Run the ratio and 77.6 billion against 150.3 billion yuan is 51.6 percent falling through to net profit [6], which is what happens when price moves against a cost base that mostly does not, and a year earlier the same product line produced a loss [2] in what the report calls a market for commodity products [7]. Annualise the half and the run rate is 300.6 billion yuan against a prior full year that must have been under 75.2 billion, since six months more than doubled it [20].
The forward numbers are where price and value part company. Goldman Sachs has the shares at roughly ten times next year's estimated earnings [14] on forecasts of 361 billion yuan of net income next year against an estimated 161 billion this year [15], which multiplies out to about 3.61 trillion yuan of implied market value [16], call it 538 billion dollars at the 6.71 the report's own dollar conversion implies [17]. This year's estimate is nearly closed already: 77.6 billion booked leaves 83.4 billion for the second half, 7.5 percent above the first [18]. So the argument is entirely about the 2.24x step to 361 billion [19].
This is probably wrong, but the ten times reads less like a governance discount than a refusal to capitalise a peak, and the counter-thesis has real teeth. Samsung, SK Hynix and Micron are all prioritising the high-end memory that AI accelerators require [8], which is capacity they are not pointing at conventional DRAM, and CXMT, the world's fourth-biggest DRAM maker [5], is working on its own high-bandwidth memory but does not have it yet [10]. The leverage it is exercising sits in the commodity tier, which is the tier that reprices fastest in both directions [7].
More interesting than passing Tencent as China's most valuable company [4] is where the money goes. The 66.6 billion yuan Shanghai raise [3] is smaller than one half-year's profit [2], so the stated plan to fund capacity expansion and catch up with larger rivals [9] is a cash-flow decision rather than a financing one, and those wafers land in 2027 (the report is dated 29 August 2026, so its "next year" is 2027) [21]. Most Western buyers will not own a CXMT die: the Apple discussions Bloomberg reported cover devices sold in China [11], and a group of US senators has already told Apple to drop Chinese suppliers [12]. They will still pay a version of the Hefei price, because the pass-through is already sitting in Nvidia's system quotes [13].
What would break the thesis: second-half profit under 83.4 billion yuan [18], which would say the pricing peaked in the first half, or incumbent capex that puts conventional DRAM wafers into 2027 faster than CXMT's expansion can [8].
