Intel says its Razor Lake-AX platform will deliver 700 TOPS of AI compute in 2028. That is the claim. It arrives as a roadmap number, not as a measured result from shipping silicon, and the gap between those two categories has burned buyers before.

The figure is large. It is also four years away. Both facts deserve equal weight.

What did Intel actually commit to?

The company named a product family, Razor Lake-AX, a target, 700 TOPS, and a year, 2028. That is the full extent of the confirmed record. The claim is a single sentence of substance stretched across a long wait.

Consider what Intel did not say. There is no confirmed process node. There is no stated wafer size, no power envelope, no shipping date for first silicon, and no breakdown of where the 700 TOPS come from. The number does not specify whether it counts NPU throughput alone, or CPU, GPU and NPU combined under a favorable measurement scheme. TOPS figures vary widely depending on the precision used for the count. Without that disclosure, the number cannot be compared against rivals on equal terms.

A vendor roadmap is a promise about intent. It is not a capacity figure, a revenue figure, or a shipment number. The distinction matters for anyone planning products around it.

Why the 2028 date invites skepticism

Four years is a long time in this market. In 2028, the competitive baseline will not stand still. Qualcomm, AMD and Apple are all scaling on-device AI throughput now, with shipping parts rather than slides.

Intel itself has recent history with roadmap slippage. The company has missed process milestones before and has restructured its foundry plans more than once. A 700 TOPS target for 2028 assumes execution across multiple generations of silicon, packaging and software enablement. Each step carries risk.

The claim also carries a commercial purpose. AI PC marketing rewards big TOPS numbers. The bigger the number, the better the headline. Buyers should ask what workload actually saturates 700 TOPS on a client device in 2028, and whether software will exist to use it. Intel's statement, as reported, does not answer that question.

How should readers treat this?

Treat it as a stated ambition with a date attached. The correct unit of account is a vendor claim, attributed to Intel, with no independent verification. It belongs in the same category as any pre-announcement: useful for tracking direction, unreliable for planning purchases.

The checklist for the coming months is short:

• A named process node for Razor Lake-AX
• A power envelope for the 700 TOPS figure
• A measurement precision for the TOPS count
• First silicon or sampling dates
• Independent benchmarks rather than vendor math

None of these items exists today. Until one appears, the 700 TOPS figure remains a promise on paper.

What happens next

If Intel details the node, the power budget and the measurement method, the claim becomes testable. Until then, rivals will keep shipping parts with smaller but verifiable numbers, and the 2028 target will be judged against whatever they ship in the meantime.