George Gilder's THE FRIDAY LETTER of June 1, 2007
gilder.com
Issue 297.0/June 1, 2007
Excerpts:
--
The Week / The Exaflood is Coming
Gilder Telecosm Forum Member #1 (5/29/07): Would you agree that EZchip (LNOP) is also unnecessary for an all-optical network? After all, it is electronic, right?
George Gilder (5/29/07): When the optical network reaches all the way to the end computer, some electronic device will be needed to accept the flow of packets from the optoelectronic transceiver, read them, and route them to the various functions within the computer. That device would have to handle packets at fiberspeed at seven layers (it would have to do TCP termination and assembly) and decryption and anti-malware, and perhaps decoding and rendering as well. It would resemble in many respects a network processor and it would perform the most exacting task in the computer.
By this time, however, the computer will have been largely hollowed out and dispatched across the net, to datacenters and application servers of various kinds, and the device with the processor might be a teleputer in your hearing aid or Microvision (MVIS) shades. In any case, the hollowed out computer will have to have NPUs in it to distribute its tasks, as the central processor becomes increasingly a communications device. NP15 anyone?
But this is all a fantasy too many years ahead to pin down (and pins too will be gone by then on Cubic Wafers).
As to the exaflood that I used to write about in 1999, it is indeed coming now as uTube et al go high def and it is creating huge new demand on the network, accelerating the day of all-optical systems. I own Infinera through Kleiner Perkins so I am happy with its IPO, but all-optical systems should obsolete it with a few years.
Still it is my view that compute functions will remain electronic. The all-optical computer is a special purpose device, as even Terry Turpin says (come and see him at Telecosm 2007). The all-optical system will need many electronic servants.
Gilder Telecosm Forum Member #2 (5/29/07): Can the in-house NPU developer's such as Acatel (ALU) develop apace with EZchip?
George Gilder (5/29/07): I doubt it. The in-house team will likely be less competent and lower on the learning curve. It will lack volume, fail to command the attention of company executives, suffer distraction from custom demands, and will not be fully exposed to the upside.
The trend is toward outside NPUs, even at Cisco (CSCO), which is the bellwether.
Gilder Telecosm Forum Member #3 (5/29/07): I think this is a classic innovator's dilemma. The NPU team at ALU is going to need to ask for corporate resources (capital and people) to maintain leading edge performance. Listen to Cisco's last call. Chambers said they are increasingly a software company. Juniper (JNPR) and Redback (RBAK) have long-maintained their differential is in the software not the hardware. In my last meeting with Juniper I asked about the trend of NPUs "hollowing out the router" and value migrating out of the router companies to the merchant silicon. The notion was roundly dismissed, again because the value is in software not the hardware.
Now consider ALU attempting over the long-run to justify and maintain an internal source of commodity hardware components. Think of EZchip like DeBeers Diamonds. Would it make sense for one jewelry retailer to maintain its own diamond mine and be vertically integrated for long? Only until the monopolist (EZchip) was able to drive his cost of production below the vertically integrated producer's cost. Then vertical integration becomes a handicap. The internal NPU provided a real window of opportunity in terms of performance for ALU and RBAK, but my guess is that window will be closed in 2008 and 2009 when the full suite of JNPR boxes is on the market and the CSCO upgrades are rolling out. In my opinion, RBAK saw the window closing when VZ decided it could wait to buy JNPR's future products, rather than RBAK's current product advantage, and sold to ERIC. ALU will see the window closing on its performance advantage and will need to then rationalize the costs of its products... getting rid of the internal diamond mine would be the way to go.
How about this for a transaction... EZchip buys the ALU NPU group and agrees to support the current ALU products through 2009 and in turn ALU makes a commitment to use the NP-3 or NP-4 in its next generation boxes.
Log on with your subscriber password at gildertech.com to read more posts by George Gilder and the GILDER TELECOSM FORUM members.
--
Readings /
The Weekly GTI
gtindex.com
YouTube for Apple TV Uses H.264, Not FLASH
arstechnica.com
Google Zooms in Too Close for Some
nytimes.com
Decoding Your DNA Destiny
forbes.com
Studies: music industry overstating threat of P2P piracy
arstechnica.com
Dell To Chop 7,000 Workers
blog.wired.com
------ |
