OpenAI Models Find Up to 14% Improvement in AMD’s Leading GPU IP
OpenAI says its models found up to 14% PPA gains on AMD’s leading GPU IP. AMD verified single-digit CPU gains, but no shipping silicon shows measured improvements.

Why it matters
- OpenAI reported up to a 14% PPA improvement on specific blocks of AMD’s leading GPU IP against heavily optimized baselines.
- AMD verified single-digit PPA gains across key blocks of its upcoming leading CPU, with Synopsys checking behavior preservation.
- OpenAI estimates its model cut Jalapeño standard-cell area by about 13% in the vector unit and 10% in the matrix unit.
- The October 2025 OpenAI-AMD deal covers six gigawatts of AMD GPUs and rights to buy up to 160 million shares.
- AMD Data Center revenue rose 97% to $8.546 billion last quarter, while its forward P/E is near 30.
By Sophie Lindqvist
OpenAI said in its research post that its models found up to a 14% improvement against heavily optimized baselines on Advanced Micro Devices’ (NASDAQ: AMD) leading GPU intellectual property. The models rewrote the SystemVerilog code that describes the chip’s circuits.
24/7 Wall St. reported that AMD closed at $530.14 that day, up 5.31%, for a market value of $877.42 billion. AMD traded at $536.99 after hours as of 7:59 p.m. ET, though the outlet said no source tied the move to the post and AMD issued no formal statement.
The stake is the pace of hardware improvement. OpenAI described the work as roughly a year of progress in less than five months, while 24/7 Wall St. said the payoff for AMD lies in sustaining a new product generation every year.
What does the 14% actually measure?
24/7 Wall St. said the “best chip design” in the headline refers to AMD’s leading GPU IP. OpenAI’s result is a ceiling measured on specific GPU blocks, not a blanket improvement across every part of a finished processor.
The source defined PPA with three measures:
- Power: How much electricity a circuit uses.
- Performance: How fast it runs.
- Area: How much silicon it takes up.
24/7 Wall St. described RTL as the code engineers write to describe a circuit before it becomes a physical layout. OpenAI did not say how much of the 14% came from power, performance or area, and the number says nothing about how much faster a finished AMD GPU will run.
OpenAI said AMD added some changes to a design it had recently sent for manufacturing, a step called tape-out, and other changes to production files for future versions. According to 24/7 Wall St., tape-out remains far from shipping, and no one has reported measured gains on AMD silicon.
For now, the 14% is a design-stage ceiling, not a shipping-GPU speed claim.
Why does AMD’s CPU result carry more weight?
AMD itself verified single-digit percentage PPA gains across key blocks of its upcoming leading CPU, using designs written by OpenAI’s model. Synopsys (NASDAQ: SNPS) tools checked that each rewrite left the block’s behavior unchanged.
24/7 Wall St. called that CPU result the strongest evidence in the post because the company had checked both the gains and the behavior of the rewritten blocks. That is a firmer result than a model’s claim about an unreleased GPU design.
The source explained the commercial effect of a smaller, more efficient design:
- A smaller chip allows more chips per wafer.
- More chips per wafer lowers the cost per chip.
- Lower power improves performance per watt in data centers limited by their electricity supply.
The power constraint also pulls grid-equipment and cooling suppliers into the same infrastructure buildout, according to 24/7 Wall St. Against AMD’s fourth-quarter revenue guidance of about $17.0 billion, block-level gains are small; their value lies in helping AMD keep shipping a new product generation every year.
The CPU work has passed an important check, but it has not produced a measured performance figure for a finished processor.
What do OpenAI’s AMD incentives mean?
Under an October 2025 deal, OpenAI agreed to use six gigawatts of AMD GPUs. The deal also gave OpenAI the right to buy up to 160 million AMD shares at a low fixed price, and those shares vest partly as AMD’s stock hits price targets.
That makes the GPU disclosure commercially entangled. 24/7 Wall St. said every figure in OpenAI’s post came from a partner that profits when AMD looks strong, while AMD’s share purchase right also gives OpenAI exposure to increases in the stock price.
AMD’s only public reaction came in a post by CEO Lisa Su. She quoted OpenAI co-founder Greg Brockman’s note that some results were in a recently taped-out design and called the work “fantastic work by engineers at both companies.”
The partnership gives the result real-world relevance, but the same relationship requires investors to separate technical validation from partner incentives.
What did OpenAI change in its own chip?
OpenAI is also building its own chips. The company estimates that its model cut standard-cell area by about 13% in the vector unit and 10% in the matrix unit of its Jalapeño inference chip.
Those results target area rather than a combined PPA claim, and they apply to one OpenAI-designed chip. They show that the model reduced the specific area measure OpenAI estimated; they do not add a separate performance or power result for a finished Jalapeño chip.
OpenAI’s own hardware test offers narrower evidence than the headline, with savings tied to specific chip units.
Does Synopsys have the stronger long-term edge?
OpenAI is working with Synopsys to train GPT-Synopsys to use Synopsys’s electronic design automation, or EDA, software. These are the tools chipmakers use to design and test circuits, and once GPT-Synopsys becomes a product, any chip designer can buy it, NVIDIA (NASDAQ: NVDA) included.
24/7 Wall St. said that wider availability would make AMD’s lead a matter of months. Synopsys gets paid by every customer using the tool, so it depends less on any one company’s product cycle.
That model gives Synopsys broader exposure to AI-assisted chip design and a stronger claim to a lasting edge. It also turns OpenAI’s work with AMD into a tool that could eventually serve AMD’s competitors.
Synopsys may gain more durable value from distributing the method than AMD may gain from owning the first implementation.
What should AMD investors watch now?
AMD’s stock is up about 148% in 2026 but remains about 18% below its $649.42 record close. The company’s core business provides the stronger valuation case: Data Center revenue rose 97% to $8.546 billion last quarter, and a forward P/E near 30 already prices in a lot of that growth.
AMD is scheduled to present at the UBS and Barclays conferences on December 1 and December 9. The source’s test is specific: management must tie the OpenAI-designed changes to a named product with a ship date for the claim to gain weight.
24/7 Wall St. was equally direct about the alternative: “If neither presentation mentions it, treat the 14% figure as marketing.” Until AMD provides silicon measurements and a product timeline, OpenAI’s 14% remains an early design benchmark rather than evidence of faster GPUs.
Source: 247wallst.com
More from Sophie Lindqvist
Show full bio
Staff writer covering regulation, policy and their impact on artificial intelligence and machine learning at AI In Context.
823 articles
Related articles
- OpenAI Reports Up to 14% PPA Gains in AMD Chip Designs
- AMD Is Up 158% While Nvidia Slips. The Case Runs Deeper Than Rival Trouble
- AMD Poised for AI Market-Share Gain as NVIDIA Faces Leadership Vacuum
- AMD’s Sell Rating Puts $196.72 Fair Value 56% Below Market
- AMD Completes $2.9 Billion Taalas Acquisition for Model-Specific AI Chips