Cerebras CEO Andrew Feldman Talks Data Center Capacity artwork

Cerebras CEO Andrew Feldman Talks Data Center Capacity

Bloomberg Talks

August 13, 2026

Cerebras CEO Andrew Feldman says limited data center availability remains a challenge across the AI industry while the company's own cloud business is seeing strong demand.
Speakers: Ed Ludlow, Andrew Feldman

Topics: Business, News, Business News

**SPEAKER_1** (0:02)
Bloomberg Audio Studios, podcasts, radio, news.

**Ed Ludlow** (0:10)
Okay, for just the second time since its IPO, Cerebras reported quarterly earnings yesterday after the closing bell, offering some investors growth that was slower than anticipated. Shares down 13.4%. I think that's on track for the biggest drop since the last week of June. Cerebras CEO, Andrew Feldman joins us now.
The first thing that I learned was there is a business renting out compute capacity that's still there. Could you give me a back story on that, Andrew?

**Andrew Feldman** (0:41)
Oh, of course. There's a huge business of renting out compute capacity. I think it's what the neoclouds do, it's what the hyperscalers do.
It's a very profitable business.

**Ed Ludlow** (0:58)
And is it outpacing the growth from the business of selling the hardware direct to others?

**Andrew Feldman** (1:04)
Well, we have two parts of our business. We have a first-party cloud. That means we design, manufacture, and deploy equipment in our own data centers, and we rent it out to our customers. That part of the business was up 281% year over year. Huge growth. Huge growth. Now, in that part of the business, we control the data centers.
There was another part of our business where we sell our AI hardware to customers who want to do the exact same thing, who want to deploy the equipment, and either use it for themselves, or break it up into little pieces and sell the compute to various customers, right? Those customers, we call that our hardware business. We're selling hardware for their premise. And in Q2, some of our customers didn't have data centers available, so we were able to deliver less to them.
And so, right now, data centers are a challenge for everybody. We had some. We were able to deploy them for our own use and for the use of our cloud customers, our first-party cloud customers. And so that business is absolutely ripping.

**Ed Ludlow** (2:24)
We're live on Bloomberg Television and Radio with Cerebras CEO Andrew Feldman. And the reason I asked for the backstory on that is that that cloud business was born out of needing to build data center capacity to test the technology worked.
It seems to be working.

**Andrew Feldman** (2:38)
It was.
We wanted to be able to provide demonstrations to customers without having to move hardware to their facilities. We didn't have a hyperscale customer at that time. And so there was a need to find a way to deliver compute services to people who didn't have a data center, didn't have a way to otherwise enjoy our blisteringly fast AI hardware. And so we set it up at first for ourselves and then for demonstrations. And there was so much demand on it over time that it's become a big business for us. And last quarter Q2, it was an extraordinary business for us. And we expected to continue to grow at tremendous rate. Now, even though now we have hyperscale customers like AWS, now that we have frontier labs like OpenAI, that business continues to grow at exceptional rates.

**Ed Ludlow** (3:39)
Okay, the hardware business, there is some confusion out there. You report on a core basis which showed that the hardware division was up, right? On a gap or non-core basis, it would be otherwise. Just explain the story for the hardware business.

**Andrew Feldman** (3:56)
The story is, one of our large hardware customers had some warrants, and those warrants are subtracted from revenue. The value of those warrants are subtracted from revenue in gap accounting. And so, it made the numbers look a little smaller than they were actually.

**Ed Ludlow** (4:19)
Demand is great. It's there.
There is a backlog. We know about it.

**Andrew Feldman** (4:24)
And we've got 25.4 billion in backlogs.

**Ed Ludlow** (4:28)
So here is the data that people are really searching for with Cerebras, which is how can you get the capacity in place faster, both on your own data centers or to get the hardware into the hands of customers for theirs? Right.

**Andrew Feldman** (4:42)
And that's a fair question. What we said to on our call yesterday is that we had signed up or had already live 600 megawatts of capacity, and that we had a pipeline of gigawatts of additional capacity. And what we said was that we expect our revenue to more than triple next year.
And so I think we have been executing at a tremendous rate, you know, adding 600 megawatts in several months, six or seven months is no small feat.
And we will add vastly more in the months to come. And so I think we are ahead of plan. We beat guidance and we beat consensus on the things we, on our core business, on core revenue, core gross margins, and beat by a lot on core operating margins. And then we guided a big raise. So for 2027 and 2028 So we feel really good about the business and are proud of last quarter's results.

3 more minutes of transcript below

Thousands of transcripts fetched by people building searchable podcast archives

Feed this to your agent

Try it now — copy, paste, done:

curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.

Using your own key:

curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/YOUR_EPISODE_ID