OpenAI’s chip shopping spree is noise, not a Nvidia nightmare
Reports that OpenAI is exploring alternative silicon have spooked parts of the market, but Wedbush argues the story is stale, the risks are overstated, and Nvidia’s grip on AI infrastructure looks intact through 2026.
Another week, another headline suggesting Nvidia’s dominance in AI chips is under threat.
This time, the catalyst was a Reuters report claiming OpenAI is actively seeking alternative silicon for certain inference workloads, citing concerns around memory configurations and access speeds. The implication was clear: if OpenAI is looking elsewhere, perhaps the GPU era is already peaking.
That conclusion, however, looks premature.
According to Wedbush, the market reaction confuses long-running experimentation with a genuine architectural shift. OpenAI has never been shy about exploring options, and none of this is new.
OpenAI has been hedging for years
OpenAI’s interest in alternatives to Nvidia predates this latest report by some distance.
The company has already announced work with AMD, has been widely reported to be collaborating with Broadcom on a custom ASIC, and Sam Altman has previously floated the idea of building dedicated fabs to support future chip needs.
Seen in that context, the Reuters story looks less like a strategic pivot and more like a reheated narrative. Wedbush describes it as “stale” and sees little evidence of a near-term shift that would materially dent Nvidia’s position.
The key point is timing. Even if alternative architectures eventually matter, they are unlikely to matter soon.
Why GPUs still win in a fast-moving AI market
Silicon Valley experts seem clear-eyed about one thing: GPUs are not a perfect solution for every AI workload. But neither are ASICs the silver bullet some investors want them to be.
AI models are evolving rapidly, and that flexibility still favours general-purpose architectures. Custom silicon can deliver efficiency gains, but it locks developers into assumptions that may not age well. Scaling SRAM-based solutions also remains technically and economically challenging.
Google’s relative success with TPUs is often cited as proof that ASICs can work, but Wedbush notes this has been the exception rather than the rule.
For now, the practical reality is that access to hardware matters more than theoretical performance advantages.
Supply chains trump architecture in 2026
Perhaps the most underappreciated part of the debate is supply.
In a note, Wedbush argues that over the next few quarters, availability will trump almost every other factor. Even if alternative chips look attractive on paper, getting them manufactured, packaged, and deployed at scale is another matter entirely.
On that front, Nvidia has a clear advantage. The firm has executed better than peers in sourcing components and materials, giving it resilience in a supply-constrained environment.
That alone makes any meaningful displacement in 2026 highly unlikely.
A quieter shift that could matter more
While the OpenAI story grabbed headlines, Wedbush flags another development that could have broader implications over time: Samsung’s decision to move glass substrates out of R&D and into a formal business unit.
Glass substrates promise lower signal loss, better structural stability, and higher chip density compared with organic alternatives. Samsung is reportedly targeting a production ramp in 2027, a timeline that suggests this is no longer just a science project.
If successful, the shift could ripple across the semiconductor ecosystem, affecting everything from advanced packaging to materials suppliers. It is a slower-burn story, but arguably a more consequential one.
The bottom line
Related reading
- Oracle shares rise as group targets up to $50bn for AI infrastructure
- Buyer’s guide identifies six top voice AI platforms for enterprise deployment in 2026
- Liberty Global strikes AI partnership with Google Cloud
The idea that OpenAI is experimenting with alternative silicon spells imminent trouble for Nvidia makes for a good headline, but a weak investment thesis.
Wedbush’s view is that GPUs remain central to AI infrastructure, supply constraints favour incumbents, and any genuine architectural transition is a multi-year story at best. For now, the noise is louder than the signal.
