# AMD delivered better stock returns than Nvidia in H1 2026 but the AI chip war is far from over > Source: > Published: 2026-07-26 06:50:27+00:00 *Nvidia still owns roughly 87% of AI data center GPU revenue, but AMD's stock surged 142% in the first half of 2026 against Nvidia's 4% gain, and Intel has effectively dropped out of the race entirely.* The AI chip war has a winner on paper. It's Nvidia. But if you owned AMD stock heading into 2026, you had a much better first half. That tension - between who dominates the market and who rewards investors - is the real story of the three-way scorecard that played out between January and June. Nvidia's numbers are staggering in absolute terms. The company reported $81.6 billion in revenue for its fiscal Q1 2027, with data center revenue of $75 billion up 92% year over year, fuelled almost entirely by Blackwell architecture systems. Its share of combined merchant GPU vendor revenue in AI data centres sat at 87.4% in Q1 2026, according to Axis Intelligence data. No other company in the history of the semiconductor industry has printed margins like this at this scale: chip-level gross margins on the H100 run around 88%, on the B200 around 84%. By every traditional measure of market dominance, Nvidia is not in a race. It has won. The problem for Nvidia investors is that everyone knew this. Entering 2026, the market had already priced in Nvidia's dominance so completely that even another quarter of record-breaking execution moved the stock roughly 4%. AMD, by contrast, was still being treated as a speculative bet on a distant alternative. Its stock gained 142% in the first half, as reported by The Motley Fool and confirmed across multiple market trackers. Same sector, opposite returns. That's what happens when a company is underpriced relative to its actual trajectory. AMD's data center revenue hit a record $5.8 billion in its most recent quarter, up 57% year over year. Its MI300X and MI355X accelerators now hold roughly 5% to 7% of AI accelerator revenue, which sounds modest until you consider where AMD was eighteen months ago. Microsoft Azure, Meta, Oracle Cloud, and Dell are all deploying Instinct GPUs at scale. In October 2025, AMD and OpenAI announced a 6 gigawatt partnership to power next-generation infrastructure across multiple Instinct generations, starting with MI450 deliveries in the second half of 2026. OpenAI also received a warrant for up to 160 million AMD shares tied to deployment milestones. That is not the kind of deal you sign with a chip supplier you plan to drop. Frankly, AMD's moat problem is real and shouldn't be papered over. Nvidia's software ecosystem, CUDA in particular, has a decade of lock-in behind it. AMD's ROCm software has improved significantly, but moving a model training infrastructure from CUDA to ROCm is not painless. Most hyperscalers use AMD as a complement to Nvidia rather than a replacement. AMD's chip-level margins, in the 64% to 68% range on the MI300 series, also trail Nvidia's by a substantial gap. Closing that gap requires yield improvements and volume that AMD is still working toward. But here's the thing: AMD doesn't need to beat Nvidia to justify its valuation. It needs to hold No. 2 as the market expands toward a trillion dollars in total addressable compute. Even a 10% slice of that is a very large business. ## Intel is the real story, and not in a good way Intel's Gaudi 3 AI accelerator never gained meaningful traction. The company's own executives acknowledged it would not generate significant revenue. Gaudi 3 missed even its modest target of $500 million in revenue for 2024, and none of the four major clouds - AWS, Azure, Google Cloud, or Oracle - deployed it at scale, citing immature software. Intel then cancelled Falcon Shores, its planned successor, for commercial release entirely, redirecting it to internal deployment. The next chip in line is Crescent Island, based on Intel's Xe3P microarchitecture with up to 160GB of memory, shown at Computex 2026. Customer sampling is expected in the second half of 2026. The odds aren't encouraging. According to analysis published by FutureSearch, the probability of any top-four cloud announcing a Crescent Island deployment by the end of 2027 sits at around 13%. One analyst at The Register put it bluntly earlier this year: Intel has missed the boat for AI in the data centre. This matters beyond Intel's own shareholders. A genuine three-way competition for AI compute would benefit every cloud buyer, every startup paying per-GPU-hour, and every company trying to build at scale. The exit of Intel from meaningful competition accelerates the duopoly forming around Nvidia and AMD. Hyperscaler custom silicon - Google's TPUs, AWS Trainium, Microsoft Maia, Meta's MTIA - is growing and collectively sits near 15 to 20% of the market. Merchant alternatives, though, are narrowing to two names. That's not a dynamic that favours buyers. AMD stock up 142%, Nvidia up 4%, Intel somewhere in between but absent from the product war that matters. The chip race winner in H1 2026 depends entirely on how you keep score. Market share says Nvidia. Investor returns say AMD. And for anyone hoping Intel would break up the duopoly, the answer is not yet - and maybe not for a while. **Also read:** [Chinese investors are using crypto derivatives to route around Beijing's AI stock controls](https://startupfortune.com/chinese-investors-are-using-crypto-derivatives-to-route-around-beijings-ai-stock-controls/) • [Anthropic's silence on Nvidia's open-weights letter is the most revealing thing about it](https://startupfortune.com/anthropics-silence-on-nvidias-open-weights-letter-is-the-most-revealing-thing-about-it/) • [Moonshot AI's Kimi K3 model sends its valuation toward $50 billion and a Hong Kong IPO](https://startupfortune.com/moonshot-ais-kimi-k3-model-sends-its-valuation-toward-50-billion-and-a-hong-kong-ipo/)