Broadcom CEO Hock Tan told a panel at the DBS Asia Future Forward Conference in Singapore on Monday that capital, not memory chips, will be the biggest constraint facing the company as AI development continues. "What I think might be the biggest constraint is capital," he said, according to CNBC. Reuters filed the story under the headline "Broadcom CEO says biggest hurdle in AI-fuelled tech growth is capital, not memory chips," and Bloomberg, Quartz, Calcalist and India Today covered the same session.
Hock Tan's argument rests on track record. "In the [last] two years we have seen a series of bottlenecks that keep coming out, but then we overcome it," he said. "Memory shortage today might not be that way a year from now." India Today reported that he expects memory shortages to improve within a year, with wafer capacity and power infrastructure stabilizing by 2028 or 2029.
"There is more demand for compute infrastructure than we can supply," Hock Tan said, adding that the imbalance could persist through 2029 and potentially into 2030. He attached a condition to that: infrastructure spending could decelerate if businesses adopt cutting-edge models more slowly than expected, or if the models stop improving as fast. On the demand side he pointed to Anthropic and OpenAI, which he said have shown they can generate significant revenue from their latest models deployed in AI personal assistants.
Intel CEO Lip-Bu Tan made the same point from a different angle on the same panel: "If there's any challenge in the front, it's capital and the cost of capital, and then the other part is in the longer run, people want to see the ROI." He does not expect a significant slowdown over the next two to three years, citing growing AI adoption and compute demand, but said sustaining growth beyond that window depends on whether the spending produces real economic benefits.
CNBC places the warnings against rising borrowing costs: higher bond yields make data center construction and chip purchases costlier to finance, pressuring projects that may take years to earn their way back. Neither company is waiting around. Broadcom launched a financing platform with Apollo and Blackstone in June, opening with a $35 billion transaction tied to roughly 20 gigawatts of planned AI capacity through 2028. Anthropic's IPO prospectus, reported exclusively by Reuters on October 1, shows Broadcom has agreed to lend Anthropic up to $42 billion, about a third of Anthropic's $125.2 billion five-year TPU lease commitment. Bloomberg has reported Broadcom assembling a roughly $60 billion debt package, split between a $42 billion senior-secured tranche and an $18 billion junior tranche led by Blackstone.
Intel has been in the market too. Quartz reported that it upsized a common stock offering to $20 billion in August, pricing 210,526,315 shares at $95 each to fund its contract-manufacturing business, after lifting its 2026 capital expenditure target to $20 billion from $18 billion in July. It has committed to high-volume production on its 14A process node by 2028.
Bloomberg reported that Hock Tan likened many AI investors to casino gamblers: "A lot of investors have no clue what they are investing in," he said, and they "set very extreme expectations only to find out that it's almost like they are in a casino." He was equally blunt about corporate AI budgets, comparing enterprise experimentation to a child burning through arcade credits: "Your parents give you $10 of tokens, you finish it in 10 minutes and you haven't learned anything. That's what happening in enterprises today."
According to AIStockWire, Broadcom shares were down 0.5% at $359.66 and Intel down 1.3% at $103.30 in premarket trading as of 8:20 a.m. ET.
The two CEOs land on the same timetable: two to three years of demand that does not need defending, followed by a test of whether capital stays affordable and whether AI returns show up. For Broadcom the test is concrete. Hock Tan has reiterated targets of roughly $115 billion in AI semiconductor revenue in fiscal 2027, doubling to about $230 billion in fiscal 2028. Hitting them will depend less on whether chips can be built than on whether the money to build them keeps flowing.