Next Tuesday, after nearly two years of debate, President Biden is expected to finally sign into law the CHIPS and Science Act, after the US Senate and House passed the bipartisan bill last week. The expansive Act, which runs to 1,054 pages, allocates $247B in spending and tax credits towards strengthening advanced technology capabilities and bolstering science research in the US. Here, we’ll break down some of the more important features of the bill, and discuss what the implications might be.

Perhaps most notable is the bill’s focus on investing in a more self-reliant US chip industry. Today, the US accounts for just 12% of global chip production capacity (down from 37% in 1990). Asia (Taiwan, South Korea, China, Japan), in contrast, represents about 75% of capacity. Furthermore, most of the chips made in the US are not cutting-edge. Taiwan-based TSMC, the industry’s largest contract chipmaker, makes 92% of the world’s most advanced chips, and Samsung makes the remaining 8%.

China’s chip industry is growing very fast, fueled by the US sanctions of the past few years and the Chinese government’s effort to reduce dependence on foreign suppliers. 19 of the top 20 fastest-growing chipmakers are Chinese, and China is the largest market right now for chipmaking equipment. Industry watchers project that just 6% of new chipmaking capacity will be based in the US over the next few years, vs. 40% in China. China’s largest chipmaker, SMIC, is reportedly now producing 7nm chips – more advanced chips than any US or European company. (Samsung is now on 3nm chips, with TSMC expected to follow later this year.)

Division A of the bill – the CHIPS Act of 2022 – allocates $52.7B over 5 years for “creating helpful incentives to produce semiconductors.” $39B of that will go towards financial assistance to “build, expand, or modernize” US chip fabs, with $2B dedicated to mature chips supporting industries deemed critical to “economic and national security interests” (e.g. automotive, military). $11B is allocated for chips R&D and workforce training, $2B will support a university-based network to translate research into military applications (CHIPS for America Defense Fund), $500M will go towards coordinating with foreign allies on security, supply chain and technology standards, and $200M will fund growth of the semiconductor workforce.

Separately, the bill includes a 25% investment tax credit for chip manufacturing investments, worth an estimated $24B. The tax credit is intended to help mitigate the 25-50% cost advantage for Asian-produced chips vs. US chips – 70% of which is attributed to foreign-government subsidies. (The remainder is the differential in cost of labor and utilities.)

While non-US companies are eligible for funds if they manufacture in the US, the CHIPS Act includes guardrails against the money going to China. It prohibits “the recipients of Federal incentive funds from expanding or building new manufacturing capacity for certain advanced semiconductors in specific countries that present a national security threat to the United States” (e.g. China, North Korea, Russia, Iran). Companies receiving CHIPS Act funds are barred from expanding their advanced-chip capabilities (any process newer than 28nm) in China for 10 years.

The Act is having an immediate effect – Samsung and SK Hynix are both now reportedly rethinking their ties to China. It will also impact Intel, TSMC, and Micron, all of which have facilities in China (though they are largely not advanced-chip plants). With CHIPS Act funding, we could see a wave of chipmakers pivot from China to the US to cash in. A number of chipmakers have already signaled their intent to invest in US chip manufacturing, including Intel, Samsung, SK Hynix, TSMC, and Micron.

Division B of the bill – “Research & Innovation” – allocates $170B across a broad range of initiatives over 5 years, for a net $82.5B increase over the prior baseline. The National Science Foundation (NSF) receives the largest allocation with an authorization of $81B (a $36B increase) for its research activities, including a new $20B Directorate for Technology, Innovation, and Partnerships for “critical technologies such as artificial intelligence, quantum computing, advanced manufacturing, 6G communications, energy, and material science.“

As part of Division B, the Department of Energy also gets $68B (a $30.5B increase) for research programs in areas such as clean energy, nuclear fusion and physics, and quantum computing, as well as innovation-oriented partnerships (e.g. energy security, technology transfer). The Department of Commerce gets $11B for 20 “regional technology hubs” and to revitalize distressed communities. Finally, the National Institute Of Standards and Technology (NIST) gets $10B (a $5B increase) to support research and standards development (e.g. in quantum information science, AI, cybersecurity, communications), as well as support manufacturers and combat supply chain disruptions.

While the Act is bipartisan, it has been criticized by members of both sides of the aisle as a “corporate giveaway” or “blank check.” In part, this is based on the sheer size and breadth of the Act, which will make governance challenging and some level of waste inevitable. The idea of giving handouts to profitable private corporations is also leaving a bad taste for some. The benefits of the CHIPS Act, for instance, will likely accrue to US-based Intel (which has been struggling of late) as well as Samsung and TSMC, advanced-chip makers which already have US facilities planned. But the US is far from the only government subsidizing chip production – China, South Korea, Taiwan, Japan, and the European Union are all pouring money into their chip industries. China alone is projected to be investing $150B from 2014-2030.

And while the numbers in the CHIPS Act are apparently large, it’s still not clear whether $53B is enough. Chip manufacturing requires eye-watering levels of capex. A single new fab might run $20B. One industry watcher estimates that growing US chip production capacity by just 5-10% would cost $40B. An additional 90K workers will be needed in the chip industry by 2025, and 300K would be needed for the US to be self-sufficient. The US education system doesn’t seem to be producing enough qualified workers – 40% of high‐​skilled US semiconductor workers were born outside the US.

In addition to the more obvious impact on the US chip industry, this sizable influx of cash and legislative support could act as an accelerator for a range of advanced technologies (e.g. nuclear fusion, satellite internet, clean energy, the race to Mars). Taking money, however, comes with strings – for instance, clawbacks if companies try to use the funds for stock buybacks or dividends. Some of the US’ strings may be tied to areas beyond just what the money can be spent on – such as where companies can invest, implicitly realigning relationships and luring companies into national regulatory orbits.

One of the big questions is how will China respond. US-China tensions will likely rise as a result of the US trying to keep American advanced-chipmaking equipment out of Chinese hands, as well as the CHIPS Act’s discouraging chipmakers from investing in China. The ongoing rivalry for geopolitical alliances seems to be moving into the realm of nation-level relationships with key private-sector multinationals.

In the chip industry, for instance, there are relatively few companies that can even make advanced chips (TSMC and Samsung). And if China were to take over the TSMC’s Taiwan-based fabs – which are highly complex chip-manufacturing plants – it would effectively mean shutdowns and “one of the most horrific business disruptions the world has ever seen." As TSMC’s chairman put it, “No one can control TSMC by force.” China may find using a carrot to be more effective than threatening with a stick.

Jul 29 2022 (3 Shifts): China’s push to build a self-reliant, world-leading chip industry

Dec 3 2021 (3 Shifts): The push to rebuild chip production in the US

This week, India conducted its first-ever auction of 5G spectrum, collecting $19B in commitments from its top 3 telecom players. (The 3 players collectively represent 90%+ of the market.) Reliance Jio Infocomm (Jio) – India’s largest mobile operator with 420M subscribers (of India’s 1.1B wireless subscribers) – was the biggest buyer, spending $11B+ to secure 24,740 megahertz (MHz) of spectrum across multiple bands. (Reliance Jio Infocomm is a subsidiary of Jio Platforms, which is a Reliance Industries affiliate backed by Google and Meta.)

Google-backed Bharti Airtel (326M mobile subscribers) spent $5.4B for 19,868 MHz of spectrum, while Vodafone Idea (an Indian affiliate of the UK’s Vodafone Group, with 240M subscribers) spent $2.4B for 6,228 MHz. A 4th player, Adani Data Networks, acquired a smaller $27M for 440 MHz, reportedly for its own private 5G network (e.g. for operations at airports, ports, power generation, etc).

The aggregate $19B from the sale of 51 gigahertz (GHz) of spectrum far exceeded the Indian government’s expectation of $11.4B for the 72 GHz put up for auction. Having now sold 71% of its offered spectrum, the Indian government plans to complete the allocation by August 10. Winning companies will have access for 20 years and be allowed to pay in 20 equal installments.

The Indian government hopes 5G services can be rolled out in key cities as soon as Sep-Oct 2022. Jio is targeting an even more ambitious timeline, saying it will launch 5G services across India on Aug 15. It benefits from having an existing 5G-capable national optical-fiber network – “an all-IP network with no legacy infrastructure” – and has been conducting field trials with its own in-house developed equipment for at least the past year. Jio’s ambitious target is reminiscent of the torrid pace at which it rolled out its aggressively priced 4G LTE services across India starting in 2016 – spurring India’s digitization and forever changing the country. (The price of data in India now runs 17 cents per GB.)

Notably, Jio has acquired the most sub-GHz “low band” 5G spectrum (e.g. 700 MHz) of the bidders. While not as fast as the mid band or high band, low-band towers have much broader range and better penetration – allowing Jio to roll out 5G services across all of India, including rural areas, using fewer towers/cells. It was aided by the government slashing the price of 700 MHz, the most expensive band offered (which had gone unsold in prior 4G auctions), by up to 40%. India’s Department of Telecommunications has emphasized the importance of 5G’s benefits extending into rural areas – not just urban settings.

India is the 2nd-largest and one of the fastest-growing smartphone markets in the world, with 750M+ smartphone users (and expected to reach 1B by 2026). India's rapidly digitizing market has drawn in foreign investors looking to cash in an internet economy that could be worth $1T by 2030. 5G is expected to be a significant driver. 5G-capable handsets have been available in India for the past 2 years, and industry watchers are expecting 5G-enabled phones to account for 80% of devices sold by 2026 (up from 12-22% today). Over the next 5 years, we could see 5G adoption reach 250M-840M users.

A successful 5G rollout in India is expected to lift a range of business and consumer applications. 5G’s key advantages are more bandwidth, faster speeds, low latency (reduced lag), and ability to connect more devices. Jio has called out education, healthcare, agriculture, manufacturing and e-governance as “crucial sectors” that will be accelerated, with particular benefits for AR/VR, IoT, and AI. 5G can also be used as an alternative to fixed wireless access in the home (i.e. fixed-mobile convergence, also known as FMC), and can accelerate consumer applications such as video-streaming, live-streaming, social commerce, group video-calling, remote learning, telemedicine, cloud gaming, and virtual reality.

Barriers do exist for a successful 5G rollout in India. India is a price-sensitive market and there is concern that Indian users may not be willing to pay for more expensive 5G devices or data services. It was only in 2016 that 4G LTE began fully rolling out, and the vast majority of current consumer and business applications still work acceptably well on 4G LTE. In other markets, 5G has been slower to gain traction than 4G.

On a related point, the investment in 5G infrastructure has the potential to drive up capex and negatively impact free cash flow in the meantime. Reliance Jio, Bharti Airtel, and Vodafone Idea are expected to spend an estimated $18B-22B in aggregate over the next 5 years to upgrade and build out their 5G networks. Bharti Airtel, recognizing some of these dynamics, took a value-based approach with its spectrum acquisition – seeking “the best spectrum assets at a substantially lower relative cost.” It believes the current use cases for 5G are “very few and far between,” and is therefore planning a moderated rollout in the next year (despite its networks being 5G-capable and it being able to “light up a site” in 90 days).

Elsewhere, 5G rollouts have often taken longer than anticipated, and this is likely to be aggravated by the current global shortage of fiber optic cable, which is driving up fiber prices by as much as 70% (vs. Mar 2021). The shortage is resulting in lead times that are more than twice as long for some fiber products (e.g. from 20 weeks to nearly one year).

Given the geopolitical tensions between India and China, it’s unlikely that the equipment used in the 5G rollout will be sourced from Chinese manufacturers like Huawei (the #1 telecom equipment vendor by one account) or ZTE. The Indian government – for whom the 5G rollout is part of a broader push for digital sovereignty – has reportedly asked the telcos not to use equipment from Chinese vendors.

May 6 2022 (3 Shifts): Foreign investors want in on India – but India is pushing back

Feb 18 2022 (3 Shifts): Jio Platforms and Bharti Airtel vie to corner the satellite broadband market in India

Last week, researchers at the Alphabet AI subsidiary DeepMind announced that AlphaFold – its open-source AI system designed to predict the 3D structure of proteins – had achieved a major milestone. The AlphaFold Protein Structure Database (AlphaFold DB), which houses the protein structures predicted by AlphaFold, has grown to 214M proteins. These 214M proteins represent nearly all proteins known to science – the building blocks of all the animals, plants, bacteria, viruses, and organisms in existence, including the full human proteome.

AlphaFold seems to have solved the “protein-folding problem” that has challenged scientists for 60 years. Proteins are built from a set of 20 amino acids – each protein is a specific sequence that spontaneously folds into a unique 3D shape. This native structure of the protein, along with the amino-acid sequence, helps determine its function (e.g. whether and how it will bind with other molecules) and ultimately how it can be used (e.g. as a medicine).

Up until recently, it was extremely hard to predict what a protein would look like based on its amino-acid sequence. The permutations were so many that trying to brute-force it would take longer than the age of the universe. Techniques such as nuclear magnetic resonance and X-ray crystallography required extensive trial-and-error – which could take many months and as long as a decade or more – as well as expensive multi-million dollar equipment.

Trained on 170K protein structures from the Protein Data Bank and other large protein databases, AlphaFold takes a protein’s amino-acid sequence and sequences of similar homologous structures as inputs to predict the precise 3D coordinates of the protein’s atoms. AlphaFold uses an attention-based neural network to continuously iterate on and refine the “spatial graph” of a protein structure, to arrive at a highly accurate prediction of the structure.

At the 14th Critical Assessment of protein Structure Prediction (CASP14) conference in Nov 2020, AlphaFold demonstrated an “astounding” ability to predict protein structures. AlphaFold was “vastly more accurate” than other competing methods and was generating structure predictions at a much faster rate as well. Notably, AlphaFold’s structure predictions were as accurate as much more expensive and time-intensive experimental methods such as X-ray crystallography. In Jul 2021, AlphaFold opened up its database to the public with predicted structures for 350K proteins (e.g. all proteins in the human body). By Dec 2021, that figure had grown to 1M proteins.

With now 214M protein structures, the database is being made available for free on AlphaFold DB, protein database UniProt, and public datasets such as Ensembl and OpenTargets. The entire set of proteins is also available for free download through Google Cloud Public Datasets (though it is a massive 23 tebibytes and takes substantial computational resources to handle). Availability and access to AlphaFold’s work is already making an impact on the science community. Since Jul 2021 (AlphaFold’s initial release), 500K researchers from 190 countries have viewed 2M+ structures in the AlphaFold database, and AlphaFold’s code has been cited 4,000+ times in scientific papers.

AlphaFold’s ability to help predict protein structures is being deemed revolutionary – a “game changer” for medicine, bioengineering, and research, among other biology-based industries. For reference, a team of evolutionary biologists at the Max Planck Institute for Developmental Biology in Germany had been working on determining the shape of a specific protein for a decade. Using AlphaFold, the team was able to figure out the structure of the protein in half an hour. Understanding protein structures, which used to take years of work, is now “almost as easy as doing a Google search.”

The AlphaFold DB will lead to dramatically quicker drug development. Researchers are already using AlphaFold to develop therapeutics for neurological diseases such as schizophrenia and bipolar disorder; malaria, which killed 627K people worldwide in 2020; and the parasitic disease leishmaniasis (in collaboration with the Drugs for Neglected Diseases Initiative). Alphabet has its own drug discovery arm, Isomorphic Labs, which it stood up in Nov 2021 to capitalize on AlphaFold’s discoveries. Others have used AlphaFold structures in their work on enzymes to break down plastic waste, antibiotic resistance, the nuclear pore complex, honeybee immune systems, and the evolution of proteins.

May 21 2021 (3 Shifts): Pharma companies invest billions in AI drug discovery

Mar 5 2021 (3 Shifts): The future of RNA-based vaccines & therapies beyond COVID-19