Elastic AI Infrastructure and Our Bitfusion Partnership + Investment

I'm surprised that folks read blogs anymore; our brains seem to have been acculturated into handling at best, an expanded 280 characters and even a Tweet storm. But as an avid reader, I'd like to believe there is some valuable expanse between great novel (recently I've read Italo Calvino's Invisible Cities and Lee, Min Jin's Pachinko) and 140 characters. Maybe I can think of this as a Tweet storm with paragraphs and punctuation.

I could write about the long arc of hardware abstraction and software virtualization, but that's been done. The more interesting thing I’d like to dwell on briefly are the lessons I observed in watching software companies navigate complex hardware and application ecosystems and then explain 1) why and 2) for what we invested in Bitfusion, a team/technology/market focused on elastic AI infrastructure for any cloud -- enabled by GPU (and remote attached FPGA or ASIC) virtualization.

Anecdote 1:

I was at VMware from 2006-2010 and observed with many others the evolving dynamic with Intel, AMD and server vendor partners. At different times, there were different views about the impact x86 virtualization was going to have on the chip vendors. Would it increase utilization, but thereby stall sales and defer purchases? Why would a chip vendor want this to happen; was it fundamentally threatening to their new sales? What I observed was that the tide was not going to be stemmed, the benefits of VMware's core product were too readily proven and provided in too stable a way for it to be stopped -- the most undisruptive, disruptive technology in a while in that space. As the segment evolved, it seems to me the chip vendors accepted this market swell and quickly turned that momentum into selling larger, multi-core chips, evolving their licensing policies (per-core) (you can see Nvidia doing this now for different reasons, but with comparable pressures to protect revenue), and also marketing the differentiating benefits of "hardware assist" (I remember the first time I learned what a nested page table was). The point is chip and hardware vendors themselves do initiate and at times are forced to find a way to take the software benefits of virtualization, emulation, hardware abstraction, and the ensuing application ecosystem benefits and turn them to an advantage.

Market observation 1:

This is a big and relevant market. Step 1: Simply look at the 5-year stock price chart for NVIDIA. Step 2: Kick yourself for not buying sooner, as it's up 18x. Also, AWS is competing with everyone. After they acquired Whole Foods and entered into direct competition with a customer, Wal-Mart announced they would build a GPU farm with Nvidia chips to rival the market leader (incidentally, Bitfusion also works well with AWS ElasticGPU). The point is GPU virtualization is an important enabling technology here for anyone making GPU server purchases, if you care about good utilization of your asset, cost savings, and running your workloads more quickly and efficiently.

Anecdote 2:

I met Bob Wiederhold first when he was CEO at Transitive. Part of Transitive's interesting history is that powered Rosetta, an invaluable bridge for Apple to transition from software built for their PowerPC processors to run on Intel-based Macs. I think Apple benefited quite a bit from their partnership with Transitive, perhaps most importantly in assuaging fears about a transition to Intel-based Macs. Then Boot Camp, Parallels, and VMware Fusion arrived, which in turn influenced the mainstream adoption of Macs in personal and office work culture (that and the Get a Mac advertising campaign with John Hodgman as PC and Justin Long as Mac - loved those). But binary translation ain't easy. Postscript: IBM acquired Transitive in June 2009 and merged the company into its Power Systems division.

The point is...

Market observation 2: This is just an inherently important footprint. The shim layer between hardware and OS or virtualization was monetized well, a few times just in my own memory by entrepreneur Benny Schnaider (Ravello, Qumranet) and many companies that were building filter drivers for desktop applications and their OS (Softricity, Thinstall). The "how" Bitfusion does what it does with its FlexDirect - intercepting Cuda or OpenCL calls at run-time, in one key example of a very extensible architecture - can be inherently valuable, just by the nature of its footprint.

What attracted us to Bitfusion:

1. Samsung SDS is an $8+bn revenue enterprise IT solutions business and we have lots of GPUs and run data centers and develop and sell software and solutions across many vertical industries. Why not benefit from increased utilization and cost savings at large scale.

2. Team - we met Subbu Rama and really enjoyed his enthusiasm and vision. They hired Michael and we've already appreciated his depth in the space, leadership, and professionalism. We met Mazhar, who is just flat-out, technically impressive. And Maciej, whom we trust to make sure things happen on time and on budget.

3. Technology - we paid for a PoC and we saw the product in action, providing the benefits we desired.

4. Market - we liked the macro trends and the strategic footprint for the product.

5. Why we invest - in this case, as in many cases, we believe in all three of the above.

6. How we invest - we've not only become equity holders in the business, we've got a strategic partnership that places trust in the product and team, so we can help ourselves as customers and our important, external customers, too.

That's really it. I’m interested in advising companies, and can do that in i) strategic partnership, in ii) venture investments and iii) through acquisitions, all at Samsung SDS. I'm based in the Bay Area, and if you want to share notes on AI, blockchain, cloud or i), ii), iii), DM me on Twitter @andrewdlee, LinkedIn, or Instagram @andrewleenyc

Finally, note: if I've invoked your personal name, it's because I've enjoyed our interactions and hope the feeling is mutual.



▶    The contents are protected by copyrights laws and the copyrights are owned by the creator.
▶    Re-use or reproduction as well as commercial use of the contents without prior consent is strictly prohibited.