Re-engineering the Semiconductor Supply Chain with Intel CEO Lip Bu Tan

2026-06-18 - 45 min - source - Read full transcript

Sarah Guo (host)Elad Gil (host)Lip Bu Tan

Key insights

Tan structured Intel's turnaround as a sequenced 'crawl, walk, run' plan, fixing the balance sheet and culture before attempting new product or foundry ambitions.

He treated the U.S. government's equity stake, Nvidia's $5 billion investment, and SoftBank's involvement as the 'crawl' phase needed to strengthen a 'horrible' balance sheet, alongside simplifying the product line and driving more accountability and faster decision-making than Intel's prior bureaucratic culture allowed. Only after that stabilization does he see Intel able to 'walk' toward next-generation products and eventually 'run' into new markets like physical and agentic AI.

intel-turnaround

Tan defused a direct threat from President Trump to resign over a conflict-of-interest allegation by securing a personal meeting to explain himself before any decision was finalized.

After being asked early one morning to resign, Tan says he first decided he did not personally need the job and was doing it purely to save Intel, which let him set the personal stakes aside. He then got a meeting within days, explained his path from Malaysia to Singapore to MIT to the U.S. with no time living outside the country, and says Trump listened and gave him the chance to stay.

intel-turnaround

Intel is deliberately importing startup speed and modern AI tooling into what Tan calls a legacy 'spreadsheet company.'

He restructured engineering so all engineers report to him directly from day one, is recruiting younger software and AI talent to complement a team whose average age is in the late 40s and 50s, and describes informally learning current AI/ML tools from his own son. He frames the goal as making Intel's decision cadence resemble a multiple of startups rather than one large bureaucratic company.

intel-turnaround

The U.S. government's equity stake in Intel mirrors the sovereign-backed model that built TSMC, and Tan argues that model is the correct playbook for capital-intensive fabs, not an aberration.

He explicitly compared the government's Intel stake to Taiwan's early state backing of TSMC and cited Japan and Singapore as having taken similar approaches, framing advanced fabs as infrastructure that government capital should support alongside strategic investors like Nvidia and SoftBank, especially as venture-scale checks (some funds now writing $1 billion rounds) increasingly can't cover fab-scale capital needs alone.

semiconductor-supply-chain

The Intel-Musk TerraFab partnership exists because both parties independently concluded that semiconductor manufacturing capacity has not kept pace with AI-driven silicon demand.

Musk is building TerraFab as a dedicated facility for his own robotics and vehicle silicon needs; Intel is sharing process and technology to help him reach production faster. Tan praises Musk as 'unconventional,' someone who questions every traditional fab practice, and says the two meet weekly, calling the collaboration refreshing after decades of working within established industry norms.

semiconductor-supply-chain

Rising demand for agentic AI inference is shifting the historical training-era CPU-to-GPU ratio from roughly 1:8 toward 1:1, reviving CPU demand.

Tan says AI developers report that CPUs handle reinforcement learning and the orchestration of many agents better than GPUs, driving CPU demand up even as GPUs remain dominant for training. He frames this shift, alongside the broader move to agentic and physical AI, as central to Intel's near-term product recovery.

ai-hardware-demand

Tan expects most crowded AI infrastructure categories to consolidate to one or two durable winners, following the pattern he saw in the internet era.

He compared the coming AI infrastructure shakeout to how Amazon and Netflix became durable 'real applications' while many other internet-era companies stagnated, were acquired, or disappeared. His framework for judging any given AI application bet is whether the underlying problem is real, whether there's a viable partner, and how big and sustainable the specific application is before doubling or tripling down.

ai-hardware-demand

Intel's foundry business depends on winning customer trust through yield, defect density, and cycle time execution, not competitive process nodes alone.

Tan describes foundry as 'a service business, a trust business' where poor yield directly costs a customer revenue and can end the relationship. He acknowledges Intel remains 'very distant' from TSMC on performance today but is investing in the fundamentals - the right IP for each customer segment (e.g. low-power IP for mobile), yield, defect density, and reliable cycle time - as the precondition for closing that gap.

foundry-strategy

As traditional silicon process scaling runs out of headroom, Tan is betting on new substrate materials and advanced packaging as the next performance lever.

He traced Intel's roadmap from 18A into 14A (1.4nm) toward planned 1nm and 0.7nm nodes, but noted that further shrinks get harder and more expensive without proportional cost or area reduction. In response he is hiring materials scientists and personally investing in gallium nitride, silicon carbide, and indium phosphide semiconductor materials, plus glass (3DGS) and artificial diamond (Diamond Foundry) as insulators for next-generation advanced packaging, alongside a new India-New Mexico packaging manufacturing program.

foundry-strategy

Tan's venture method is to identify a validated bottleneck in the value chain and back the company solving it, then get that company its first hyperscale customer to enable scaling.

Examples he cites include Credo Semiconductor and Celestial AI for the interconnect/optical bottleneck in AI clusters, InPhi (indium phosphide materials, later acquired by Marvell) and Empower (DC-to-DC power conversion, acquired by Analog Devices) for materials and power bottlenecks. He says landing one hyperscale customer willing to pay millions of dollars and grant warrants is the formula that lets a bottleneck-solving startup scale, and that he still personally recruits and negotiates these relationships rather than delegating to a search firm.

venture-investing-philosophy

About nine of every ten companies Tan backs pivot their business plan mid-course, so he prioritizes team quality and open-mindedness over the original plan.

He said he looks for an entrepreneurial team rather than a single founder and prefers founders who listen to customer and investor feedback and draw their own conclusions, rather than founders who insist on executing a fixed thesis, because markets change enough that the original plan rarely survives contact with the market. His broader track record across roughly 200 semiconductor investments includes 159 IPOs and 126 M&A exits, with 38 percent of those investments in the U.S.

venture-investing-philosophy

Companies

Intel - The company Tan took over as CEO 14 months before this episode; central subject of the whole conversation.
Cadence - Tan's prior company - 13 years as CEO plus 2 as executive chairman (15 total), during which he says he delivered 76x return as CEO and 85x total to shareholders.
Walden International - Tan's venture capital firm, referenced as the base for his roughly 200-company, 20-plus-year semiconductor investing track record.
Nvidia - Jensen Huang, described as Tan's old-time friend, personally invested $5 billion in Intel, which Tan says has since grown to a paper value of $25 billion.
SoftBank - Masayoshi Son, whose board Tan previously sat on, helped strengthen Intel's balance sheet alongside the government stake and Nvidia's investment.
TSMC - Cited as the government-backed model Intel's own U.S. government equity stake is modeled on, and as the foundry leader Intel is still 'very distant' from on yield and performance.
Tesla / TerraFab (Elon Musk) - Musk is building a dedicated fab (TerraFab) for his own silicon needs; Intel is collaborating with him, sharing process and technology, which Tan calls a refreshing, unconventional partnership.
Credo Semiconductor - A Walden portfolio company Tan backed to address the interconnect bottleneck in AI clusters.
Celestial AI - A Walden-backed optical interconnect company, part of Tan's bet that optical technology becomes critical as cluster interconnect speed matters more.
Synopsys - Cadence's chief EDA rival, cited as also pursuing agentic AI in design tools and as having received Nvidia investment; also credited with acquiring Ansys.
Ansys - Named as the simulation company Synopsys acquired to move into full-system design.
InPhi - A Walden-backed indium phosphide materials company, later acquired by Marvell.
Marvell - Acquired InPhi, one of Tan's materials-bottleneck investments.
Empower - A power-management/IVR startup Tan backed to address the DC-to-DC voltage conversion bottleneck; acquired by Analog Devices.
Analog Devices (ADI) - Acquired Empower, Tan's power-management bottleneck investment.
3DGS - A glass-substrate packaging venture Tan invested in, betting on glass as a heat insulator for advanced packaging.
Diamond Foundry - An artificial-diamond materials company Tan invested in as another advanced-packaging insulator bet.
AMD - Lisa Su, described as a good friend, cited in a market-cap comparison (~$800 billion) alongside Intel's own (~$600 billion at the time).
Broadcom - Cited alongside TSMC as a roughly $2 trillion market-cap semiconductor company, part of Tan's point about how the industry's valuation ceiling has changed.
Samsung - Named as one of the few corporate investors (with Arm and SoftBank) willing to back Tan's semiconductor deals 15-20 years ago when VCs would not.
Arm - Same point as Samsung - an early corporate co-investor in Tan's semiconductor bets when mainstream VC avoided the sector.
Amazon - Cited alongside Netflix as an example of an internet-era company that became a durable 'real application' rather than fading or being acquired.
Netflix - Same point as Amazon - used as the model for what a durable, focused AI application should look like.
Anthropic - Named alongside OpenAI as an example of a startup moving at 'the speed of light' to become a dominant player by finding a more elegant approach.
OpenAI - Same point as Anthropic - cited as a fast-moving startup example of how a smaller company can out-execute an incumbent.
India (government) - Intel announced a new advanced-packaging manufacturing program with the Indian government, alongside a New Mexico facility.

Techniques and frameworks

Crawl, walk, run - Tan's repeated framing for sequencing Intel's turnaround: stabilize the balance sheet and culture first (crawl), then rebuild products (walk), then expand into new markets like physical and agentic AI (run).
Bottleneck-driven investing - Tan's venture method: identify a real, customer-validated bottleneck in the semiconductor value chain (interconnect, optical, materials, power conversion) and back the startup solving it, rather than investing off a broad thesis.
First hyperscale customer as the scaling formula - Tan's playbook for early-stage semiconductor startups: land one hyperscale customer willing to pay millions of dollars and grant warrants, because that single reference customer is what enables the company to scale.
Team over plan - Tan's investment filter: back an entrepreneurial team rather than a single founder and prioritize open-mindedness, since roughly nine of ten companies he backs change their business plan mid-course as the market shifts.

Summary

Sarah Guo and Elad Gil host Lip Bu Tan, the former longtime Walden International venture investor and Cadence CEO who took over as Intel CEO 14 months before this recording, for a conversation about rebuilding Intel and re-engineering the broader semiconductor supply chain. Tan opens by explaining why he took "the hardest job in the industry" at 66: Intel's importance to both the semiconductor ecosystem and the United States. He recounts being asked directly by President Trump to resign over a conflict-of-interest concern, and describes defusing it by first deciding he didn't personally need the job, then securing a meeting within days where he explained his path from Malaysia to Singapore to MIT to the U.S. and was given the chance to stay. From there he lays out his turnaround framework - "crawl, walk, run" - starting with strengthening a "horrible" balance sheet (via the U.S. government taking an equity stake, a $5 billion investment from Nvidia's Jensen Huang that has since grown to a paper $25 billion, and help from SoftBank's Masayoshi Son), simplifying the product line, and pushing faster, more accountable decision-making by having all engineers report to him directly.

A recurring thread is how AI demand is reshaping Intel's product mix and go-to-market. Tan says agentic AI and inference workloads are pushing the historical training-era CPU-to-GPU ratio from roughly 1:8 toward 1:1, since developers report CPUs handle reinforcement learning and multi-agent orchestration better than GPUs - a shift he calls central to Intel's near-term demand recovery. On the foundry side, he is candid that Intel remains "very distant" from TSMC on yield and performance, framing foundry as fundamentally "a service business, a trust business" where a bad yield directly costs a customer revenue. To close that gap he's investing in new substrate materials (gallium nitride, silicon carbide, indium phosphide) and advanced packaging (glass via 3DGS, artificial diamond via Diamond Foundry, plus a new India-New Mexico packaging program) as traditional process shrinks toward 14A, 1nm, and 0.7nm get harder and more expensive without proportional gains. He also details Intel's TerraFab collaboration with Elon Musk, who is building a dedicated fab for his own robotics and vehicle silicon needs; Tan calls Musk "unconventional" for questioning every traditional fab practice and says the partnership - built on weekly working sessions and shared process technology - has been refreshing.

Tan repeatedly frames the U.S. government's equity stake as consistent with, not a departure from, how other advanced economies built their chip industries: he compares it directly to Taiwan's early government backing of TSMC, and cites Japan and Singapore as having taken similar approaches, arguing capital-intensive fabs are infrastructure that deserves government and sovereign capital support alongside strategic investors. The hosts press him on where compute equilibrium settles - centralized data centers versus edge and client compute - and Tan says the answer follows the application, not a fixed architectural preference, drawing an analogy to how Amazon and Netflix became durable "real applications" in the internet era while many peers stagnated or were acquired; he expects most crowded AI infrastructure categories to consolidate similarly to one or two winners.

The conversation closes on Tan's venture investing philosophy, honed across roughly 200 semiconductor investments with 159 IPOs and 126 M&A exits (38 percent in the U.S.). His method is to find a validated bottleneck in the value chain - interconnect (Credo Semiconductor), optical (Celestial AI), materials (InPhi, later acquired by Marvell), power conversion (Empower, acquired by Analog Devices) - back the company solving it, and then get that company its first hyperscale customer, since one reference customer willing to pay millions and grant warrants is what lets a bottleneck-solving startup scale. He says he still personally recruits every hire rather than delegating to a search firm, and that because roughly nine of ten portfolio companies change their business plan mid-course as the market shifts, he prioritizes an open-minded, coachable team over any specific original plan. He sets his own bar at Intel modestly relative to his Cadence record (76x return as CEO, 85x total across 15 years) - roughly 10x shareholder return over five to ten years - citing Intel's much larger revenue base as the reason the multiple has to come down even as the dollar opportunity stays large.

Notable Quotes

Note: the source transcript (podscripts ASR) carries no speaker diarization. Guest quotes below are attributed to Lip Bu Tan based on first-person content (his own history, decisions, and investments); host quotes are not separately distinguishable and are not quoted here.

"This is an iconic company, and it's so important for the semiconductor ecosystem, and also so important for the United States." - Lip Bu Tan

"It's a service business. It's a trust business... If the yield is not good, they will be toast in terms of revenue or miss." - Lip Bu Tan

"Nine of the ten companies I invest, halfway they change their business plan because market will change. So I like to have an entrepreneur as a team, not just one person." - Lip Bu Tan

"I always look for 10x. You know, being a venture capitalist at heart, you want to look for 10x." - Lip Bu Tan

"He's very, I call it, unconventional. And he basically questioned every step and why this traditional way of doing things. And in some way, it's very refreshing." - Lip Bu Tan, on collaborating with Elon Musk