Last September we suggested that the ailing Transmeta's haemorrhaging bank balance would not allow it to continue for much longer. It lost $591 million over the previous four years, $25.5 million of it in its second quarter of 2004, more than the $22 million lost in the equivalent quarter of 2003.
Sadly, we have been proved right, as Transmeta this week announced it was pulling out of the chip fabrication business and moving instead to a licensing model in a bid to stem the flow of red ink. Specifically, it will "complete a critical evaluation of the economics of its current business model of designing, developing and selling x86-compatible microprocessor products."
So, Transmeta is to adopt a model like ARM's, designing not making processors. But this might make it an acquisition target -- but who would the buyer be?
Transmeta, which launched a processor called Crusoe back in 2000, was a darling of the start-up industry. Its initial chip offering, Crusoe, appeared to have everything a maker of mobile systems could want. It used very little power compared to products from the big chip makers and allowed very long battery life.
Since then, though, things have changed. Intel's Pentium M arrived, and extended laptop battery usage by a considerable proportion. Today, it's not unusual to extract four or more hours of normal usage from a single battery charge. AMD is following suit with its recently announced Turion, aimed right at Intel's Centrino, itself a Pentium M with added wireless.
And the incentive to cut processor power consumption is now driven, not just by laptop users, but also by the blade server market. Processor and systems designers are running out of options when it comes to keeping their power-hungry technology cool while satisfying the demand both from enterprises to increase server densities, and from consumers who baulk at sitting next to machines whose cooling fan noise rivals Concorde's.
Add this level of competition to one other problem: Transmeta's original chip, Crusoe, was not x86-compatible. Instead, it used a VLIW (very long instruction word) architecture that used code-morphing software to convert the code into x86-compatible. While it dramatically reduced the amount of power consumed by the chip because much of the processing was moved from transistors to software, performance suffered.
By the time Transmeta overcame some of the problems with its second generation Efficeon, the window of opportunity had all but closed.
Licensing is much lower-cost model to operate, so Transmeta's move makes sense, especially as even fab-less chipmaking still means you have to test and distribute product, with all the cost and logistical issues that raises. Instead, the company at least has a future.
But with the barrier to entry into the CPU market even higher than ever, Transmeta still faces a momentous struggle to make headway in today's more cautious market. When vendors such as HP refuse to make statements about its bc1000 blade server range -- and HP is one of the few mainstream vendors selling Transmeta products into enterprises -- it's clear that the path forward could well be boggy.
On the other hand, it would not be surprising if a larger chip company either licenses Transmeta's technology wholesale, providing a chunky, steady revenue stream or, better still from its point of view, buys the company outright.
Given Transmeta's connection with Linus Torwalds -- he was involved in the launch -- and Microsoft's current bullish attitude to IP acquisition, maybe the Redmond giant will enter a bid Transmeta and its shareholders would be foolish to reject. This would then deny the technology to Linux vendors. On the other hand, Intel is also on the acquisition trail and would make a better fit.
The jury's out, but Transmeta's tale is by no means finished.