Samsung's in-house benchmark is a negotiating position with a 14 trillion won price tag

The 14 trillion won does not evaporate if Exynos wins ^[2]. It moves. Substitution shifts money from a supplier's revenue line to Samsung's own System LSI design costs and its foundry's loading ^[4], which is why the process node is the load-bearing detail rather than the Geekbench delta: the Exynos 2700 is slated for SF2P, the second generation of Samsung's 2nm process, with launch expected next year ^[12]. Yield and thermals there decide whether the transfer is accretive or just relocated.

The mix is the variable, not the benchmark. The Exynos 2600 reached only some Galaxy S26 models plus the Galaxy Z Flip 8 ^[5], so the live question is what share of Galaxy S27 units Samsung routes through its own part ^[13].

The leverage does not require the swap to happen. First-half purchasing of mobile APs ran at 7.44 trillion won, 17.9% of the DX division's raw-material bill, which implies a total raw-material spend of about 41.6 trillion won for the half ^[15][5]. Annualized, the AP line runs to 14.88 trillion won ^[1], roughly 7.6% above last year's 13.83 trillion ^[2][14]. Last year was itself up 26.5%, an increase of 2.90 trillion won ^[14][3]. A buyer on that curve does not need to move a single wafer to profit from a credible alternative: a 5% concession across the annualized run rate is worth about 744 billion won ^[4].

The test results deserve the discount their provenance earns them. Samsung's handset division measured Samsung's chip against a Qualcomm part that has not shipped, the results reached the press through industry sources, and a Samsung official said the company could not confirm them ^[3][16]. Several margins are thin enough to come from unit selection or test setup: 5% at GPU peak, 5% on prompt processing ^[8][10]. The figures that are harder to manufacture are the low-power ones, 24% and 22% more GPU performance in a 2.5W envelope ^[9], and 161 mA against 185 mA in a simulation of real use ^[11]. Both point at the same claim, a part tuned for the thermal budget of a phone rather than a bench run. One footnote on that current draw: 185 mA to 161 mA is a 12.97% reduction, not the 12.7% quoted ^[11][6].

For Qualcomm, the reading is arithmetic. A customer whose order book grew from 10.93 trillion won to an annualized 14.88 trillion, about 36% in two years ^[14][7], has now let it be known that it has an in-house option it prefers on CPU, GPU, AI throughput and power ^[6][7]. Whether that option takes the S27 socket matters less than its presence in the room when the next supply contract is priced.