Following a third consecutive record-breaking sales quarter reported by programmable logic vendor Xilinx Inc. Wednesday (July 22), at least one Wall Street analyst reiterated his contention that extending lead times by the leading FPGA supplier put it at risk of double ordering.
During an analyst conference call Wednesday, Xilinx CFO Jon Olson acknowledged that lead times have grown longer from the 10 to 12 weeks the company reported during its last earnings call.
"We note Xilinx stated its lead times have extended to over 12 weeks for some Virtex-5 devices (estimated 26 percent of F1Q11 sales), above last quarter’s 10-12 week range," Danely wrote in a report circulated Thursday morning. "As a result, we believe Xilinx is at risk of double ordering. The company stated lead times are unlikely to be reduced until the December quarter due to ongoing supply constraints."
Danely earlier issued the same warning about the other big programmable logic vendor, Altera Corp., which also reported record quarterly sales this week. He noted that Altera reported extended lead times for some parts of up to 26 weeks.
Double ordering is the dreaded phenomenon where customers hedge their bets by placing orders with multiple vendors, fully intending to cancel some orders depending on who can deliver the parts first. It's bad news for everyone and creates a misrepresentation of actual demand that can give suppliers fits.
Danely has warned about Xilinx being at risk for double ordering before, most recently following the company's last quarterly report in May. And analysts have generally been concerned about inventories at Xilinx and Altera for several months.
The disconnect here is that Danely is the same Wall Street analyst who earlier this week raised a red flag because Texas Instruments Inc. was pulling in lead times (though they remain high by historical standards), noting that one of his firm's rules for semiconductor investing is, simplistically, "Lead times coming in is bad."
To recap, Danely warned TI investors that the company was pulling in lead times (much to the relief of TI customers), then later warned Xilinx and Altera investors that the companies were reporting extending lead times. No disrespect to Danely intended, but the logical conclusion is that in his view, the only safe situation is for lead times to remain unchanged. But lead times for most parts are continually fluctuating, as supply and demand struggles to find equilibrium.
While it's difficult to reconcile Danely's advice on TI with his advice on the programmable logic vendors, it should be noted that he and other analysts gave generally very positive reviews to the quarterly results turned in by both Xilinx and Altera, though Danely himself maintains a "neutral" rating on both stocks, citing fears of inventory build and correction (also difficult to reconcile with warnings about extending lead times).
"Despite the lack of improvement in lead times, Xilinx continues to meet critical customer demand and management does not believe they have lost any share to competitors as a result," wrote Hans Mosesmann, an analyst with Raymond James & Associates. "Xilinx expects to be wafer constrained for at least another two quarters, but sees no unusual buffer inventory accumulation."
Mosesmann maintains a "strong buy" rating on both Xilinx and Altera.
Olson said Xilinx could have shipped an additional $20 million or so in revenue during the quarter if not for supply constraints. "We think we can hold at that level through the September quarter and not expand that number, and I think that will be a good thing," he said.
In his research note on Xilinx earnings, Ian Ing, an analyst with Gleacher Co., suggested that the shortfall and supply constraints at 65-nm may be due to Xilinx' declining priority at foundry supplier United Microelectronics Corp. (UMC). After years of being a top-tier customer at UMC, with corresponding privileges, Ing said he believes Xilinx "has little leverage to get highest-tier priority at UMC, as foundry supplies remain tight and UMC best serves its long-term partners."
Xilinx announced earlier this year that it would use Taiwan Semiconductor Manufacturing Co. Ltd. and Samsung Electronics Co. Ltd. as its foundry suppliers for the 28-nm node, forsaking UMC after a more than 10 years of partnership. UMC will presumably continue to manufacture TSMC parts at other process nodes.
Ing maintains a "buy" rating on Xilinx and a "neutral" rating on Altera. He increased his revenue estimates and share price targets for both companies following this week's earnings announcements.
Xilinx traded at $28.50 in midday trading Thursday, up 2.5 percent from Wednesday's close. Altera traded at about $28.30, roughly flat with its closing price Tuesday, when it announced second quarter earnings.
There is some logic to both being issues (too long or too short lead times) as the situations represent significant changes in trends and new challenges for management. The consequent increase in business risks for each company requires the analyst to call attention to the reliability of the financial forecasts. Basically, any time a company gets slapped too far off center, recovering their balance adds the risk they stumble or fall. Of course, it could just be wishy-washy CYA too!
I find the arguments, put forward by Danely, hard to comprehend. Xilinx being mostly a sole-source, should not be too bothered about double-ordering. Unless, it can keep the lead times below a certain threshold, beyond which, customers might find value in platform change. I absolutely don't agree with the correlation between pulled-in lead times and low demand. In times, where customers were struggling with time-to-market due to extended lead times, if companies, such as TI, come up with efforts to actually put the lead times within acceptable brackets, it does not reflect in any way on the demand. It actually augurs well in the long run as customer confidence is retained and it acts as a magnet for customers in growing economies where time to market is essential.
Xilinx makes unique parts; typically the parts are neither software compatible nor pin compatible with other competitor parts, so Xilinx's end-customers will need anywhere from 6 months or more to design in a different part. Moreover Xilinx's main competitor Altera has the same issue of extending Lead times, so it's not like end-customers can double order Xilinx with Altera. Thus the analyst's concern that Xilinx's extending Lead times puts it at risk of double ordering, is not convincing.
On the other side of the coin, shorter Lead times are bad if it is caused by end markets softening, resulting in lower demand. (But it's obviously a good thing if end markets are strong, and a company has been able to match capacity and production with customer and market demand).
In the case of TI, the issue is more complex. Though TI's end-market of Analog is growing, TI is still overweighted towards specific customers and products in its Wireless business unit, where market-share and end-demand is starting to flatten.
@John- I like the idea. I wish I'd thought of it. Very creative.
I nearly changed the title after I read your suggestion, but in the event that this column gets made into a movie I don't want to have to share the proceeds.
Apparently Mr. Danely is unaware of market share with "noting that one of his firm's rules for semiconductor investing is, simplistically, "Lead times coming in is bad.""
With standard component lead times at 26 weeks on many parts, and major improvement would justify changing vendors.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.