system, which it says delivers up to 30 times faster inference, and previewed the CS-5, targeted for early summer 2027 and designed to deliver up to 10,000 tokens per second per user, with the Wafer Scale Engine 5 already back from the fab and in system testing less than five months after tape-out. Cerebras was a day-zero launch partner for Kimi K3.1, Moonshot AI's 2.8 trillion-parameter frontier-class open-weight model, which the company says runs at up to 2,000 tokens per second on its hardware, the fastest in the industry. The claim that matters for the business is speed per megawatt: the company's disaggregated inference offering, the first production-scale version in the market, is designed to deliver up to five times the number of tokens a data centre can produce for the same power. In a market where power is the binding constraint on artificial intelligence growth, speed per megawatt is the pitch, and Microsoft, Amazon and Oracle are the customers who would know whether it is true.

The capacity to deliver on those deals is the other half of the story, and it is where the execution risk sits. Cerebras said it has secured 700 gross megawatts of data-centre capacity for delivery by the end of 2027, signed contracts for additional capacity in Norway and Finland, and expects to exit 2026 with more than ten times the manufacturing capacity it had at the start of the year, scaling with Flex and Sanmina. The company also said it plans to add factories and further increase capacity in 2027. That is a buildout on the scale of a utility, funded from the IPO proceeds, and it is the reason the cash pile exists and the reason it is being spent. The question for the next two quarters is not whether the demand is there. The backlog says it is. The question is whether the capacity arrives on the schedule the guidance assumes, because the fourth-quarter guide of 310 to 330 million dollars in core revenue depends on it, and the OpenAI experience, in the next section, shows what happens when it does not.

The OpenAI concentration: a third of revenue and most of the backlog

Behind the Microsoft headline sits a concentration that is larger than the deal itself. In December 2025 Cerebras agreed to provide OpenAI with 750 megawatts of computing capacity over several years, in a multi-year agreement described by Reuters as worth 20 billion dollars, and OpenAI accounted for 34 percent of Cerebras' third-quarter revenue, according to its quarterly report. Remaining performance obligations, the contracted revenue not yet recognised, stood at 26.1 billion dollars at 30 September, up from 25.4 billion dollars at 30 June. The backlog is roughly 28 times the company's full-year 2026 revenue guidance, which is the kind of number that makes a growth story and also the kind that makes a concentration problem, because a backlog that large, attached mostly to one customer, is a promise that has to be kept by building data centres, and building data centres is where this story has already slipped.

The slip is disclosed, not rumoured, and it is worth reporting precisely. Cerebras disclosed in its quarterly report that delivery delays under the OpenAI agreement could trigger service credits or partial contract termination, and on the earnings call Feldman said the delays stemmed from industry-wide shortages of land and power, rather than chip manufacturing constraints. The company expects delivery schedules to normalise in the second half of 2027. That is a year away, and it is the reason the fourth-quarter cloud revenue line is the one to watch: the company has said new data-centre capacity will not be online in time to lift it, and the 310 to 330 million dollar fourth-quarter guide depends on hardware and on the capacity that is already running. A backlog of 26.1 billion dollars is not revenue. It is a list of promises, and the promises are being kept late, for reasons the company says are outside its control, which may be true and is still a risk.

The concentration question has two sides, and both belong in the article. On one side, a 20 billion dollar multi-year agreement with the most important artificial intelligence company in the world is the strongest demand signal a chip startup can have, and it is the reason the backlog exists at all. On the other side, a company that gets a third of its revenue and most of its backlog from one customer is a company whose fortunes turn on one relationship, one delivery schedule, and one counterparty's own funding and strategy. The Microsoft deal, the Amazon agreement, the Oracle integration and the Bilt platform are all real diversification, and they all point in the right direction. But none of them is yet large enough to change the shape of the backlog, and the shape of the backlog is the shape of the risk. A halal investor who screens the company has to screen the business as it is, not as it will be when the diversification arrives, and as it is, the business is a bet on one customer, one technology, and one buildout, in that order.

There is also a governance thread inside the concentration, and it is the warrant. The Amazon affiliate's warrant for up to 2,696,678 shares at 100 dollars each, and the customer warrants more broadly, are the mechanism by which Cerebras' largest customers also become shareholders, and the amortisation of those warrants is the reason core revenue is higher than GAAP revenue by 20.1 million dollars in the quarter. Warrants align a customer with a supplier, and they also dilute existing shareholders when they are exercised, and they are a reminder that the cloud deals in this story are not arm's-length purchases of a commodity. They are strategic relationships, negotiated with equity attached, and the equity is part of the price. The screen does not measure warrants. It measures debt, cash and income. But the warrants are part of the business a halal investor is being asked to understand, and they belong in the article for that reason.

The IPO journey: from 311 dollars on day one to 78 dollars last week

Friday's 15 percent rise is a large move, and it is a small move against the stock's own history. Cerebras listed on Nasdaq on 14 May 2026 at 185 dollars a share, and closed its first session at 311.07 dollars, a 68.1 percent premium to the offering price. By Thursday's close at 97.75 dollars, the stock was down 47.2 percent from its IPO price and 68.6 percent from its first-day close. It reached an intraday low of 77.84 dollars on 5 November, and even after Friday's rise to about 104 dollars by late morning, it remains about 44 percent below the price the public was first offered. The first-day close of 311 dollars is the number that frames everything else, because it is the price the market put on the company when enthusiasm was highest, and the 97.75 dollar close on Thursday is the price it put on the company when the enthusiasm had gone, and the gap between them is the market's verdict on six months of execution, spending, and dilution.

The fall has a specific cause, and it is not a mystery. Investors grew wary of the massive infrastructure spending required to support artificial intelligence growth, and of a business that was spending 1.05 billion dollars on property and equipment in nine months, tripling its research and development budget, and still losing money on a GAAP basis, while carrying a backlog that depended on data centres that did not yet exist. The main IPO lock-up expired at 6 a.m. Eastern on Monday 9 November, releasing the remaining restricted shares, and the stock traded at 93.79 dollars that morning, up 4.58 percent, about 49.3 percent below its IPO price. On Tuesday 10 November it closed at 93.66 dollars, down 3.1 percent, after Morgan Stanley cut its price target to 178 dollars from 279 dollars. The stock had fallen for most of the six months since listing, and the third-quarter results were the first event in that period that gave buyers a reason to come back. The question is whether the reason is durable, and the answer to that question is the fourth-quarter guide, not the Friday morning price.

The IPO journey also matters for the screen, in a way that is easy to miss. A company that lists at 185 dollars, closes day one at 311 dollars, and then falls to 98 dollars has a market capitalisation that has fallen by roughly two-thirds from its first-day peak, while its cash pile has stayed roughly where the IPO left it, at 7.8 billion dollars. The debt and cash ratios in a Shariah screen are measured against market capitalisation, so when the price falls, the ratios rise, and when the price rises, they fall. A stock that has halved is a stock whose ratios have roughly doubled, and a stock that jumps 15 percent in a morning is a stock whose ratios have just fallen by about 13 percent. That mechanical relationship is the subject of the cash-pile section below, and it is the reason the screen result for Cerebras cannot be stated once and left alone. It moves with the price, and the price has moved a lot.

The comparison with the other artificial intelligence infrastructure name this site has covered is instructive, because it shows the two ways the same business model can fail a screen. CoreWeave surged on a 168 percent revenue beat on 11 November and still fails this site's screen at the debt gate, with debt of 74.7 percent of market capitalisation and about 47 billion dollars of borrowings. Cerebras, by contrast, has almost no debt, and fails, if it fails, at the other end of the balance sheet, on the cash pile. The two companies do the same thing, rent out artificial intelligence computing, and they fund it in opposite ways: one with debt, one with equity raised in an IPO. The screen catches both, for different reasons, and the lesson is that the funding model is part of the screening result, not separate from it. A company can be debt-free and still fail, if the money it raised sits in interest-bearing securities earning yield.

Friday's context: a cooler PPI, a falling Nvidia, and a Microsoft outage

Friday's session was not a broad rally, and Cerebras rose into a market that was otherwise going down. October producer prices rose 0.3 percent, below the 0.4 percent economists polled by Dow Jones expected, though the annual rate climbed to 6.5 percent, the highest since November 2022. The 10-year Treasury yield was 5.16 percent at 9:22 a.m. Eastern, up from 5.10 percent on Thursday. At 9:38 a.m. the S and P 500 was down 0.09 percent at 7,585.65, the Nasdaq Composite was down 0.38 percent at 25,269.40, and the Dow was up 0.27 percent at 52,436.17. The market opened higher on the cooler producer price print and then faded, and the fade is the context for the Cerebras move: a stock that rises 15 percent while the Nasdaq falls is a stock-specific story, not a market story, and the stock-specific story is the Microsoft deal and the raised forecast, not the inflation print.

The rest of the artificial intelligence trade went the other way, and the split is the clearest signal of the day. Nvidia opened at 167.22 dollars, up about 0.6 percent from Thursday's close of 166.19 dollars, and then slid to 163.81 dollars, down 1.4 percent, putting it on course for a sixth straight weekly loss, after shedding nearly 500 billion dollars in market value this week, according to MarketWatch. AMD fell 4.7 percent to 557.71 dollars after rising on Thursday, and Super Micro slipped 0.6 percent. Microsoft, the customer at the centre of the Cerebras story, fell 0.9 percent to 518.10 dollars, after its Microsoft 365 outage, which knocked out Outlook, Teams and Copilot for many users, ran into a second day. The market sold the established artificial intelligence names and bought the challenger, which is the same rotation that ran on Thursday, when AMD and Broadcom rose as Nvidia fell. The rotation is a bet that the buildout continues and that the winners inside it change, and Cerebras is the name the rotation bought on Friday.

The other large mover of the morning was Modine Manufacturing, the data-centre cooling equipment maker, which rose more than 25 percent after Bloomberg reported that Samsung Electronics is in talks to buy it, and the pairing with Cerebras is worth noting because the two moves share a theme. Both are bets on the physical infrastructure of artificial intelligence: one on the chips that do the computing, one on the cooling that keeps them running. The market is buying the buildout, not the software, and it is buying it in the names that are furthest from the governance questions that have dominated the week. That is the week's pattern in one sentence: the demand for the buildout is real, the leadership of its biggest name is not, and the money is looking for the next name that can take the work. Cerebras is that name today, and the question is whether it can hold the work, or whether Friday is the high-water mark of a relief rally.

For the halal investor, Friday's context is also a rates story, and the rates story is the screen story. The producer price print was cooler than expected on the month and hotter than expected on the year, and the 10-year yield above 5 percent is the level at which the interest income of cash-rich companies becomes a screening question in its own right. Tuesday's hot consumer price print made a December rate hike near-certain, and the December decision, on 8 to 9 December, is the next macro date that matters for the screen, because higher rates raise the interest income of every company holding a large cash pile, and interest income is exactly what the third gate measures. Cerebras, with 7.8 billion dollars of liquidity, is one of the companies most exposed to that mechanism, and the exposure grows with every rate hike. The market's Friday fade is a rates story. The Cerebras screen is a rates story too, and they are the same story.

The screen results: no result yet, and three gates pointing in different directions

The first thing to say about Cerebras and the halal screen is that there is no screen result yet. Cerebras does not appear in this site's registry of 939 screened companies, and there is no screener page for it, because it listed only in May 2026 and the pipeline has not reached it. That is not a pass, and it is not a fail. It is an absence, and a halal investor should treat it as one. What follows is not a verdict. It is the raw arithmetic from the company's own filings, gate by gate, with the parts that can be computed and the parts that cannot, so that a reader can see exactly where the screen would land and exactly where it cannot land yet. The method is the one set out in the screening methodology, and the full registry is on the screener hub.

Gate one, the business-activity gate, passes on the facts available. Cerebras designs wafer-scale artificial intelligence processors and sells access to artificial intelligence computing as a cloud service, to enterprises, research institutes and governments, across healthcare, energy, financial services and scientific computing. That is a permissible business, and no prohibited revenue stream, alcohol, gambling, conventional finance, pork, or weapons manufacturing, decides the result at gate one. The customer list includes governments, and the company's own description says governments choose its systems to run their workloads, but a government customer is not a weapons business, and no documented military or weapons revenue stream has surfaced in the review. The gate-one question is therefore straightforward, and it is the only gate that is.

Gate two has two parts, debt and cash, and they point in opposite directions. The debt part passes comfortably. The interest-bearing debt on the 30 September balance sheet is the working capital loan, 706.5 million dollars current and 122.6 million dollars non-current, a total of 829.1 million dollars, and the 850 million dollar revolving credit facility is unused. Finnhub put the market value at about 22.6 billion dollars at Friday's late-morning price of about 104 dollars, and the same share count implies about 21.2 billion dollars at Thursday's close of 97.75 dollars. The debt ratio is therefore about 3.7 percent at Friday's price and about 3.9 percent at Thursday's close, against a ceiling of roughly 33 percent, with a reported disagreement between providers over whether it is 30 percent or about 33 percent. At either ceiling and either price, the debt gate passes by a wide margin. The cash part does not. Cash and cash equivalents of 3.07 billion dollars plus investments of 4.29 billion dollars is 7.36 billion dollars, and that pile is about 32.5 percent of the market value at Friday's late-morning price and about 34.7 percent at Thursday's close. That is above the 30 percent ceiling at both prices, above the roughly 33 percent ceiling at Thursday's close, and just inside the roughly 33 percent ceiling at Friday's price. Including the 444.0 million dollars of restricted cash, the pile is 7.8 billion dollars, about 34.5 percent at Friday's price and about 36.8 percent at Thursday's close. On the cash gate, Cerebras is either failing or a rounding error from failing, and the answer depends on the price on the day and on which ceiling a reader applies.

Gate three, the non-compliant income gate, cannot be verified from the release, and that is the honest answer. The income gate is measured on interest income against revenue, with a 5 percent ceiling, and the third-quarter release reports only other income, net of 63.9 million dollars, without splitting out interest income. The nine-month figure is 93.4 million dollars, also unsplit. Zoya's page gives the fiscal 2025 inputs, revenue of 509.991 million dollars and interest income of 26.802 million dollars, and describes the ratio as 5.26 percent. On the revenue denominator the ratio is 5.26 percent, above the 5 percent ceiling; on the combined revenue-plus-interest denominator it is 4.99 percent, below it. Zoya rates the stock compliant, and the page does not say which denominator produced its verdict. The 2025 numbers are also the wrong year for the question that matters now, because the cash pile in 2025 was 701.7 million dollars and the pile today is 7.8 billion dollars, eleven times larger, and the third quarter's other income of 63.9 million dollars, if most of it is yield on that pile, would put the 2026 run-rate far above 5 percent of revenue. That is an inference from the size of the pile, not a verified figure, and this site does not publish inferences as screen results. The income gate will be decided by the fiscal 2026 annual report, due in early 2027, when the interest income line is filed. Until then, the gate is open, not passed.

Two flags belong in the article, reported as context and not as a screen change. Cerebras does not appear on the BDS movement's published boycott, divestment or pressure lists, as checked in public search on 13 November 2026. That flag is informational only. It does not change any screening result, and it is not a religious ruling. The site's moral screen, which checks documented business ties to the Israeli and United States militaries and to weapons manufacturing, publishes no findings for Cerebras. A row with no finding means none surfaced in the review, not proof of none, and the screen results stand on the financial gates alone. Cerebras pays no dividend, so there is nothing to purify, and the purification question does not arise for this name. The screen, as it stands today, is: business gate passes, debt gate passes, cash gate borderline to failing depending on price and ceiling, income gate unverifiable, and no result published by this site yet.

The cash-pile paradox: the rally itself moves the verdict

The cash gate deserves its own section, because it is the most interesting screening question in this story and the one most likely to be misunderstood. The pile exists because the IPO worked: Cerebras raised 6.23 billion dollars of net proceeds in May, plus 1.01 billion dollars from a Series H preferred share sale, and the money is sitting in cash, restricted cash and short-term investments while the company spends it on data centres, manufacturing capacity and research. That is what IPO money is for. It is also, under the screen, a problem, because the cash and the investments earn interest, and interest is the income the third gate measures, and the pile itself, measured against market capitalisation, is the quantity the second gate measures. The company raised the money to build the business, and the screen measures the money as if it were the business. That is not a flaw in the screen. It is the screen doing what it was built to do: ask whether a company's balance sheet looks like a lender's, and a company holding 7.8 billion dollars of interest-bearing securities looks, on the numbers, like one.

The paradox is that the rally moves the verdict, and the verdict moves with the price in the opposite direction from the one a buyer would expect. A cash pile of 7.36 billion dollars needs a market value of about 24.5 billion dollars to sit at the 30 percent ceiling, and about 22.3 billion dollars to sit at the roughly 33 percent ceiling. At Thursday's close the market value was about 21.2 billion dollars, below both, so the ratio was above both ceilings. At Friday's late-morning price the market value was about 22.6 billion dollars, above the 33 percent line and below the 30 percent line, so the ratio was inside one ceiling and outside the other. A further 10 percent rise in the share price would take the market value to about 24.9 billion dollars and the ratio to about 29.6 percent, inside both ceilings. A 10 percent fall would take it to about 20.3 billion dollars and the ratio to about 36.2 percent, outside both. The screen result for Cerebras is therefore not a fact about the company. It is a fact about the company at a price, and the price changes every morning. A halal investor who checks the screen on a Friday and finds a pass may find a fail on a Monday, with nothing about the company having changed except the denominator.

That price-sensitivity is not unique to Cerebras, but it is unusually large here, because the pile is unusually large relative to the company. The contrast with the names near the ceiling shows why. Regeneron sits at 4.998 percent non-compliant income, a rounding error below the 5 percent ceiling, and the ratio that threatens it is the income gate, which moves with rates, not with the share price. Cerebras is threatened by both: the income gate moves with rates, because higher rates raise the yield on the pile, and the cash gate moves with the price, because the price is the denominator. The December rate decision, on 8 to 9 December, after Tuesday's hot consumer price print made a hike near-certain, threatens the first. A bad week for the stock threatens the second. The disciplined response is to treat the screen as a moving picture, not a snapshot, and to recheck it at the refresh, when the annual report lands and the ratios are measured on audited figures rather than on a Friday morning's price.

There is a second, quieter paradox inside the first, and it is the one the earnings call did not address. The company is spending the pile as fast as it can, on property and equipment, on data-centre capacity, on research, and the spending is the reason the pile will shrink. Operating cash flow was negative 388.9 million dollars for the nine months, and purchases of property and equipment were 1.05 billion dollars. If the spending continues at that pace, the pile falls, the ratios fall with it, and the screen improves, not because the business changed but because the money was spent. If the revenue triples in 2027 as the chief executive expects, the income ratio falls too, because the denominator grows. The screen for Cerebras is therefore likely to look better in a year than it does today, on both gates, if the company executes. That is a reason to watch the name, not a reason to buy it, and the distinction is the whole of the framework in the next section but one. A screen that will probably pass next year is not a screen that passes today, and a halal investor does not buy a probable pass.

The provider split: Zoya says compliant, Musaffa says not halal

The two rating services that cover Cerebras disagree, and the disagreement is worth reporting in full, because it is the clearest illustration in this story of why the site reports provider disagreement and never picks a position. Zoya's page for Cerebras, updated 13 November 2026, says the stock is Shariah-compliant and therefore considered halal to invest in, based on the company's most recent financial reports, under AAOIFI guidelines. It gives the fiscal 2025 inputs, revenue of 509.991 million dollars and interest income of 26.802 million dollars, and describes the ratio as 5.26 percent. Musaffa's page for Cerebras, as of November 2026, says the company is classified as not halal under its AAOIFI methodology, based on an assessment of the company's business activities and financial ratios against Islamic investment guidelines, and it does not publish the ratios behind its verdict on the public page. One service says compliant. One says not halal. Both say they apply AAOIFI. The disagreement is the fact, and the fact is reported here without resolution, because there is no resolution available from the public pages.

The disagreement is not a contradiction, and it is worth explaining why, because the explanation is the methodology itself. The two services use different data vintages, different denominators, and different treatments of the same numbers. Zoya's inputs are fiscal 2025, the last full year, when the cash pile was 701.7 million dollars and the interest income was 26.8 million dollars. Musaffa's assessment is dated November 2026, after the third-quarter results, when the cash pile is 7.8 billion dollars and the third quarter's other income alone was 63.9 million dollars. A screen run on 2025 data and a screen run on 2026 data are screens of two different companies, and they can legitimately disagree. The denominator question adds a second layer: on the revenue denominator the 2025 ratio is 5.26 percent, above the ceiling, and on the combined denominator it is 4.99 percent, below it, and the two services do not publish which denominator they use. The honest read is that the disagreement is real, it is driven by data vintage and methodology, and it will probably resolve itself when the fiscal 2026 annual report is filed, because the 2026 numbers are large enough that the denominator question stops mattering.

What a reader should do with the split is the same as what this site does with every provider split: read both, note the vintage, and do not treat either as a ruling. Zoya's compliant rating is a rating of the 2025 numbers. Musaffa's not-halal rating is a rating of the company as it stands in November 2026. Neither is this site's screen, and this site has not published one. The contrast with the names where the providers agree is instructive: Cloudflare fails at 6.05 percent non-compliant income, and Zoya independently rates it not compliant on the same figures, so the agreement reinforces the result. Here the providers disagree, and the disagreement is itself the information, because it tells a reader that the name is on the edge, that the edge is moving, and that the next filing will decide it. A split is not a tie to be broken. It is a warning to wait for the filing.

One more thing belongs in this section, and it is about the site's own practice. This site's screeners are measured on audited annual figures, not on quarterly releases and not on prices, and that is why there is no screen result for Cerebras yet: there is no annual report on file, because the company listed in May and its first annual report as a public company will not arrive until early 2027. The quarterly numbers in this article are labelled arithmetic, computed from the 10-Q and the earnings release, and they are presented as such, not as a screen result. That distinction matters, because a reader who confuses labelled arithmetic with a verdict will either buy a name that has not passed or sell a name that has not failed, and both mistakes come from the same confusion. The arithmetic says the debt gate passes, the cash gate is on the line, and the income gate is unverifiable. The verdict says nothing yet, because there is no verdict to give.

How a halal investor should approach an unscreened IPO name

A 15 percent rise in an unscreened IPO name triggers a specific set of instincts, and they are worth naming before acting on any of them. The first is to buy the story, on the theory that a Microsoft deal and a raised forecast are enough. The second is to buy the dip that is no longer a dip, on the theory that a stock 47 percent below its IPO price is cheap. The third is to treat Zoya's compliant rating as a pass, on the theory that a rating service has done the work. None of the three is a framework. A framework is a set of questions, answered in order, before any order is placed, and for an unscreened IPO name the questions are these.

First, no screen result means no pass. Cerebras has not been screened by this site, and a rating from one service, against a not-halal rating from another, is not a substitute for a screen. A halal investor who buys on Zoya's compliant rating has let one provider do the screening, and a halal investor who sells on Musaffa's not-halal rating has done the same thing in the other direction. The screen is the first question, and a provider rating is an input to it, not a replacement for it. Second, wait for the annual report. The fiscal 2026 10-K, due in early 2027, will give the interest income line, the full-year revenue, and the year-end balance sheet, and those three numbers will decide the income gate and settle the cash gate. Until then, the screen is arithmetic on quarterly figures, and quarterly figures are not the basis for a verdict. Third, do not chase a 15 percent move. A stock that rises 15 percent in a morning and eases to 7 percent by late morning is a stock whose price is moving faster than its facts, and the facts that matter, the fourth-quarter delivery, the Microsoft revenue in the first quarter of 2027, the OpenAI delivery schedule in the second half of 2027, all arrive later. The disciplined position is to note the screen arithmetic as it stands, note that a rally moves the cash ratio in the investor's favour and does not change the income question at all, and wait for the filing.

Fourth, watch the concentration, not the headline. A third of the quarter's revenue and most of a 26.1 billion dollar backlog from one customer is a risk the screen does not measure, because the screen measures what a company does and how its balance sheet looks, not who its customers are. A halal investor who understands the screen knows where its boundary lies, and the concentration sits outside it, which means it has to be watched separately, as a qualitative layer on top of the screen. Fifth, remember that you do not use margin. A halal portfolio carries no margin debt, which means no lender can liquidate it on a bad morning, and a 15 percent rise that eases to 7 percent by late morning is a price, not a forced sale, in either direction. That is the practical difference the screen's debt gate buys you: the companies you hold do not borrow heavily, and neither do you. Sixth, recheck at the refresh. If the fourth-quarter guide is met, if the Microsoft revenue arrives in the first quarter of 2027, if the OpenAI deliveries normalise in the second half of 2027, the next screen refresh will show it in the numbers. The framework is the position: screen first, wait for the filing, do not chase the move, watch the concentration, no margin, and wait for the refresh.

There is one more question the framework has to answer, and it is the one the cash-pile paradox raises directly. If the screen improves as the company spends the pile and grows the revenue, is the right move to buy now, before the screen catches up? The answer is no, and the reason is the same as the reason for the framework itself: the screen is a gate, not a forecast, and a gate that will probably open next year is not a gate that is open today. A halal investor does not buy a probable pass, because the probability is not the compliance, and the compliance is the only thing the screen certifies. The company may execute, the revenue may triple, the pile may shrink, and the screen may pass, and all of that may happen, and none of it has happened yet. Read the filing when it lands, run the numbers, and keep the house in order.

What to watch next

The next number that matters is the fourth-quarter print, and the next document that matters is the annual report. Watch first for the fourth-quarter results, due in early 2027, against the guide of 310 to 330 million dollars in core revenue, because the guide is the steepest step in the story and the capacity to deliver it is the question the OpenAI delays have already raised. Watch for the fiscal 2026 annual report, also due in early 2027, because it will give the interest income line, the full-year revenue, and the year-end balance sheet, and those three numbers will decide the income gate and settle the cash gate, and they will decide whether this site publishes a screen result for Cerebras and what it says. Watch for the Microsoft revenue to begin showing up in the first quarter of 2027, as Feldman said it would, because a deal with no disclosed value is a promise until the revenue arrives, and the revenue is the test of the promise.

Watch for the OpenAI delivery schedule, because the company has said it expects delivery to normalise in the second half of 2027, and the service credits disclosed in the quarterly report are the cost of the delay, and the delay is the risk. Watch for Nvidia's earnings on 17 November, the first under interim chief executive Colette Kress, because the chip trade's direction is set by its largest name, and Cerebras rose on a day Nvidia fell, and the rotation between them is the market's judgment on the buildout. Watch for Anthropic's IPO pricing on 18 November and trading on 19 November, because the appetite for artificial intelligence paper is the market Cerebras is selling into, and a weak debut would say something about that appetite. And watch the Federal Reserve's decision on 8 to 9 December, and the Bank of Canada's on 9 December, because the rate path is the screen weather, and the screen weather is the income gate. For readers watching Cerebras, the practical next step is the same as it was after the IPO, after the lock-up expiry, and after the third-quarter beat: note the arithmetic as it stands, note that a rally moves the cash ratio and does not change the income question, and recheck the numbers when the annual report lands. Run it through the screening methodology, check the company on the screener hub, and build the rest of the decision from there. A 15 percent rise is a market reading. The screen is the house. Read the rise, and keep the house in order.

FAQ

What did Cerebras report on 12 November 2026?

Cerebras reported third-quarter 2026 results after the market closed on Thursday 12 November, for the quarter ended 30 September. Core revenue was 230.9 million dollars, up 70 percent from a year earlier and above the 214.7 million dollar consensus, adjusted earnings were 12 cents a share against an expected loss of 14 cents, and the company raised its full-year 2026 core revenue forecast to 940 to 960 million dollars from 880 to 890 million dollars. GAAP revenue was 210.8 million dollars, up 55 percent, and the GAAP net loss was 161.2 million dollars, or 67 cents a share.

How did Cerebras shares react on Friday 13 November?

The shares rose about 17 percent in extended trading on Thursday to about 114 dollars, after closing at 97.75 dollars, added about 15 percent in early trading on Friday to about 113 dollars, then eased to about 104 dollars by late morning, still up about 7 percent on the day. The stock had closed Thursday about 47 percent below its 185 dollar May IPO price, and even after Friday's rise it remains about 44 percent below that price.

What is the Microsoft deal, and when does it start paying?

Cerebras announced an important cloud deal with a top-three hyperscaler, without naming the customer or giving a value, and chief executive Andrew Feldman said in an interview that the customer is Microsoft, that the deal was signed in October 2026, and that revenue should begin to show up in the first quarter of 2027, with more details to come next year. The company also signed a commercial agreement with Amazon Web Services in June, and Oracle is integrating Cerebras into Oracle Cloud Infrastructure in 2027.

Has Cerebras been screened for halal compliance?

Not by this site. Cerebras does not appear in this site's registry of 939 screened companies, and there is no screen result yet, because it listed only in May 2026 and its first annual report has not been filed. The raw arithmetic from its 30 September filings says the debt gate passes at about 4 percent of market capitalisation, the cash pile of about 7.4 billion dollars sits at or above the ceiling depending on the share price, and the income gate cannot be verified because the release does not split out interest income. Zoya rates the stock compliant as of 13 November 2026, and Musaffa rates it not halal as of November 2026, a split this article reports and does not resolve.

Why does the cash pile decide the screen?

The screen measures cash and investments against market capitalisation, with a ceiling of 30 percent, or about 33 percent under the reported provider disagreement. Cerebras holds about 7.4 billion dollars of cash and investments from its May IPO, and that pile is about 32.5 percent of the market value at Friday's late-morning price and about 34.7 percent at Thursday's close, so the verdict moves with the share price: a rise takes the ratio inside the ceiling, and a fall takes it outside. The pile also earns interest, which is what the income gate measures, and the third quarter's other income of 63.9 million dollars is not split out in the release, so the income gate cannot be verified until the fiscal 2026 annual report is filed in early 2027.

Is anything in this article a fatwa or a buy or sell call?

No. Everything here is facts and screening methodology only, with no fatwas and no personalised buy or sell calls. Rulings on your own situation belong with a qualified scholar, and investment decisions belong with advice suited to your circumstances. Not financial advice. Facts and screening methodology only, no fatwas.

Affiliate disclosure. This page contains no affiliate links, and the site currently earns no affiliate revenue. If we join affiliate programs in the future, this disclosure will be updated and links will be clearly labeled. Commissions never influence our scores or rankings — this review is based on official documentation and public data, and every product is Shariah-screened before review. Nothing on this site is financial advice.