Lumenum, a company with a tumultuous history rooted in the telecom bubble, is experiencing a significant resurgence driven by the demands of the AI data center boom, particularly its critical role in supplying high-power, low-noise lasers essential for advanced optical networking.
Mind Map
Click to expand
Click to explore the full interactive mind map
Lumenum makes lasers, fiber optics, and
some other stuff. It has a wild history.
During the peak of the telecom bubble,
this company with less than $2 billion
in revenue was worth more than GM and
Ford combined.
After it all came crashing down, the
company languished in obscurity for over
a decade. But now, thanks to the massive
AI data center boom, it is so back. In
today's video, the glorious rebirth of a
fiber optic highf flyier and why. I want
to thank Irrational, author of the
irrational analysis newsletter, for
inspiring me to do this video and
working with me on the AI and laser
section. Check out his newsletter in the
description below.
Act one, scene one. We opened at Bell
Northern Research or BNR, the research
laboratory lab of the telecom giant
Nortell in the 1970s.
Canadians Yseph Strauss, Gary Duck,
Philip Geral Jones, and Bill Sinclair
are part of a team working on components
for the first optical fiber
technologies. Their efforts relate to
reducing noise and increasing data
capacity for carrying multiple video
streams. One technology they focused on
was dense wavelength division
multipplexing or DWDM.
This lets a type of optic fiber called
single mode fiber carry multiple signals
simultaneously by encoding them in
different light wavelengths. Such
signals can be switched, routed and
separated without fear of interference.
Fiber optics are a wondrous technology
but also have an infamous reputation as
a destroyer of capital. Prior to the
1990s, Nortell's fiber optic business
lost money for 10 years straight. So
Nortell somewhat understandably didn't
want to make these fiber optic
components in house. When that became
clear, Strauss, Duck, Geralt Jones, and
Sinclair left in 1981 to start JDS
Optical. JDS Optical produced passive
fiber optic network components, meaning
those that control light without using
any power. Optical couplers, splitters,
filters, and isolators.
In 1990, they struck a distribution and
technology sharing deal with the
Japanese giant Furukawa Electric.
Furukawa bought half of the company for
about $9 million. As part of the deal,
JDS Optical changed their name to JDS
Fitel. Vital being a Furukawa brand.
Over the next few years, the company
produced optical switches, couplers, and
lasers, essential items for optical
fiber networks. Overall, people knew
them as a quiet, low-key company that
kept their heads down.
In this manner, the company emulated its
co-founder and second CEO, Yseph
Strauss, a private individual who
immigrated to Canada from Czechlovakia
prior to the 1968 Soviet invasion. He's
known around Ottawa for wearing a black
wool beret.
Demand for fiber optic components grew
steadily but linearly throughout the
1980s. And when you're carrying just
telephone calls, faxes, and even emails,
like with the early internet, capacity
demand expanded at just about 8% a year.
Nothing to write home about. Then in the
early 1990s, we had the rise of the
worldwide web and suddenly telecoms were
in a race to build out fiber optic
networks for all the internet's
revolutionary functions. Streaming video
and audio, e-commerce, faxes, what have
you. CEO Joseph Strauss explained the
core trend in a 1998 interview. The
entire telecommunications industry is
growing like a fast spinning vortex
driven by the need for more bandwidth at
lower costs and greater flexibility.
JDS Fitel found itself in a fortuitous
position. They were the leading supplier
of DWDM equipment, which let telecoms
raise the data capacity of their
networks without needing to tear out and
replace the fiber, which is
prohibitively expensive. JDS's revenues
grew at 65% annual growth rate between
1994 and 1998. In March 1996, they went
public, raising about $93.7 million. The
stock quickly surged 52% to about $18.25
in August a few months later. Then in
January 1999, JDS Fitel announced that
it would merge with another
Californiabased rocket ship called
Unifase Corporation in a stock swap
transaction worth about $6 billion.
Unif is a classic Silicon Valley
entrepreneur story. The company was
founded by an engineer named Dale Crane
from the laser maker Spectra Physics.
Spectrophysics is one of the OG laser
companies. They were founded a year
after the laser's invention and helped
commercialize it. One of their products
was a small helium neon gas laser for
scanning supermarket barcodes. It helped
speed up the checkout process, but had
to be integrated into the counter. Crane
felt that it was possible to build a
similar laser but smaller. Small enough
to put into a handheld. His bosses were
not interested in that. So in 1979,
of savings. After a year, he finished
his laser and sold tens of thousands of
laser units to manufacturers. His vision
of a handheld laser-based grocery store
scanner produced by Symbol technology
became reality. UNIFA's revenues grew
rapidly, allowing Crane to build more
complicated laser products, moving up
wavelengths from helium neon red lasers
to argon blue and even CO2 lasers.
In 1992, Unifas's board appointed a bold
software executive named Kevin Calhovven
as its CEO. Calovven's mandate was to
accelerate the company's plotting growth
and IPO so that investors can get a
financial return on these random laser experiments.
experiments.
Calovven first led an entry into the
semiconductor industry, producing a
defect scanning inspection tool called
the Ultra Point. It scans a blue argon
laser across the silicon wafer to find
defects as small as 0.1 microns in
diameter. Ultra Point was a fairly
successful business that eventually
boosted Unifiase towards their 1993 IPO
which valued them at about $30 million.
Then in 1994, Calovven read magazine
article profiling how MCI the telecom
used an amplified laser beam to send
data from Chicago to Sacramento. The
beam was sent via optical fiber with
relatively little loss. no need for
electrical equipment to regenerate the
signal through copper wires. Calovven
immediately became a true believer in
the technology and proposed to reorient
the company to this new space. At the
time, this was an unpopular move.
Calovven personally had no experience in
telecommunications and UNIFAS had no
experience in the semiconductor lasers
used for telecom. Their expertise was in
gas lasers as I mentioned earlier. So
more than a few people were skeptical.
Paul Crane, for his part, thought the
man was crazy, sold off all his shares,
and disembarked to Nevada to raise a
family. But Calovven pushed forward. In
1995, he paid $8 million to buy defense
contractor United Technologies Photonix
Division, UT, was then going through the
early 1990s posts Soviet defense bust
and needed to slim down.
The purchase was as risky as can be.
Calovven was betting the company, but it
gave Unifase their first critical fiber
optics component, a reliable low power
lithium nobate modulator that could
modify a laser beam at gigahertz speeds
to encode data signals. UNIFAS then went
on to build up the company's offerings
with a raft of acquisitions, including
IBM's laser diode division in
Switzerland in 1997.
In 1998, they swallowed up Philips's
opto electronics group in a complex
stock swap. The transaction helped
complete Unifas's lineup of
semiconductor lasers from the fiber
optic industry. Previously, they only
offered 980 nanometer lasers. Philips
provided two others, 1310 nanometers and
1550 nmters.
Then in 1998, Calovven got a call from
JDS's Strauss asking whether he might be
interested in doing something together.
So the two CEOs went on a hike in the
Colorado Rockies as rich CEOs do and
hashed out how a combination might look
like. The deal was finalized in
September 1998 near Lake Tahoe and as I
said earlier was announced in January 1999.
1999.
The combined company called JDS Unifase
or JDSU began life in July 1999.
Calovven would serve as CEO and
co-chairman. Yseph Strauss became COO
and the other co-chairman. The latter
celebrated by giving away thousands of
his trademark black beret. The company
JDSU had sales of about $587 million in
profits of 124 million. More
importantly, the combination brought
together JDS's strengths in passive
fiber optic components and Unifias'
strengths in active components. The
company can now serve as a one-stop shop
for buyers like Lucent and Nortell, who
use those components to produce
equipment to fulfill the seemingly
insatiable demands of telecoms like
Quest, Global Crossing, Level 3, and 360 networks.
networks.
Investors loved the merger. The stock
soared 12% when it was announced and
they kept loving it throughout the year.
JDS Unifas's stock went from a split
adjusted 1725 per share in January 1999
to $161.35
by December and it just kept going.
Using its ballooning stock price, JDSU
went on one of the telecom bubble's
wildest acquisition sprees, which is
saying something considering that this
was the peak of the telecom bubble.
Companies like Corning, Sarant, and
Lucent were all shelling out billions to
buy startups at inflated valuations.
But JDSU, you should see them feast.
They were like a mantis. In the second
half of 1999 alone, JDSU bought six
companies, including a $2.8 8 billion
all stock deal with Optical Coating
Laboratory, a maker of optical thin film
coatings. Then in January 2000, they
announced they would purchase a rival
fiber optic components and modules maker
called E Teet Dynamics for about $15
billion in stock. Despite the high
price, analysts hailed the deal because
it gave JDSU a dominant twothird share
of the fiber optics market. In other
words, it now made them the Intel of
fiber optics. Yes, they are like Intel.
Now, in March 2000, the NASDAQ hit an
intraday high of 5,132.
JDSU's buoyant stock hit an all-time
high of $293,
rising 82% in just 3 months from
January. It had risen a ridiculous 220
times from what it was 4 years prior.
March marks the beginning of the end for
these twin bubbles in the late 1990s.
From there, the NASDAQ index began to
descend as various gimmicky.com stocks imploded.
imploded.
Despite the ongoing.com meltdown, the
fiber optics stocks stayed hot for a
while. One infamous IPO in late July
2000 valued a fiber optics component
startup called Corvvis at 11 billion
despite them not having a single
shipping product. Founder David Huber's
stake alone was worth 1.86 billion.
Calovven decided then to step down from
JDSU. He would later get into racing
cars. Strauss took over as the company's
CEO and swiftly led the company to its
biggest buy yet. In mid July 2000,
JDSU's market cap was $110 billion, more
than GM and Ford combined. Their
trailing 12-month revenues were just
$1.4 billion.
And the company showed a net loss of
$170 million, though most of that was
from advertising the company's goodwill
from its purchases.
Strauss then negotiated to buy rival
optics fiber components maker SDL for a
staggering $ 41 billion in stock or
about 76 billion today. It was the
largest non- Telecom technology company
acquisition up until then. This
particular transaction was wild. SDL's
trailing 12-month revenues were only 187
million. Profits just about 25 million.
Paying $41 billion for so little, even
if it was all paid in stock, boggled the
mind. Business Week ran an article with
the title, "Is JDS Unif bonkers?" and
asked if management had lost their
minds. If your stock is overvalued,
which it was, "You use it to buy
undervalued assets. Don't use it to buy
other overvalued companies." In late
July 2000, JDSU's split adjusted stock
price had retraced a bit from the peak
of an adjusted $293,
but was still at a pretty good $135.
Unfortunately, worse things were yet to come.
come.
Telecom capex continued in the third
quarter of 2000, but the engine was
sputtering. Lucen and Nortell had
controversially lent billions to
telecoms to buy their equipment. When
those small companies collapsed due to
the fiber overbuild that came back to
bite them, the pain took no time to
filter down to JDS Unifase. By late
September 2000, the stock had fallen to
$95. By December, it had spiraled down
to just $41 and now the alarm bells were
seriously ringing. In January 2001, JDSU
laid off 700 contract workers in Ottawa.
A month later, Nortell abruptly cut
their forecasts. So then, a week later,
in February 28th, JDSU announced a 10%
cut to their workforce, 3,000 layoffs.
The stock price fell to $29. A company
spokesman said, "We still feel very
strongly about the need and demand for
bandwidth. We still think it's there."
Unfortunately, he was wrong. Demand for
fiber optic components was vanishing
faster than an ice cube in the Sahara. A
week later in March, JDS admitted that
they will miss their third and fourth
quarter earnings forecasts. Internally,
it was chaos down and up the ranks.
Employees were hearing about job cuts on
the radio before the managers can be
informed. Considering that the rest of
the telecom industry was imploding too,
Nortell alone would cut an astonishing
50,000 jobs in just the year 2001, it
caused a lot of anxiety.
In late April 2001, JDSU announced that
they would cut another 5,000 jobs, half
of which would be in Canada, where JDSU
had many core research and engineering
staff. The stock continued its march to
Hades by falling 14% to $20.
In July 2001, they declared a $50
billion loss for the 2001 fiscal year
and another 7,000 job cuts, a third
round on top of what had already been
announced. At their peak, JDSU had
29,000 employees globally. 16,000 of
those were chopped in one year. By late
August 2001, the stock price was $7. A
year later, in August 2002, it reached
$2. Even as the stock stumbled, analysts
declared their faith in the seemingly
quotequote stratospheric future of
communications and the internet.
Meanwhile, management rad in millions
selling their shares.
I know that some today argue that there
are good and bad bubbles, pointing to
the telecom bubble as one of the good
ones because it created internet
capacity. Considering the social and
economic annihilation of entire
communities in Canada and the United
States that the bubbles burst left
behind, I am skeptical.
For the next few years, JDSU continued
to show losses. A year after their first
immulation, they announced another $8.7
billion loss for the fiscal year 2002,
mostly due to billions in write downs
from the 17 or so acquisitions made
during its wild and woolly years.
Fortunately, the company did almost all
of its purchases using stock rather than
cash. They had no debt and were cash
flow positive. So ultimately after
cutting enough jobs, the company managed
to limp through these hard times.
In August 2003, Strauss retired. His
legacy for Canada is mixed. He helped
create thousands of highpaying
engineering jobs in his adopted home of
Ottawa, but under his purview, JDSU
hyped the fiber bubble and then
annihilated billions in people's life
savings and cut 95% of the jobs in said
adopted hometown.
The fiber overhang meant that JDSU can
no longer rely on new fiber component
sales. So they pivoted to providing
testing and maintenance services for
those customers with the 2005 purchase
of Actturn. By now revenue had shriveled
to just $166 million. The stock plopped
to $150.
But this pivot was the right move. And
over the next decade, the company
quietly expanded its portfolio of stable
but boring service businesses. Stuff
like fiber optics cleaning and network
testing services. The latter became
especially attractive as cloud computing
became more prevalent. They also seem to
have spent some money to get into the
business of anti-counterfeit stuff like
those special colorshifting optical inks
that you can see on currencies. Pretty cool.
cool.
In September 11th, 2014, JDSU announced
that they were going to spin off their
optical components and commercial lasers
businesses. The old JDSU entity, later
to be renamed to Vavi Solutions,
retained the various network service,
installation, and test businesses that
helped the company survive the past 10
plus years. Management explained the
move has given investors more clarity.
JDSU had been three businesses under one
umbrella and it was clear on the
investor call that management valued one
of those three less than the other two.
The optical hardware business was strong
but faced heavier competitive pressures
and since it was also a manufacturing
business with fabs and capital
requirements for them and all that its
profit margins were rougher. The various
network and service enablement
businesses on the other hand are more
software ccentric with 60% gross
margins. Management also felt that great
opportunities laid ahead for that thanks
to the ongoing rise of the cloud and
data center. Thus, the laser business
was being sent off on a raft to fend for
itself. Fortunately, without any debt,
it is most telling that the old JDSU
management opted to stay with Vavi.
In February 2015, it was announced that
the spin-off would be named Lumenum.
Lumenum's core business involved lasers
for fiber optics. But for a long time,
that wasn't growing much. So, the
company experiences next big expansion
thanks to the rise of 3D sensing. These
rely on small near infrared dots
generated by an array of compact little
lasers called Vixels. The technology
first gained some prominence when
Microsoft used it to produce the Xbox
Connect, a device that let you use
gestures to game. I never used it, but I
heard it was fun. JDSU talked it a lot
in 2011. But the real turning point
occurred when these 3D sensing tools
were adapted to do facial recognition on
the mobile phone. These first broke
through with Apple's iPhone 10, which
hit the market in November 2017.
Lummentum supplied all the vixels for
those devices. Though the Certino fruit
company quickly diversified to a second
supplier by investing in Finnar, now
part of another optics company called
Coherent. The company continues to work
on Vixels for LAR production, extoing
their potential benefit for applications
like self-driving.
But 3D sensing is not behind Lum's rapid
growth as of late. AI is that reason of
course, but what are they doing that is
so special? I am no expert. So, as I
mentioned earlier, I invited the
anonymous author of the irrational
analysis newsletter to help. What
follows is his analysis, though I made
some edits so I can narrate it.
Lumenum's value in the AI boom is tied
to a shortage of highquality, as in low
noise, high density photons. Let's start
with understanding density. co-ackaged
optical or CPO systems like those shown
by Nvidia and Broadcom want ultra high
power continuous wave lasers. For the
lasers inside traditional optical
transceivers, it was historically
sufficient for them to have an optical
power of around 20 m. C-ackage optics,
however, require much higher optical
output power per laser, typically in the
range of 200 to 400 m. This is because
CPO solutions use external pluggable
light sources called ELSFP modules
rather than have them directly
integrated into the system.
Disagregating the lasers from the rest
of the system makes sense because these
lasers are typically the most unreliable
and failureprone element of an optical
system. They're also very temperature
sensitive necessitating precise active
cooling using thermo electric coolers.
So moving the lasers to a dedicated
pluggable module the same form factor as
a transceiver greatly helps reliability.
However, there is a cost drawback with
doing it this way particularly involving
the fibers. The fibers connecting a
continuous wave laser and the optical
engines must be of a particular type. Polarization
Polarization
maintaining PM fibers. What are those?
Imagine a pair of sunglasses. Sunglasses
reduce glare by filtering out light that
is in the wrong quotequote orientation
aka polarization.
Polarization maintaining fiber is
designed so that the light coming in
stays on the same axis. Bad polarization
leads to massive optical losses. So PM
fibers intentionally build in internal
stress elements to force the glass core
into a desired shape. This helps ensure
that the most light possible makes it
from the laser to the silicon photonix
optical engine. It also makes PM fibers
expensive. So people want to minimize
how much they use it for each system.
Attaching those PM fibers is also
expensive and quite painful. Typically
the attachment process means aligning
the fiber as well as a micro lens to a
target surface, usually an indium
phosphide laser or a silicon photonix
photonic integrated circuit. Once
aligned, we use epoxy to hold it all
together. This is not easy.
So more PM fiber attachments mean more
cost plus additional potential points of
failure. So using four 100 matt lasers
is not the same as one 400 matt laser.
Economic and reliability requirements
place significant pressure upon a laser
vendor like Lumenum to deliver as much
power as possible within a single laser
diode. Lumenum's ultra high power lasers
deliver up to 24 dBm optical output per
wavelength at 50° C. That 24 dBm
converts to about 250 ms of output
power. This is the power that makes it
onto the fiber. Every coupling interface
induces optical loss. Thus, a
non-trivial amount of light is lost
between the indium phosphide laser and
the fiber within the ELSFP module.
In general, the ultra high power class
of lasers is around 350 to 400 m optical
output power at facet. meaning before
any coupling loss. This corresponds to
25.5 to 26 dBm.
As of this writing, only three companies
claim to have ultra high power class of
continuous wave lasers. Those are Lum,
Coherent, and Applied Opto electronics.
Broadcom has an internal laser division
that pro presumably has a solution, but
they have not advertised their precise
optical power specs.
But power is not the only story. The
quality of the photons is also critical
for modern CPO systems, particularly
those using ring modulators. Two key
specs determine how quoteunquote noisy a
laser is. Line width and relative
intensity noise or Ren. Let us go
through those one by one.
Lasers don't stay at one exact
frequency. They actually jitter across
the optical spectrum. Line width is the
width of that jitter, thus telling you
the laser's sharpness. The sharper the
better. It is typically measured at the
3dB full width half magnitude point. So
how wide the plot is when it is at 50%
of peak power. Many factors can
influence a laser's line width,
including thermal noise, power supply
noise, or even optical reflections back
into the laser itself. But the most
important factor is the quality of the
laser's design as well as the
manufacturing process that made it.
Achieving a line width of 1 megahertz or
less is highly desired by the co-ackaged
optics boom fueled by AI requirements.
It is extremely difficult to hit this
number at high power because of how DFB
distributed feedback lasers work. Unlike
normal lasers, DFB lasers use many tiny
mirrors called gradings that bounce
light back and forth in a feedback loop.
This loop raises the laser's power while
also sharpening/conentrating
light. The issue at higher power levels
is that the DFB laser becomes much more
sensitive to very small imperfections in
the grading design or fabrication. The
effects of any imperfections amplify
creating a much noisier laser.
In addition to line width, the other
measure of noisiness is ring or relative
intensity noise. Is it RN? I like to
call it re. If line width is noise going
side to side, then REN is amplitude
noise. Laser power intensity fluctuates
up and down randomly over time. Ren
measures that intensity noise. Read
measurements are also in frequency
domain, but the final number is an
average between two values. The lower
bound is defined by the optical systems
tolerance to low frequency noise and is
typically around 0.1 to 10 MHz while the
upper bound is often determined by the
Nyquist frequency of the communication
system. So a system designed for 112
Gbits per second Ethernet might measure
read from 10 MHz to 28.5 GHz.
Again, many factors contribute to laser
ring, particularly the quality of the
electrical driver and thermal control.
Yet, the most important factor again is
the quality of the laser design and
fabrication itself.
Lum is the only independent laser vendor
who can achieve these two key low-noise
aspects, line width and re while also
achieving high power. This makes it a
key player in the AIdriven co-ackage
optical revolution.
As of this writing, Lumenum, the
company, has a market cap of around $50
billion. Quite volatile. So, be careful
about that number. And it was just
announced as of this writing that
Lumenum and Nvidia are collaborating to
enable next generation AI silicon
photonic systems along with building new
manufacturing capacity.
I don't know where the AI data center
boom goes from here, but I am fascinated
to see how it has empowered entire
technology fields or given them new
life. Fiber optics is back again after
20 plus years in the goolog. Let's go.
All right, everyone. That's it for
tonight. Thanks for watching. Subscribe
to the channel. Sound for the Patreon.
And thanks again to Irrational for
spending time with me on this. All
Click on any text or timestamp to jump to that moment in the video
Share:
Most transcripts ready in under 5 seconds
One-Click Copy125+ LanguagesSearch ContentJump to Timestamps
Paste YouTube URL
Enter any YouTube video link to get the full transcript
Transcript Extraction Form
Most transcripts ready in under 5 seconds
Get Our Chrome Extension
Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.
