How long has the IDM got? Five years? Ten? The current consensus is that only the specialist guys, the foundry guys, the DRAM guys and Intel will still have fabs in five to ten years time.
And even Intel, with its outsourcing deal to TSMC, has conceded that in certain circumstances its IDM model is inappropriate.
Last year the fabless guys had sales of $55 billion up 3 per cent on 2007 and the IDMs had sales of $197 billion which was an 8 per cent decline on 2007.
Of course the top ten have always had a declining market share of the overall semiconductor industry. That's a trend going back 30 years which gives the lie to the occasional emergence of the theory that the industry is 'consolidating'.
Now the screw is tightening on the IDMs. Eight out of the top ten semiconductor under-performed the market last year which meant they declined faster than the average 5.2 per cent for the industry, while fabless companies, with growth of 1.4 per cent in 2008, out-performed the market.
Meanwhile the IDMs suffered the ultimate indignity of having a fabless company, Qualcomm, in the top ten.
Of fabless companies in the top 25, Qualcomm, Broadcom, Marvell and MediaTek grew their revenues between 10.2 per cent and 23.9 per cent in 2008 while only nVidia declined.
Fabless suppliers, with growth of 1.4 per cent in 2008, out-performed the overall semiconductor market and led the growth among the Top 25 semiconductor suppliers in 2008.
Qualcomm, Broadcom, Marvell and MediaTek each grew their revenues between 10.2 per cent and 23.9 per cent in 2008. Out of the five fabless companies in the Top25 rankings, only nVidia saw its revenues decline in 2008.
If you look at the Global Semiconductor Association's FABL Index for fabless companies with capitalisation of $500 million plus, the index is down only 5 per cent from October 1st last year.
While the mini FABL - what GSA calls the mFABL Index - for companies with a market cap of $100 million to $500 million is only 3 per cent down since October 1st.
So the evidence points one way: IDMs are dying. For the VCs, this is the biggest opportunity for decades.
TOMORROW MORNING: TEN FASTEST GROWING APPS
Are you saying that for instance Grenoble/Crolles fab near future is surely compromised?
Not at all, Djonne, the Byzantine complexities of the ownership and funding of Crolles will keep it going through 45nm, and 32nm, and very likely 22nm. Will ST build another fab there after 22nm? Well if the French government through all its various R&D Institutes and bodies, and France Telecom, wish it, and if Pres Sarkozy keeps up with his empathy for hi-tech stuff, then there could be a sub-22nm fab at Crolles.
But I don't think ST would pay, entirely out of its own resources, for a 32nm fab, let alone a 22nm one.