The editorial argues that when measured on server CPUs and AI accelerators sold to hyperscalers — the segment where margins actually live — AMD's datacenter business is now larger than Intel's. The MI300 accelerator ramp combined with EPYC Turin and Bergamo displacing Xeon in Meta, Microsoft, Oracle, and Google fleets has turned a share-gain story into a revenue-line reality.
By submitting the Tom's Hardware piece framing AMD's Q4 guide as the first time in history it will outsell Intel, rbanffy signals endorsement of the narrative that this is a historic inflection point. The 255-point score and 88-comment thread reflect broad community agreement that the competitive balance has tipped.
The editorial explicitly flags that Intel's $12.2–13.2B Q4 projection bundles foundry, client, and 'everything else,' while AMD's $9.6B comparison point is company-wide. The honest apples-to-apples comparison is datacenter-only, which is where AMD has actually pulled ahead — the top-line 'outsell Intel' framing overstates the picture.
The editorial notes Lip-Bu Tan has kept the foundry carve-out moving and the client business still prints cash from the OEM laptop installed base. The loss is specifically of Intel's identity segment — being the default server CPU — not the entire company, which still has multiple viable lines of business.
The editorial argues that the two-decade procurement default compounded through OEM board designs, hyperscaler roadmaps, and enterprise software validation, letting Intel charge a premium even with inferior parts. But hyperscalers run TCO spreadsheets rather than making default choices, which is why the gravitational pull has officially inverted and AMD is now the incumbent in the segment where margins live.
AMD's Q4 revenue guidance now implies the company will book more quarterly revenue than Intel for the first time in the two companies' 56-year coexistence. Tom's Hardware, summarizing AMD's guide and Intel's own Q4 outlook, put the comparison plainly: AMD is pointing at roughly $9.6 billion for the quarter against Intel's projection in the $12.2–13.2 billion band — but Intel's number includes foundry, client, and everything else, while AMD's datacenter segment alone is tracking toward about $17 billion for the full year.
Measured on the fight that actually matters — server CPUs and AI accelerators sold to hyperscalers — AMD's datacenter business is now larger than Intel's. The company's MI300-series accelerator ramp, combined with EPYC Turin and Bergamo displacing Xeon in Meta, Microsoft, Oracle, and Google fleets, turned what was a share-gain narrative in 2022 into a revenue-line reality in 2025.
Intel, for its part, is not collapsing. Pat Gelsinger's successor Lip-Bu Tan has kept the foundry carve-out moving, and the client business still prints cash from the installed base of OEM laptops. But the segment Intel built its identity on — being the default server CPU — is now a two-horse race where AMD is the horse with momentum.
For two decades, "nobody got fired for buying Intel" was a procurement default that compounded. OEMs designed boards around Xeon sockets, hyperscaler capacity plans assumed Intel's roadmap, and enterprise software was validated on Intel first. That gravitational pull was worth real money — it meant Intel could charge a premium even when AMD had the better part, because the switching cost on the buyer side was non-trivial.
The Q4 guide is the moment that gravitational pull officially inverts: AMD is no longer the challenger brand asking for a seat at the table, it's the incumbent in the segment where margins actually live. Hyperscalers don't make default choices. They run TCO spreadsheets, and those spreadsheets have been pointing at EPYC for three generations now. Turin's 192 Zen 5c cores per socket at competitive power envelopes isn't a marketing bullet — it's the reason Meta publicly disclosed that the majority of its new server deployments are AMD.
The AI accelerator side is more interesting than the headlines suggest. MI300X and MI325X are not catching Nvidia — nobody is catching Nvidia in 2025 — but they don't have to. AMD only needs to be the credible second source that hyperscalers can point to during their Nvidia negotiations, and the OpenAI deal announced in October (a multi-year, 6-gigawatt commitment) established exactly that. The value of being the credible number two in an Nvidia-dominated market turned out to be worth more revenue than being the dominant number one in a commoditizing CPU market.
Intel's structural problem is that its response has to come from a foundry that isn't yet competitive at the leading node, with a product team that has lost its best architects, against an opponent (TSMC via AMD) that gets first dibs on N3 and N2 capacity. Granite Rapids is a respectable part, but "respectable" doesn't claw back 40 points of hyperscaler share. And the foundry pivot is a decade-long bet that depends on customers who aren't yet signed up at volume.
Community reaction on Hacker News captured the vibe accurately: this is less "AMD won" and more "Intel lost the thing it was built to do." The 255-point thread was notably light on AMD triumphalism and heavy on postmortems — what Krzanich's manufacturing missteps cost, what the 10nm delay compounded into, what the foundry carve-out means for the combined-margin story that justified Intel's old valuations.
If you're running workloads on a cloud provider, the practical implication is already priced in: AWS, GCP, and Azure have been quietly standardizing new general-purpose instance families on EPYC for two years. The m7a, c7a, and n4 families are AMD; the equivalent Intel SKUs exist but tend to launch later and price higher per vCPU for the same workload. If your IaC still pins Intel-family instance types by default, you are paying a tax on inertia — benchmark the AMD equivalent and rebook.
For on-prem buyers, the socket conversation is different now. A Turin-based 2U can replace three to four older Xeon 2Us in dense-core workloads (Java, Go services, Kubernetes nodes, PostgreSQL), and the power delta matters enough that colo facilities are starting to adjust their per-rack power allocations around it. The playbook that worked in 2019 — "refresh Xeon with newer Xeon" — now leaves 25–40% TCO on the table depending on workload profile. Run the numbers before signing the next three-year lease.
On the AI side, don't overcorrect. MI300X is viable for inference workloads where you have ROCm engineers or a vendor doing the integration work for you (Databricks, Together, a few others). It is not yet a drop-in replacement for H100/H200 for training, and the CUDA moat is still real. The useful read on AMD's accelerator business isn't "switch today," it's "Nvidia's pricing power has a ceiling now, and your 2026 capacity contracts should reflect that."
The x86 duopoly isn't ending — Intel is too large, too deeply embedded, and too strategically important to the US industrial base to simply fade. What's ending is the era when Intel set the price and AMD took whatever share was left over. For the first time since the Athlon 64 era, the pricing power in server silicon sits on the other side of the table, and every cloud bill, colo contract, and hardware refresh over the next 18 months will reflect that. Plan accordingly.
Love them or hate them, from a customer's point of view AMD is the only company in a position to stand up to NVIDIA in the GPU market and force those sky-high prices down a bit. I'm eagerly waiting for their new models, now backed by a much more mature and well-thought-out ROCm.I'm no
I have avoided Intel products since the time some law enforcement forensic team used DCILeech to read the RAM of my server hosted at Hetzner and extracted a LUKS key from it. That was a lifelong lesson. There is no guarantee AMD won't come up with similar shenanigans like Intel's DCI soon
AMD specifically calls out high sales of Instinct products as leading the sales growth. I wonder how their Epyc and Ryzen lines compare to Intels Xeon and Core lines. AMD's Instinct sales are practically nothing compared to competing Nvidia products, but no one† buys Intel Arc accelerators, so
Good. I was bitten by Intel's voltage issues in their 13th gen CPUs. Then again in the 14th gen replacement. They appear bloated, complacent and I hope AMD eats their lunch.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
That says more about Intel's shambolic efforts than it does about AMD's road to damascus moment.Honestly both of them have demonstrated that they have little interest in AI. Intel gave up on Optane right as (a much-improved successor) would've helped with MoE offsetting or even as a h