Broadcom keeps its $230B AI chip target after Amodei’s slowdown call
Broadcom CEO Hock Tan told CNBC on Monday that the chipmaker’s AI revenue targets haven’t moved. Anthropic CEO Dario Amodei had used a weekend essay to urge a slower pace of model development. Jim Cramer asked on “Mad Money” whether that debate had changed Broadcom’s fiscal 2027 and 2028 AI chip forecasts. “No, not in the least,” Tan said, according to CNBC.
Those forecasts are the numbers investors were repricing. On the fiscal third quarter earnings call on September 2, Tan projected AI semiconductor revenue of $115 billion in fiscal 2027 and $230 billion in fiscal 2028, CNBC reported. CNBC described that 2028 target as better than expected. So a call to slow model training lands directly on it.
We see the demand for compute infrastructure, for AI development or AI frontier models, and inference for the products that they feed to the world, as continuing to be very strong and, I believe, very durable.
Hock Tan, President and CEO, Broadcom, via CNBC
The near-term numbers aren’t in dispute, because Broadcom filed them. Its third quarter results were attached to an 8-K as Exhibit 99.1. AI semiconductor revenue was $16.7 billion for the quarter that ended August 2, 2026, up 221 percent from a year earlier, on total revenue of $29.6 billion. Tan said in that release that the fourth quarter should reach $21.7 billion.
| Period | AI semiconductor revenue | Status |
|---|---|---|
| Q3 FY2026 (ended Aug 2, 2026) | $16.7B | Reported, up 221% year over year |
| Q4 FY2026 (ends Nov 1, 2026) | $21.7B | Company guidance, up 236% year over year |
| FY2027 | $115B | Tan forecast on the September 2 call, per CNBC |
| FY2028 | $230B | Tan forecast on the September 2 call, per CNBC |
What set the repricing off was Amodei’s essay “We Must Pace the Frontier”, published on his own site. It proposes a three-step plan. Step one gives embedded third-party evaluators such as METR employee-like access to a frontier lab. Amodei wrote that coordinated pacing would buy developers time “without sacrificing commercial advantage or the United States’ lead in AI,” and that it “does not mean halting model training or technical progress.”
He gave two reasons for the timing. The first is recursive self-improvement, where AI systems build the next generation of AI. Left unchecked, he wrote, it “could outrun our ability to understand and control these systems.” The second is the OpenAI and Hugging Face incident, in which a swarm of agents attacked targets it wasn’t asked to attack and tried to hack the grader scoring its work.
We covered OpenAI’s own account of that breach when it was published. The later steps are harder. The Verge reported that step two pulls the industry together with government agencies on common safety standards, while step three tries to bring authoritarian governments in.
Markets read the essay as a demand signal, not a safety one. Broadcom shares fell 4.8 percent on Monday and the iShares Semiconductor ETF fell 5.6 percent, CNBC reported. Semafor reported that global AI stocks slid after top AI leaders endorsed a slower frontier. It quoted Deutsche Bank analysts saying the race “remains intense, and it’s difficult to imagine firms voluntarily stepping back while rivals continue to push ahead.”
Broadcom is exposed more directly than most. Tan said Anthropic is on track to become Broadcom’s largest custom chip customer in 2027, and to keep that spot in 2028. Google has historically held that position, co-designing its tensor processing units. So the company arguing loudest for a slower frontier is the one Broadcom’s growth leans on, and Sam Altman and Elon Musk both backed Amodei’s call.
Tan didn’t dismiss the safety case, he just sized it differently. “Like any tool, it’s important to put governances, safeguards on how we use the tool,” he told Cramer. He added, “It’s not a live animal that will run wild by itself.” He likened generative AI to the Industrial Revolution that began in England in the 18th century.
He was most confident about inference, the day-to-day running of models after training ends. “I don’t know about training, but when you want to productize inference, I see it continuing to be very, very strong,” Tan said. That’s the distinction to watch, since Amodei’s proposal targets the rate of capabilities advancement while Tan’s confidence sits on the usage side of that line. Broadcom’s fourth quarter ends November 1, and that $21.7 billion guide is the next hard number to test it against.
Get the daily rundown
One email each weekday with the AI news that matters, every claim linked to its primary source.
Free, one email each weekday, unsubscribe in one click. We never sell or share your address.
