OpenAI has been busy deepening its push into the semiconductor world, striking a sweeping new partnership with Broadcom Inc (NASDAQ:AVGO, ETR:1YD) this week to design and produce custom AI accelerators. The move adds to a flurry of recent chip-supply and development tie-ups — including deals with Advanced Micro Devices Inc (NASDAQ:AMD, ETR:AMD) and ongoing collaboration with Nvidia Corp (NASDAQ:NVDA, ETR:NVD) — that underscore how the ChatGPT maker is racing to secure the hardware muscle behind its fast-growing ecosystem.
But even as OpenAI tightens its grip on the physical backbone of artificial intelligence, rising worry over an AI bubble remains a theme, with some of the world’s largest fund managers now beginning to voice doubts about the sustainability of the AI boom. A new Bank of America survey widely reported today shows an increasing number of investors see tech valuations as stretched, with more than half now describing AI stocks as being in bubble territory.
Building the backbone
Broadcom’s newly announced partnership will see it co-develop and manufacture custom accelerators for OpenAI’s next generation of large-scale models. The chips are expected to begin rolling out in the second half of 2026, with full deployment extending into 2029.
Rather than simply purchasing GPUs off the shelf, OpenAI will collaborate on the architecture itself — tailoring hardware to the unique demands of its training clusters and data-centre operations. The arrangement gives OpenAI a measure of independence from Nvidia’s supply constraints while allowing Broadcom to flex its strength in networking and application-specific integrated circuits (ASICs).
The deal also fits into a broader strategy that’s been building for months. In early October, OpenAI agreed to a multibillion-dollar GPU supply arrangement with AMD, reportedly giving it the option to acquire a minority stake in the chipmaker. And it continues to work closely with Nvidia, whose processors still power much of OpenAI’s current training infrastructure. Collectively, these agreements suggest OpenAI is hedging its bets — locking in near-term capacity from established suppliers while laying the groundwork for a longer-term, custom-silicon roadmap.
There is risk in that approach. Chip development is capital-intensive, fraught with yield challenges, and highly dependent on execution. But if successful, it could help OpenAI bring costs down, boost energy efficiency, and squeeze more performance out of its sprawling model-training operations — advantages that might compound over time.
Market jitters over AI valuations
Meanwhile, the broader market context appears increasingly uneasy. Bank of America’s latest global fund-manager survey indicates that optimism toward technology shares has reached its highest level in months — yet so have worries that valuations have overshot. Roughly half of respondents now believe AI stocks are in bubble territory, making it the most frequently cited risk factor in the poll.
Strategists at the bank have pointed to valuation metrics edging toward extremes reminiscent of the late-1990s tech boom. Meanwhile, cash allocations are falling, and overall exposure to equities — particularly in the US tech sector — remains elevated. In other words, investors are still chasing the rally even as they question its foundation.
That tension is starting to play out in price action. After months of relentless inflows into AI-linked names, recent sessions have shown more volatility across chipmakers and big-cap software firms alike. It’s not yet a reversal, but the conviction that “AI can only go up” is showing cracks.
What to watch next
- Early hardware milestones: The first generation of OpenAI-Broadcom accelerators will be a key proof point. If they deliver meaningful efficiency gains, competitors could follow suit.
- Capital intensity: Developing and fabricating custom chips is a multiyear, multibillion-dollar exercise. How OpenAI manages those costs will influence investor sentiment towards the broader AI ecosystem.
- Market discipline: If earnings growth fails to justify current multiples, a pullback in high-flying tech could ripple through global indices.
- Competitive realignment: Success for OpenAI’s custom silicon could loosen Nvidia’s dominance and accelerate a shift towards vertically integrated AI stacks.
OpenAI’s hardware ambitions mark a new phase in the race to define AI’s physical infrastructure — one that could reshape both the economics of computing and the competitive order among chipmakers. But with investors already whispering “bubble,” execution will matter more than ever. If OpenAI’s chip bets pay off, they’ll be seen as visionary. If they stumble, they could come to symbolise the very excess investors are beginning to fear.