Charles Fan and the trouble with starting over

Charles Fan’s early work asked a wonderfully inconvenient question: could several ordinary machines behave like one dependable system? It is the kind of question that sounds tidy in a research proposal and becomes considerably less tidy when a machine stops cooperating. Fan made a career in the space between those two situations. The customer, sensibly, would prefer never to notice the difference.

At Caltech, he worked on distributed computing with Shuki Bruck. Their collaboration helped produce Rainfinity, a software company that took research about reliability into the commercial world. Fan later built inside EMC and VMware, joined Cheetah Mobile as chief technology officer, and returned to startup life with MemVerge. The company he now leads is working on another everyday failure of continuity: an AI assistant that meets you anew each time you open a conversation.

There is a temptation to arrange that career as a parade of fashionable technologies. Networking. Storage. Cloud. AI. A more useful reading is that Fan repeatedly works on the connections that make a system usable. Who keeps the files available while they move? Who helps applications use a new kind of memory? Who keeps the context when the conversation ends?

A company before the doctorate

Fan graduated from Cooper Union in 1995 with an electrical engineering degree, summa cum laude. He earned his master’s at Caltech in 1996 and his doctorate in 2001. The dates matter because his commercial work began before the academic work was finished. Rainfinity opened a research office in Pasadena’s Old Town in September 1998. The laboratory and the business were running on overlapping calendars.

The research behind the company involved a Reliable Array of Independent Nodes, or RAIN. A cluster could route work around a failed component rather than allow that component to interrupt the whole service. In a Caltech demonstration, one of five interconnected servers could be unplugged while an onscreen movie continued playing. For the audience, the achievement was a film that stubbornly refused to become interesting.

That is an underappreciated pleasure of infrastructure. The engineering succeeds when the drama disappears. An application continues, a connection survives, a user carries on. The people who built the system understand how many things might have gone wrong; everyone else sees a normal afternoon.

Rainfinity’s first product, RainWall, arrived in 1999. The company established its headquarters in San Jose while retaining its research arm in Pasadena. Fan was becoming an entrepreneur without having to abandon the problems that interested him as an engineer. Research gave him something to build. A company gave it somewhere to go.

The files had to keep moving

Rainfinity eventually moved into file virtualization. Its RainStorage software made different storage systems appear together as a single logical resource. The practical benefit was continuity: people could still access files while administrators moved them or changed the equipment underneath. It addressed the awkward interval when a system needs maintenance but the work using it has no intention of taking a holiday.

Fan recalls a difficult pivot after the internet bubble burst and a partner became a competitor. He and two engineers worked on a new storage product with cash running short. Three purchase orders arrived on the planned shipping day. It is a compact reminder that a startup’s elegant technical argument eventually has to become somebody else’s order.

On August 17, 2005, EMC announced its agreement to acquire Rainfinity. The attraction was file virtualization across heterogeneous storage environments, including Windows, Linux and Unix. EMC’s subsequent annual filing put the cash purchase at approximately $90 million. That figure belongs to the company transaction; the more revealing detail for Fan’s story is what customers were buying: a way to alter their infrastructure without interrupting access to their information.

The transaction brought Fan into EMC. It also changed the scale of his work. Selling a product and building an organization are different kinds of engineering, even if both reward an ability to understand how the pieces depend on one another.

Building inside somebody else’s company

At EMC, Fan established the China R&D center. At VMware, he became a senior vice president and general manager, founding the storage business unit that developed Virtual SAN. He led teams working on storage and big data. The work had moved beyond a startup product into the organization needed to develop, ship and support infrastructure for other businesses.

A title can make that transition look effortless. The substance is more demanding. A team needs a shared direction, enough technical depth to make difficult decisions and a way to connect its work with customers’ existing systems. Fan’s career passes through both the small company that needs its next order and the larger company that needs a new business to function within it.

In February 2016, Cheetah Mobile appointed him chief technology officer. His remit covered global research and development and the creation of a larger Silicon Valley presence. The move put an infrastructure engineer inside a mobile internet company with a geographically distributed technology operation. He would lead work across AI, big data and cloud, as well as the engineering platform team.

The connection across those roles is organizational as much as technical. Fan repeatedly took responsibility for building something within a changing environment: a research center, a business unit, a new engineering presence. His later return to a startup carried experience of how ideas travel through institutions. A good invention still needs colleagues, customers and a place on the budget.

The opening in the hardware

MemVerge began in 2017 with Fan, Bruck and Yue Li. Bruck had been Fan’s Caltech adviser and Rainfinity collaborator; Li brought research experience in non-volatile memory. The founding team joined an old working relationship with a newer line of research. There was history around the table, and a fresh technical opening in front of it.

Intel’s Optane SSD, released that year, used 3D XPoint media. Fan and his colleagues expected persistent-memory modules to follow. The prospect was a different relationship between memory and storage. Working memory could be fast; storage could keep data. New hardware offered a way to bring those properties closer together, but ordinary applications would need software to benefit from it.

MemVerge’s answer was Big Memory software. Its Memory Machine combined memory technologies with services such as snapshots, replication and tiering. The aim was to let existing applications use new infrastructure without an expensive rewrite. Fan’s experience with virtualization had a new setting: software would manage complexity so that the application could keep doing its job.

Customers get a vote

Fan’s description of company culture gives a glimpse of how he wants that work conducted. In 2023, he described an OPEN culture: originality, a positive working environment, external demands, urgency and open communication. Customers and partners were meant to influence priorities. Technical discussions were meant to welcome different perspectives, with ideas judged on their merits.

Those principles sit beside a real tension in infrastructure startups. A company can see a technical opening before customers have settled on exactly how to use it. That leaves the founder with two jobs: develop the capability and learn where it earns its place. Enthusiasm for an architecture does not finish the second job.

Fan’s public presentations have repeatedly made the architecture available for discussion. In December 2020, he introduced Big Memory at Tech Field Day. In March 2023, he presented a CXL computing-cell architecture, a memory-centric design for data-intensive applications. CXL, short for Compute Express Link, became another setting for his work on how computing resources connect and share capacity.

There is a social dimension to this supposedly impersonal field. Hardware vendors, system builders, software developers and customers all need parts of the arrangement to work. Fan’s task as CEO includes keeping those conversations connected. A diagram may contain clean arrows. The business behind it contains people with different priorities.

A conversation with a past

In September 2025, MemVerge launched MemMachine, an open-source memory layer with enterprise offerings. The subject had moved closer to the person at the keyboard. AI applications needed a way to retain useful information across sessions, agents and models, rather than treat each interaction as a fresh introduction.

“AI without memory is incomplete.”

Charles Fan, September 2025

By December, MemVerge was announcing additional memory capabilities, developer tools and integrations. Fan emphasized third-party builders choosing the system for their own agents. The proposition depended on what other people could make with it. An open-source project has to become useful beyond the organization that released it.

In April 2026, Fan appeared among the authors of a MemMachine research paper. It described a system that preserves whole conversational episodes and retrieves surrounding context, reducing reliance on lossy extraction. The detail is revealing: a remembered fact can become less useful when severed from the exchange that explains it. Saving a fragment and preserving its meaning are separate engineering decisions.

Then came MemoryBox’s beta announcement in August 2026. MemVerge was offering individuals a way to keep context persistent and under their control across AI tools. The setting had changed from administrators managing storage to people managing the information that makes an assistant useful. The irritation was easy to recognize: explain your goals, open another conversation, explain them again.

Fan’s career now reaches from the reliability of clustered machines to the continuity of a personal conversation. Those are different problems, with different demands. They share an insistence that useful work should survive a change in the system around it. After years spent helping computers accommodate one another, he is asking them to carry a little more of the conversation forward.