Every generation of tech giants falls into the same trap: self-developed chips burn cash to the point of questioning life choices.

Apple has been working on baseband chips since 2018, and Intel's team bought in, pouring $7 billion into it, with mass production continually delayed. Google’s TPU saw its first three generations nearly break even, only to truly figure out the economics in the fourth generation.
Now it’s OpenAI’s turn.
On May 8th, the news broke: OpenAI's collaboration with Broadcom on a custom AI accelerator—a project planned for 10GW-level computing power, claimed to partially replace the aging GPU—hit a funding snag. The issue isn’t with the tech; it’s about the cash. An $18 billion manufacturing contract needs to lock in old Zhang's production capacity ahead of time, but OpenAI's current cash flow can't cover this scale of capex.
Interestingly, Broadcom isn’t sweating it. In Q1 FY2026, AI chip revenue hit $8.4 billion, up 106% year-over-year, and they still hold two mega contracts with Meta MTIA and Google TPU. The delay with OpenAI doesn’t affect their goal of hitting $100 billion in AI chip revenue by 2027.
But whether OpenAI is worried is another story. If the timeline for self-developed chips slips, continuing to rely on old GPUs means their cost structure will forever be at the mercy of others. And as for the pricing power of the old GPUs… well, those in the know understand the situation.
The historical script keeps flipping through the same few pages: technology has never been the hardest part; it’s the money that is. Those who can burn cash until mass production day are the winners; those who can’t will just end up back in line to buy GPUs.
#TradfiTradingChallenge