It wasn’t the Intel monopoly that dealt Transmeta, one of the highest-flying chip companies to come out of Silicon Valley, its mortal blow. The wound was self-inflicted. The startup let down its customers, chip buyers for computer makers who had stuck their necks out to get their companies to use Transmeta’s unproven but promising low-power processors.
When the Santa Clara, CA, company came out of stealth mode in January 2000, it was swept up on a wave of hype. It had the right technology to pioneer low-power microprocessors, which, because they use less battery power and throw off less heat than other chips, allow laptops to be both speedy and thin. If Transmeta had hit its performance and power targets, it could have taken leadership away from Intel – which had focused mainly on cranking up the megahertz – in the fastest-growing segment of the PC industry. Laptops are now close to half the market.
David Ditzel, one of the designers of Sun Microsystems’ first Sparc chip, which helped create the workstation/server market, founded Transmeta in 1995 and raised money from the likes of Paul Allen and George Soros; the company’s initial public offering in late 2000 brought in $273 million. The game plan was clear. Transmeta was supposed to break in with laptop chips and then pioneer low-power gadgets of a type that hadn’t previously existed, such as one-pound laptops and handhelds that could run Windows. The company could have become profitable with just a fraction of the $27 billion PC microprocessor market, selling chips to the likes of Toshiba, IBM, and Hewlett-Packard. But it made mistakes.
First, it took a long time to design its chips. It was five years before Transmeta announced its first product and almost another year before its first chip shipped. Also, the company wasn’t conservative enough in its choice of manufacturer. It bet that Taiwan Semiconductor Manufacturing would be able to roll out an advanced manufacturing process on schedule. But Taiwan Semiconductor faltered, as do many chip makers when implementing new fabrication equipment and materials. This caused a big delay in chip deliveries, a consequent loss of face for customers, and the erosion of Transmeta’s lead over Intel.
“It’s almost a Greek tragedy,” says Nathan Brookwood, an analyst at Insight 64 in Saratoga, CA. “All they needed was the chorus in the back. They promised a tremendous amount and didn’t really deliver on those promises.”
Problems with execution were magnified by a high-maintenance design. Transmeta had taken the novel step of moving some functions ordinarily performed in hardware into software, which reduced the size of its chip and lowered power consumption; it expected to make up the difference with increased processing speed. But the task of developing and constantly updating the added software proved to be a drain on company resources.
When Transmeta lost its lead in low-power chips, it awoke a sleeping giant in Intel, which launched a low-power Pentium M chip in 2003 and took most of the market with the Centrino, which aggregated everything computer makers needed to build wireless-network laptops.
Then the economy fell into a recession. Transmeta’s sales slipped from $35.6 million in 2001 to $17.3 million in 2003. The company tried again with its second-generation chip, the Efficeon, but it never won back customer loyalty, and Intel’s momentum grew. The cutting-edge portables where Transmeta dominated never became a high-volume market.
Transmeta has been up for sale for at least five months. But with a market capitalization of around $200 million, it has no takers at present. Advanced Micro Devices, the number-two PC microprocessor maker behind Intel, could use Transmeta’s technology but hasn’t been willing to pay the steep price. And while Intel could have used Transmeta’s technology five years ago, it doesn’t need it now.
The company’s last resort is to license its LongRun2 technology, which saves power by reducing the “leakage” of electrical current across transistors that are supposed to be turned off. NEC, Fujitsu, and Sony have licensed LongRun2. But the revenues that have resulted – a few million dollars a quarter – won’t solve Transmeta’s problems. The company has told its customers that they can license its chip designs, but that it will fill orders only for existing chips. At press time, Transmeta had only $53 million in cash and had been burning through $25 million a quarter.
“This is the way Silicon Valley works,” says Paul Saffo, research director for the Institute for the Future in Palo Alto, CA. “Companies make mistakes, and we learn from failure, not successes. Innovation doesn’t always win.”
Meet Altos Labs, Silicon Valley’s latest wild bet on living forever
Funders of a deep-pocketed new "rejuvenation" startup are said to include Jeff Bezos and Yuri Milner.
Meta’s new learning algorithm can teach AI to multi-task
The single technique for teaching neural networks multiple skills is a step towards general-purpose AI.
All charges against China Initiative defendant Gang Chen have been dismissed
MIT professor Gang Chen was one of the most prominent scientists charged under the China Initiative, a Justice Department effort meant to counter economic espionage and national security threats.
A horrifying new AI app swaps women into porn videos with a click
Deepfake researchers have long feared the day this would arrive.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.