$Cerebras Systems(CBRS)$  

Cerebras' First Profit Came From Interest, Not Chips

Mathematical Money | November 16, 2026


Cerebras makes AI chips the size of a whole silicon wafer, and it rents out computing on them. It listed in May at $185 a share. On Friday it closed at $108.72.


Last Thursday it reported its first profit as a listed company. The CFO's words were "positive non-GAAP net income for the first time as a public company". The stock closed up 11.2% the next day. So is the turnaround real?


Is the profit real?


Not yet. Even on the company's own adjusted numbers, the business of selling chips and renting out computing lost $33M in the quarter. The company still reported a profit of $30M. What turned it positive was $64M of other income, mostly interest on its cash. Most of that cash came from the IPO.


There's a second catch. Those adjusted numbers are what "non-GAAP" means. They leave out about $190M that was paid in shares, not cash. Staff got $155M of it. Customers got $36M as a discount for buying, mostly G42 and Amazon. Almost none went to OpenAI.


Paying in shares costs no cash. But it creates new shares, so everyone who already holds the stock owns a smaller slice. If your employer paid part of your salary in shares, you'd still call it salary. Count it as a cost, the way standard accounting does, and Cerebras lost $161M.


Is it cheap after falling 41%?


No. At Friday's price the market pays about 31 times a year of sales, taking this quarter's sales times four. You can run that sum on any stock. It's the value of all the shares, divided by a year of sales. A 41% fall doesn't make 31 times cheap. It only tells you the IPO price was higher still.


Run the same sum on a different stock. CoreWeave rents out AI computing too. At Friday's close of $53.13, all its shares were worth about $30 billion. A year of sales, on last quarter times four, is about $15 billion. That's about 2 times, against 31 for Cerebras.


That doesn't settle which stock is better. CoreWeave owes about $47 billion of debt, and this sum ignores debt. What the gap shows is how much future growth is already in the Cerebras price.


So the price is a bet on growth, and the first test is the next report. The company guides sales of $310M to $330M for the coming quarter, on its own adjusted measure. That's 34% to 43% more than this quarter on the same measure, in three months.


Is relying on three customers normal?


A few big buyers is common in AI hardware, but Cerebras is at the far end. Three customers made up 79% of its revenue. They are a UAE university called MBZUAI, OpenAI and G42. The filing lists G42 and the university as related parties, so count those two as one group.


Compare CoreWeave again. In the same quarter its two biggest customers made up 54% of revenue, and nobody else reached 10%. At Nvidia, the biggest direct customer was 16% of revenue in the quarter to July.


Cerebras is less lopsided than it was. A year ago the university alone was 70% of revenue. It still matters, because when one buyer is a third of sales, one delayed order moves the whole quarter. You can check this on any company. Open the quarterly report and search for "10% or more".


When can IPO holders sell?


They already can. After an IPO, staff and early investors are barred from selling for a set period, called a lock-up. When it ends, more shares can come up for sale, and that can push a price down.


For Cerebras the lock-up ended on Monday 9 November, 180 days after the prospectus and three days before these results. Most of the shares didn't wait for that date. The prospectus let them out in stages from May, nearly four-fifths of the locked shares by 28 October. It put those staged releases at up to 171 million shares. The IPO itself sold 34.5 million.


So the 41% fall happened while the number of shares free to trade kept growing. The filings don't say how many were actually sold. What you can check is the calendar. The dates are in the prospectus under "Shares Eligible for Future Sale". For Cerebras the last one has passed.


Who's right about the chips?


Nobody outside the company knows yet. Culper Research, a short seller, says the wafers "last just 1.5 to 2.5 years in the field". I couldn't open its report, so that quote is from its X thread. The company spreads the cost of its equipment over 3 to 5 years. The CEO's answer to Culper was "That number is wrong." Culper makes money if the stock falls, and the CEO owns the stock.


It matters because a shorter life means a bigger cost every quarter. On my own arithmetic, the equipment already running costs about $26M a quarter over 4 years. Over 2 years it costs about $52M. If you want to redo it, divide the equipment on the balance sheet by the years, then by four. And another $1.1 billion of equipment is still being built, so none of its cost shows up yet.


What do I watch next?


Three things in the next report. The first is sales against that $310M to $330M guide. The second is whether the business makes money before interest. The company guides it to land between -1% and +1% of sales, which is break-even. The third is when the $1.1 billion of new equipment gets switched on and starts adding cost.


If the business breaks even without the interest, the turnaround is real. If the interest is still doing the work, it hasn't happened yet.


How do you check a "first profit" yourself?


It takes three steps and the company's own press release. You can run it on any company that announces a first profit.


1. Find which line turned positive. Is it the business itself, or only the line after interest? Here the business lost $33M.

2. Add the share payments back as a cost. Here that turns a $30M profit into a $161M loss.

3. Check how much interest the cash earns. If it's bigger than the profit, the cash is doing the work. Here it was $64M against $30M.


[PROPOSED] I don't hold Cerebras.


The chip question is the one no filing can settle for me. The people who know are the ones running the machines. If you've run AI hardware in a data centre, any brand, how long did it last before you pulled it?


Sources: Cerebras' earnings release and quarterly report for the quarter to 30 Sep 2026, filed 12 Nov 2026, and its May 2026 prospectus. Dollar figures are rounded. The 31 times is about $26.4 billion of market value over four times the quarter's $210.8M of sales. Prices are Tiger closes on 12 and 13 Nov 2026. CoreWeave: quarterly report for the quarter to 30 Sep 2026, filed 12 Nov 2026; about 556M shares at $53.13 over four times $3.66 billion of sales. Nvidia: quarterly report for the quarter to 26 Jul 2026. The equipment cost run is my own arithmetic on the quarter-end balance of $414.2M.


$CBRS$

#Cerebras #AI #semiconductors #earnings #investing


Stop guessing. Start calculating.


Live to fight another day. 🤙

# CBRS cloud business surged by 272%! Surging 18% after earnings results, is the competitive landscape of AI semiconductors changing?

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Report

Comment

  • Top
  • Latest
empty
No comments yet